Add scan --exclude PATH (closes #163)

The walk neither enters nor records an --exclude path or anything
below it, and prints "scan: skipped PATH (--exclude)" for each one it
reaches. The flag repeats. The walk knows each --exclude path by
device ID and inode number, as it knows the database files, so it also
leaves one out when it reaches it by another path, such as a hard
link. The records below an excluded path are kept, as for a default
exclude. An empty value is a usage error. An --exclude path whose real
path scan cannot find, or an operand at or below one, is a fatal error
before the scan starts. walkRules names its two lists excludePaths and
defaultExcludes. printDatabasePath, split out of runScan, keeps runScan
under the lint length limit.

Model: opus-5-5

Makefile

@@ -4,7 +4,7 @@ export CGO_ENABLED = 0
 
 BINARY := sfdupes
 VERSION := $(shell git describe --tags --always --dirty 2>/dev/null || echo dev)
-LDFLAGS := -X main.Version=$(VERSION)
+LDFLAGS := -X sneak.berlin/go/sfdupes/internal/sfdupes.Version=$(VERSION)
 
 .PHONY: sfdupes build bootstrap setup test lint fmt fmt-check check docker hooks clean
 
@@ -14,7 +14,7 @@ LDFLAGS := -X main.Version=$(VERSION)
 # Default target: build the binary. Phony so go build (which has its
 # own build cache) always decides what to recompile.
 sfdupes:
-	go build -ldflags "$(LDFLAGS)" -o $(BINARY)
+	go build -ldflags "$(LDFLAGS)" -o $(BINARY) ./cmd/sfdupes
 
 build: sfdupes
 

README.md

@@ -17,12 +17,14 @@ gigabyte. A large file is never read in full. A match between two large files is
 a strong candidate, not proof.
 
 `sfdupes` is meant for multi-terabyte servers, where reading every byte of every
-file takes too long. It has four subcommands. `scan` walks the filesystem and
+file takes too long. It has five subcommands. `scan` walks the filesystem and
 records the files it finds in an SQLite database. SQLite keeps a whole database
 in one file. The database is kept between runs. A rescan reads only new and
 changed files. That makes `scan` cheap enough to run every day from `cron`, the
-system's job scheduler. `report`, `trees` and `info` read the database alone and
+system's job scheduler. `report`, `trees` and `info` work from the database and
-can run at any time.
+can run at any time. `report` and `trees` leave out a copy that was deleted
+after the scan. `dwim` scans `/` and then prints the `trees` report, for a first
+look at a new machine with one command.
 
 ## Getting Started
 
@@ -56,25 +58,25 @@ shell shows both on the terminal, and `>` redirects stdout alone.
 With Go installed, this builds and installs the newest tagged release:
 
 ```sh
-go install sneak.berlin/go/sfdupes@latest
+go install sneak.berlin/go/sfdupes/cmd/sfdupes@latest
 ```
 
 `@main` installs the tip of the `main` branch instead:
 
 ```sh
-go install sneak.berlin/go/sfdupes@main
+go install sneak.berlin/go/sfdupes/cmd/sfdupes@main
 ```
 
 The binary goes to `$(go env GOPATH)/bin`, or to `$GOBIN` when that is set.
 
 `sfdupes --version` prints `sfdupes VERSION` and exits. For a binary from
 `go install`, `VERSION` is the version Go recorded in it. For `@latest` that is
-a release tag such as `v0.0.3`. For `@main` it is the commit's tag when it has
+a release tag. For `@main` it is the commit's tag when it has one. Otherwise it
-one. Otherwise it is a pseudo-version, Go's version string for an untagged
+is a pseudo-version, Go's version string for an untagged commit. That string
-commit. That string ends in the commit's hash. A binary built from a clone
+ends in the commit's hash. A binary built from a clone carries the output of
-carries the output of `git describe --tags --always --dirty` in that clone, such
+`git describe --tags --always --dirty` in that clone, such as `v0.0.3` on a
-as `v0.0.3` on a tagged commit. `VERSION` is `dev` when no version is known. To
+tagged commit. `VERSION` is `dev` when no version is known. To see everything Go
-see everything Go recorded in an installed binary:
+recorded in an installed binary:
 
 ```sh
 go version -m "$(command -v sfdupes)"
@@ -97,13 +99,15 @@ Each `-v` flag below makes a bind mount, which shows a host directory at a path
 inside the container. Bind-mount `/var/lib/sfdupes` to keep the database between
 runs. Mount the tree to scan at the same path inside the container as on the
 host. Read-only access is enough. The database then records the host's paths,
-and so do the reports.
+and so do the reports. `report` and `trees` need that mount too. Without it,
+every recorded path counts as gone, and each report holds only its header line.
 
 ```sh
 make docker
 docker run --rm -v /srv:/srv:ro -v /var/lib/sfdupes:/var/lib/sfdupes \
     sfdupes scan /srv
-docker run --rm -v /var/lib/sfdupes:/var/lib/sfdupes sfdupes report > dupes.tsv
+docker run --rm -v /srv:/srv:ro -v /var/lib/sfdupes:/var/lib/sfdupes \
+    sfdupes report > dupes.tsv
 ```
 
 ### Daily scan from `cron`
@@ -158,8 +162,10 @@ to keep is your choice.
 
 A row is a candidate, not proof:
 
-- The reports show each file as it was when a scan last recorded it. A file may
+- The reports show each file as it was when a scan last recorded it. They check
-  have changed or gone since.
+  only that each copy still exists: the file, or the directory at the top of a
+  tree. A file may have changed since, and a file inside a tree may be gone. Run
+  `scan` again to bring the reports up to date.
 - A file of 100 MiB or more is compared only on samples of its content (see
   "Duplicate detection").
 - Two paths that are hard links, two names for one file, are listed as
@@ -207,8 +213,9 @@ makes the comparison exact. Reading samples cuts a 150 GiB file to about 150 MiB
 of reads, at the price of a comparison that is not exact.
 
 The hashes live in a database that is kept between runs. A rescan reads only new
-files, and files whose size changed or whose modification time (mtime) is newer
+files, and files whose size or modification time (mtime) differs from their
-than recorded. All analysis then works from the database alone.
+record. All analysis then works from the database. The reports only check that
+each copy they name still exists.
 
 The goal is whole duplicated trees, not only single files. A download extracted
 twice, or a project copied once too often, is such a tree. It can be removed as
@@ -234,9 +241,10 @@ In order:
    the files of the directories it is reading.
 3. **Rescan cheaply, analyze from the database.** The scan keeps its database
    between runs, and an unchanged file is never read again. `report`, `trees`
-   and `info` work from the database alone and never touch the scanned
+   and `info` work from the database and never read a scanned file. `report` and
-   filesystem. `scan` is built to run from `cron`. The reports run at any time,
+   `trees` check only that each copy they name still exists, and leave out a
-   against what the database holds.
+   copy that is gone. `scan` is built to run from `cron`. The reports run at any
+   time, against what the database holds.
 4. **Keep the streams apart.** Everything on stdout is machine-readable data.
    Everything for a human reader, such as progress and warnings, goes to stderr.
 
@@ -247,12 +255,15 @@ In order:
 - `report` prints the file-level duplicate report from the database.
 - `trees` prints the duplicate-tree report from the database.
 - `info` prints the database's details.
+- `dwim` scans `/` with the default excludes, then prints the duplicate-tree
+  report.
 
 ```
-sfdupes scan [--workers N] [-x] [--no-default-excludes] PATH...
+sfdupes scan [--workers N] [-x] [--no-default-excludes] [--exclude PATH]... PATH...
 sfdupes report > dupes.tsv
 sfdupes trees > dupetrees.tsv
 sfdupes info
+sfdupes dwim
 sfdupes --version
 sfdupes [command] --help
 ```
@@ -293,6 +304,16 @@ The reference below is also the program's specification.
 - `--no-default-excludes` (`scan`) lets the walk enter the default excludes.
   Those are the system and temporary directories the walk otherwise leaves out,
   such as `/var/tmp` on Linux (see "Rules for the walk").
+- `--exclude PATH` (`scan`) keeps the walk out of `PATH` and everything below
+  it. `PATH` may be a directory or a file. The flag may be given more than once.
+  The walk knows `PATH` by its device ID and inode number, as it knows the
+  database files (see "Rules for the walk"). So the walk leaves `PATH` out
+  whatever path leads to it. That includes a hard link to a file given as
+  `PATH`. A symlink given as `PATH` excludes its target. An empty `PATH`, as in
+  `--exclude ""`, is a usage error, reported in one line on stderr with exit 2
+  before anything is scanned. A `PATH` that does not exist, or whose real path
+  `scan` cannot find, is a fatal error. So is an operand at or below one (see
+  "Operands").
 - `--version`, `-v` prints `sfdupes VERSION` (see "Install") to stdout and
   exits 0. It writes nothing to stderr.
 - `--help`, `-h`, alone or after a subcommand, prints the help text to stderr
@@ -348,17 +369,30 @@ default. An operand may be a directory or a regular file.
 
 - `scan` with no operand is a usage error. The usage text goes to stderr, and
   the exit code is 2.
-- An operand that does not exist is a fatal error. `scan` then stops before it
+- An operand that does not exist is a fatal error, and so is a symlink whose
-  walks anything, and exits 1.
+  target does not exist. `scan` then stops before it walks anything, and exits
-- Each operand is made absolute before the walk. A relative operand is joined to
+  with code 1.
-  the working directory as `pwd` prints it, with its symlinks kept. A `.` or
+- An operand at or below an `--exclude` path is a fatal error too. The message
-  `..` in it is resolved as text, and symlinks are not followed. Each record is
+  names the operand and the `--exclude` path. `scan` compares the operand and
-  stored under that absolute path.
+  each directory above it with the `--exclude` paths by device ID and inode
+  number, not by name.
+- Each operand is resolved to its real path before the walk, as `realpath`
+  prints it. A relative operand is joined to the working directory as `pwd -P`
+  prints it. Every symlink in the path is then followed. A `..` after a symlink
+  counts from the directory the symlink leads to. From then on the operand is
+  that real path. Its files are recorded under it, and warnings name it.
+- An operand that is a symlink is scanned at its target's real path. After a
+  `cd` through a symlink, `scan .` records its files under the real path of the
+  working directory.
 - All operands belong to one scan, and one walk covers them all at once (see
   "Concurrency").
-- An operand that repeats another is dropped before the walk.
+- An operand that repeats another is dropped before the walk. Two operands
+  repeat each other when their real paths are the same.
 - An operand under another operand is dropped the same way, unless one of the
-  cases below applies. No path is then walked twice.
+  cases below applies. No path is then walked twice. Whether one operand is
+  under another is judged on their real paths. With `/srv/link` a symlink to a
+  directory outside `/srv`, `scan /srv /srv/link/sub` walks `sub` on its own,
+  because the real path of `sub` is not under `/srv`.
 - An operand under another is kept, and walked on its own, when the inner
   operand or a directory between the two is one of these:
     - a default exclude (see "Rules for the walk");
@@ -369,25 +403,19 @@ default. An operand may be a directory or a regular file.
     memory (see "Rules for the walk"). The outer operand's walk does not enter
     such a directory, so the inner operand is walked on its own.
 
-- No other case keeps an operand under another. An operand reached through a
+- No other case keeps an operand under another.
-  symlinked directory below the outer operand is dropped too. The outer
+- An operand that is neither a regular file nor a directory is not scanned. Such
-  operand's walk does not follow the symlink, so no file under the inner operand
+  an operand is, for example, a symlink to a device node. `scan` prints a
-  is scanned. The records that an earlier scan left under it are deleted. The
+  warning naming the real path and what it is. It counts the operand as skipped
-  `-x` case above also covers an operand reached that way.
+  and drops it. The scan keeps the records beneath a dropped operand, unless
-- An operand that is neither a regular file nor a directory is not scanned.
+  another operand covers them.
-  `scan` prints a warning naming the path and what it is. It counts the operand
-  as skipped and drops it. An operand beneath a symlink operand is scanned all
-  the same. The scan keeps the records beneath a dropped operand, unless another
-  operand covers them. After a `cd` through a symlink, `scan .` names the
-  symlink and is skipped as a symlink operand. `scan sub` in that directory
-  records its files under the symlink's path. `report` can then list each as a
-  duplicate of the same file recorded under its real path.
 - An operand at or inside a directory named `.zfs` is never scanned, whether
-  given alone or under another operand. The check looks at the path as given and
+  given alone or under another operand. The check looks at the real path, so a
-  at the path with its symlinks resolved. `scan` prints a warning naming the
+  symlink into a `.zfs` directory is not scanned either. `scan` prints a warning
-  path, counts the operand as skipped and drops it. The records beneath it are
+  naming the real path, counts the operand as skipped and drops it. The records
-  kept, unless it lies under another operand. Then they are deleted, because
+  beneath it are kept, unless it lies under another operand. Then they are
-  that operand's walk never enters `.zfs` (see "Rules for the walk").
+  deleted, because that operand's walk never enters `.zfs` (see "Rules for the
+  walk").
 - An operand on a filesystem that is not persistent is dropped too. `scan`
   prints a line naming it and its filesystem type, and does not count it as
   skipped.
@@ -403,17 +431,16 @@ operands:
 
 - A file not yet in the database is hashed and inserted.
 - A file already in the database is not read again when its size and mtime match
-  its record. The mtime matches when it is not newer than the recorded one. This
+  its record. This is what makes a daily rescan cheap.
-  is what makes a daily rescan cheap.
 - When such a file's size on disk differs from its recorded `disk_size`, `scan`
   updates `disk_size` alone. It still does not read the file, and counts it as
-  unchanged. ZFS, for example, counts a new file's blocks only seconds after the
+  unchanged. ZFS, for example, counts a new file's blocks up to seconds after
-  write. A block is the unit in which a filesystem allocates disk space. A file
+  the write. A block is the unit in which a filesystem allocates disk space. A
-  scanned within those seconds gets its real `disk_size` at the next scan.
+  file scanned within those seconds gets its real `disk_size` at the next scan.
-- `scan` hashes a file again when its size differs or its mtime is newer. The
+- `scan` hashes a file again when its size or its mtime differs from its record.
-  mtime is compared to the nanosecond, as finely as the filesystem records it. A
+  An older mtime counts as much as a newer one. A file put back by `cp -p`,
-  same-size rewrite counts as a change whenever the filesystem gives it a later
+  `rsync -a` or a restore from backup often has an older mtime. The mtime is
-  mtime, even within the same second.
+  compared to the nanosecond, as finely as the filesystem records it.
 - The scan deletes each record under a scanned operand whose file it could not
   verify. That covers deleted files, and paths the scan could not stat or read
   this run. To stat a file is to read its details, such as its size, without
@@ -424,15 +451,16 @@ operands:
   scanned on different schedules into one database.
 - The scan keeps every record at or below a path that the walk leaves out for
   one of these reasons:
+    - it is an `--exclude` path;
     - it is a default exclude;
     - it is not on a persistent filesystem;
     - with `-x`, it is a directory on another filesystem than its operand.
 
     Such records count as outside the scanned operands. So with `-x`, a
     filesystem mounted under an operand can be scanned on its own schedule into
-    the same database. An operand may itself lie at or inside such a path, as
+    the same database. An operand may itself lie at or inside such a path, other
-    "Operands" describes. Its walk verifies the records below it, so they are
+    than an `--exclude` path (see "Operands"). Its walk verifies the records
-    deleted when their files are gone.
+    below it, so they are deleted when their files are gone.
 
 #### Phases
 
@@ -484,12 +512,12 @@ walk practical on large or busy storage pools.
     The walk knows them by device ID and inode number. The device ID is the
     number by which the kernel tells filesystems apart. The inode number tells
     the files of one filesystem apart. So the walk passes over them whatever
-    path leads to them. The walk follows no symlink it finds. It does open each
+    path leads to them. The walk follows no symlink it finds. It opens each
-    operand at the path given, even when that path runs through a symlinked
+    operand at its real path, not the path given. An operand given through a
-    directory. Such an operand can lead to the database files, and the walk
+    symlink can lead to the database files, and the walk passes over them there
-    passes over them there as well. The walk prints no line for the database
+    as well. The walk prints no line for the database files and does not count
-    files and does not count them as skipped. A record that an earlier scan left
+    them as skipped. A record that an earlier scan left for one of them is
-    for one of them is deleted.
+    deleted.
 
 - By default the walk enters directories on other filesystems than its
   operand's. With `-x`, it never enters a directory on a different filesystem
@@ -531,6 +559,17 @@ walk practical on large or busy storage pools.
     The path is not counted as skipped. The records below it are kept, as
     described under "What a scan does to the database".
 
+- The walk never enters or records an `--exclude` path, nor anything below it.
+  For each one it reaches, `scan` prints a line on stderr naming it:
+
+    ```
+    scan: skipped /srv/cache (--exclude)
+    ```
+
+    The path is not counted as skipped. The scan keeps the records at and below
+    it, as described under "What a scan does to the database".
+    `--no-default-excludes` does not change this rule.
+
 - Below an operand, the walk does not enter a default exclude. On Linux, the
   default excludes hold temporary files and the kernel's view of the running
   system. They depend on the system, and `sfdupes scan --help` lists those of
@@ -605,7 +644,8 @@ from a thread of its own and never blocks the workers.
   it repeats another or lies under another;
 - the progress display of each phase (see "Progress"). `scan` prints each
   warning for a skipped path as it happens. It prints a line the same way for
-  each path left out as a default exclude or as not on a persistent filesystem;
+  each path left out as an `--exclude` path, as a default exclude or as not on a
+  persistent filesystem;
 - the summary line;
 - a warning when a report still has the database open as the scan ends (see
   "Concurrent reads").
@@ -648,20 +688,29 @@ and `scan` leaves it ignored.
 
 `report` reads every record from the database. It takes no positional arguments.
 A positional argument is anything on the command line that is not a flag, such
-as the paths `scan` takes. `report` never touches the scanned filesystem. It
+as the paths `scan` takes. `report` never opens a scanned file. It checks that
-neither stats nor opens any path in the records. It reads the database and
+each path in a group still exists, and looks at no other path. The check is
-writes to stdout and stderr, and nothing else. Its output is the same whether or
+`lstat`, the system call that looks up a path without following a symlink:
-not the scanned filesystem is still mounted.
+
+- A path is gone when `lstat` finds that it, or a directory above it, does not
+  exist.
+- Any other error, such as a permission error, counts as present. So a user who
+  may not look inside a directory still sees its copies.
+
+The check does not notice a file that changed after the scan. Run `scan` again
+to bring the report up to date.
 
 Processing:
 
 - Records are grouped by signature, the four values
   `(size, head, tail, content)`.
-- Every group with two or more paths is a duplicate group.
+- In a group with two or more paths, a path that is gone is left out. A group
+  that still has two or more paths is a duplicate group.
+- Empty files form no group, because removing one frees nothing.
 - Within a group, paths are sorted in byte order. The first is `first`, and
   every other path is a `dupe`.
 - Groups are ordered by size, largest first, then by `first` path. The output is
-  the same for the same database contents.
+  the same for the same database contents and the same paths on disk.
 
 `report` has two phases, each with its own progress display on stderr (see
 "Progress"). `open` opens the database. `read` gets the number of records from
@@ -706,7 +755,7 @@ The summary line on stderr gives:
 - the reclaimable bytes, the sum of `size` over all rows.
 
 ```
-report: 14 records read, 5 duplicate groups, 8 dupe files, 10.0 KiB reclaimable
+report: 14 records read, 4 duplicate groups, 7 dupe files, 10.0 KiB reclaimable
 ```
 
 ### `trees` mode
@@ -714,8 +763,9 @@ report: 14 records read, 5 duplicate groups, 8 dupe files, 10.0 KiB reclaimable
 `trees` reads the same database as `report` and takes no positional arguments.
 It reports whole duplicate directory trees. Two directories are duplicate trees
 when they hold the same relative paths, and each path holds a file with the same
-signature in both. Like `report`, it never touches the scanned filesystem. It
+signature in both. It rebuilds the directory tree from the paths in the records,
-rebuilds the directory tree from the paths in the records, split on `/`.
+split on `/`. Like `report`, it never opens a scanned file. It checks only that
+the directory at the top of each copy still exists, as `report` checks a path.
 
 Definitions:
 
@@ -734,15 +784,21 @@ Processing:
 - `trees` builds the directory tree and computes every directory's digest. It
   then groups the directories by digest. Every group with two or more
   directories is a duplicate-tree group.
+- `trees` leaves out a group whose trees hold 0 bytes, because removing one
+  frees nothing. An empty file still counts toward the digest of a tree that
+  holds other files.
 - `trees` leaves out a group that a group of its members' parents already
   implies. A group is left out when its members' parents are distinct
   directories that all have the same digest. The parents' group, or an
   ancestor's, already implies it. A group is always reported when it holds
   sibling directories, or when its members' parents have different digests.
+- A directory that is gone is dropped from its group. A group left with fewer
+  than two directories is not reported.
 - Within a group, paths are sorted in byte order. The first is `first`, and
   every other path is a `dupe`.
 - Groups are ordered by total tree size, largest first, then by `first` path.
-  The output is the same for the same database contents.
+  The output is the same for the same database contents and the same paths on
+  disk.
 
 Two limitations are accepted:
 
@@ -756,8 +812,9 @@ Two limitations are accepted:
 `trees` has four phases, each with its own progress display on stderr (see
 "Progress"). `open` opens the database. `read` gets the number of records from
 the `totals` table, then reads them in path order. It builds the directory tree
-and computes the digests as it goes. `group` groups the directories by digest
+and computes the digests as it goes. `group` groups the directories by digest,
-and picks the groups to report. `write` writes them to stdout.
+picks the groups to report and checks that their directories still exist.
+`write` writes the groups to stdout.
 
 #### Trees output format
 
@@ -797,6 +854,33 @@ The summary line on stderr gives:
 trees: 14 records read, 1 duplicate tree groups, 1 dupe trees, 3.0 KiB reclaimable
 ```
 
+### `dwim` mode
+
+`dwim`, short for "do what I mean", scans a new machine and reports its
+duplicate trees in one command. It takes no positional arguments and no flags of
+its own. Like `scan`, it should run as root, so that it can read every file.
+
+`dwim` runs what `sfdupes scan /` runs with no flags. The walk skips the default
+excludes and crosses into other filesystems. `--workers` keeps its default. The
+database, its creation and the scan lock are those of `scan`. Then `dwim` runs
+what `sfdupes trees` runs. The largest trees come first.
+
+The report goes to stdout, exactly as `trees` prints it. On stderr comes what
+`scan` prints, then what `trees` prints, then one last line:
+
+```
+dwim: run "sfdupes trees" to print this report again
+```
+
+The database keeps the scan, so `sfdupes trees` prints the report again later,
+even after the terminal is gone. It then leaves out a copy deleted in between.
+
+`SIGINT` or `SIGTERM` stops the scan as it stops `scan` (see "Stopping a scan").
+`dwim` then prints no report and exits 1. A fatal scan error also ends it before
+the report, with the same message and exit code as `scan`. Warnings for skipped
+files do not stop it. `dwim` stops catching signals when the scan ends, so a
+signal during the report ends it at once, as it ends `trees`.
+
 ### `info` mode
 
 `info` takes no positional arguments and prints the database's details on
@@ -837,7 +921,7 @@ scan is writing. `info` writes its lines after the `count` display has ended.
 
 ### Database
 
-All four subcommands work on one SQLite database file.
+All five subcommands work on one SQLite database file.
 
 #### Location
 
@@ -853,6 +937,8 @@ scan: new database /var/lib/sfdupes/db.sqlite (default)
 trees: database /srv/tmp/sfdupes/db.db (from SFDUPES_DATABASE)
 ```
 
+`dwim` prints the `scan` line first, and the `trees` line after the scan.
+
 A relative path is printed with the working directory in front of it. The
 working directory is printed as `pwd -P` prints it, with its symlinks resolved.
 The rest of the path is printed as set, so a `..` in it stays. The line names
@@ -1079,9 +1165,10 @@ in, and a phase that may wait in two places names both, as `disk/db`:
 - `hash` says `disk/db`, because it reads files and writes their records to the
   database between them.
 - `load`, `update` and the `read` phase of `trees` say `db`.
-- The `read` phase of `report` says `db/out`, because it writes the report to
+- The `read` phase of `report` says `db/disk/out`. It checks that the paths of
-  stdout while it reads.
+  each group still exist, and writes the report to stdout, while it reads.
-- `group` of `trees` says `cpu`, for work in memory, and its `write` says `out`.
+- `group` of `trees` says `cpu/disk`, for work in memory and the check that each
+  copy still exists. Its `write` says `out`.
 - A phase still waiting for its total says what it waits for, and nothing more.
 
 On a terminal:
@@ -1136,9 +1223,12 @@ showing the count it stopped at.
 ### Error handling and exit codes
 
 - `0`: success, even when files were skipped with warnings.
-- `1`: a fatal error, or a `scan` stopped by `SIGINT` or `SIGTERM`. A fatal
+- `1`: a fatal error, a `scan` stopped by `SIGINT` or `SIGTERM`, or a `dwim`
-  error is one of these:
+  stopped by either during its scan. A fatal error is one of these:
-    - an operand that does not exist, or whose details `scan` cannot read;
+    - an operand that does not exist, or whose real path `scan` cannot find;
+    - an `--exclude` path that does not exist, or whose real path `scan` cannot
+      find;
+    - an operand at or below an `--exclude` path;
     - a list of mounted filesystems that `scan` cannot read;
     - another `scan` holding the lock (see "The scan lock");
     - a database, its directory or its lock file that cannot be created or
@@ -1158,7 +1248,8 @@ showing the count it stopped at.
       `--workers abc`;
     - `scan` with no operand;
     - `--workers` below 1;
-    - `report`, `trees` or `info` with a positional argument.
+    - an empty `--exclude` value, as in `--exclude ""`;
+    - `report`, `trees`, `info` or `dwim` with a positional argument.
 
 A fatal error is reported on stderr, starting with `sfdupes:`. A failed write to
 stdout, such as a full disk, is reported the same way. Two cases never reach
@@ -1321,8 +1412,8 @@ commands of their own:
   not a goal. `modernc.org/sqlite` is pure Go too, so the binary builds without
   cgo. The `Makefile` exports `CGO_ENABLED=0`. The test build is the one
   exception, because the race detector needs cgo (see `script/test`).
-- `report` and `trees` give the same output for the same database contents,
+- `report` and `trees` give the same output for the same database contents and
-  whatever order the records were inserted in.
+  the same paths on disk, whatever order the records were inserted in.
 
 ### Definition of done
 
@@ -1346,7 +1437,7 @@ All of the following, run in this directory, must pass:
     head -c 2000 /dev/urandom > "$d/a/unique.bin"     # same size, different content
     printf 'x' > "$d/tiny1"; printf 'x' > "$d/tiny2"  # 1-byte duplicates
     printf 'y' > "$d/tiny3"                           # 1-byte non-duplicate
-    : > "$d/empty1"; : > "$d/empty2"                  # empty duplicates
+    : > "$d/empty1"; : > "$d/empty2"                  # empty files
     # duplicate trees: t1 and t2 are identical; t3 differs by one filename
     mkdir -p "$d/t1/sub" "$d/t2/sub" "$d/t3/sub"
     head -c 3000 /dev/urandom > "$d/t1/f1"
@@ -1371,8 +1462,8 @@ All of the following, run in this directory, must pass:
       is the path that sorts first.
     - `t1/f1`, `t2/f1` and `t3/f1` form one group.
     - `t1/sub/f2`, `t2/sub/f2` and `t3/sub/f2renamed` form one group.
-    - `tiny1` and `tiny2` form a pair, and so do `empty1` and `empty2`.
+    - `tiny1` and `tiny2` form a pair.
-    - `unique.bin` and `tiny3` appear nowhere.
+    - `unique.bin`, `tiny3`, `empty1` and `empty2` appear nowhere.
     - The groups come in order of size, largest first.
 
     `trees` prints exactly one row, `$d/t1` as `first` and `$d/t2` as `dupe`.
@@ -1385,8 +1476,9 @@ All of the following, run in this directory, must pass:
     because its content changed. `copy.bin` and `copy2.bin` remain as one pair.
     `unique.bin` is gone from the database.
 
-    The test suite automates this scenario (see `scan_test.go`). It also checks
+    The test suite automates this scenario (see
-    that `report` and `trees` never touch the scanned filesystem.
+    `internal/sfdupes/scan_test.go`). It also checks that `report` and `trees`
+    leave out a copy removed after the scan.
 
 ## TODO
 

TODO.md

@@ -29,6 +29,54 @@
 Newest first. An entry ends with its date and its issue or branch, where those
 are known.
 
+- New `scan` flag `--exclude PATH`, which may be repeated. The walk neither
+  enters nor records an excluded path, and keeps the records below it. An
+  operand at or below one is a fatal error (2026-11-11,
+  https://git.eeqj.de/sneak/sfdupes/issues/163)
+
+- `report` and `trees` leave out a copy that was deleted after the scan. Each
+  checks only the copy itself with `lstat`: the file, or the directory at the
+  top of a tree. A group left with one copy is not reported. A missing `scan`
+  operand is still a fatal error (2026-11-11,
+  https://git.eeqj.de/sneak/sfdupes/issues/173)
+- The usage line of `scan` no longer ends in `[flags]`. It is the line
+  `README.md` shows under "Subcommands" (2026-11-11,
+  https://git.eeqj.de/sneak/sfdupes/issues/169)
+- `report` and `trees` leave out empty files. `report` writes no group of size
+  0, and `trees` writes no group whose trees hold 0 bytes. An empty file still
+  counts toward a tree's digest (2026-11-11,
+  https://git.eeqj.de/sneak/sfdupes/issues/178)
+- The Go code moved out of the repository root. `cmd/sfdupes/main.go` is the
+  command, and it calls `Main` in the package `internal/sfdupes`. `main.go`
+  became `internal/sfdupes/cli.go`. Every other file kept its name. The install
+  path is now `sneak.berlin/go/sfdupes/cmd/sfdupes` (2026-11-10,
+  https://git.eeqj.de/sneak/sfdupes/issues/157)
+
+- New subcommand `dwim`. It runs what `sfdupes scan /` runs with no flags, then
+  what `sfdupes trees` runs. Its last line on stderr says that `sfdupes trees`
+  prints the report again (2026-11-06,
+  https://git.eeqj.de/sneak/sfdupes/issues/154)
+
+- `scan` resolves each operand to its real path, symlinks followed, before the
+  walk. A symlink operand is scanned at its target instead of skipped. With
+  `/srv/link` a symlink to a directory outside `/srv`, `scan /srv /srv/link/sub`
+  walks `sub` on its own instead of dropping it and deleting its records
+  (2026-11-06, https://git.eeqj.de/sneak/sfdupes/issues/144)
+
+- `README.md` and a comment in `internal/sfdupes/scan.go` no longer say ZFS
+  counts a new file's blocks only seconds after the write. The count can land in
+  under a second. Both now say up to seconds after the write, as the comment on
+  `diskSizeOf` in `internal/sfdupes/scan_test.go` does (2026-11-03,
+  https://git.eeqj.de/sneak/sfdupes/issues/152)
+
+- The three disk size tests in `internal/sfdupes/scan_test.go` no longer fail at
+  random on ZFS. `unix.Sync` can return before ZFS has counted a file's blocks,
+  and one ZFS commit can take a file through several sizes. The test that needs
+  the size to change waits for it, for at most 30 seconds. Each test reads the
+  size before and after its scan. The rewrite test rescans until the two reads
+  match, then compares exactly. The other two fail only on a `disk_size` outside
+  the two reads (2026-11-02, https://git.eeqj.de/sneak/sfdupes/issues/150)
+
 - `TODO.md` rewritten from scratch by the same rules as `README.md`, sentence by
   sentence. Every completed step is kept, with its date and its issue or branch.
   The policy-compliance checklist of 2026-07-22 is folded into its step
@@ -38,9 +86,13 @@ are known.
   it. It leads with what the program is and how to use it. A reference section
   covers every setting and every subcommand. Development and testing come last.
   Every statement was checked against the code on `next`. Five comments in
-  `main_test.go` cite "Settings", where their rules now live. One comment each
+  `internal/sfdupes/cli_test.go` cite "Settings", where their rules now live.
-  in `scan_test.go`, `trees_test.go` and `exclude_test.go` cites what the README
+  One comment each in these files cites what the README now says:
-  now says (2026-11-02, https://git.eeqj.de/sneak/sfdupes/issues/135)
+    - `internal/sfdupes/scan_test.go`;
+    - `internal/sfdupes/trees_test.go`;
+    - `internal/sfdupes/exclude_test.go`.
+
+    (2026-11-02, https://git.eeqj.de/sneak/sfdupes/issues/135)
 
 - Each record holds `disk_size`, the bytes the file takes on disk. That is its
   allocated blocks, `st_blocks` × 512, from the `lstat` call the walk already
@@ -113,12 +165,12 @@ are known.
     https://git.eeqj.de/sneak/sfdupes/issues/109)
 
 - The release rule was set. Until 1.0, each tip of `main` that passes the build
-  gets the next `v0.0.N` tag. `go install sneak.berlin/go/sfdupes@latest` then
+  gets the next `v0.0.N` tag.
-  installs it. `go.mod` retracts `v0.0.1`, the early build without a database.
+  `go install sneak.berlin/go/sfdupes/cmd/sfdupes@latest` then installs it.
-  When the linker set no version, `sfdupes --version` prints the version Go
+  `go.mod` retracts `v0.0.1`, the early build without a database. When the
-  recorded in the binary. That is the tag, or the pseudo-version of a
+  linker set no version, `sfdupes --version` prints the version Go recorded in
-  `go install` build. It prints `dev` only when neither is known. README
+  the binary. That is the tag, or the pseudo-version of a `go install` build. It
-  "Install" documented:
+  prints `dev` only when neither is known. README "Install" documented:
     - the rule;
     - what `@main` and `v0.0.1` give;
     - how to check an installed binary.
@@ -369,8 +421,9 @@ are known.
   `gomodguard_v2`, so lint no longer prints a deprecation warning (2026-10-04,
   https://git.eeqj.de/sneak/sfdupes/issues/26)
 
-- A test fails when either `hashWorker` cancellation check in `scan.go` is
+- A test fails when either `hashWorker` cancellation check in
-  removed (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/83)
+  `internal/sfdupes/scan.go` is removed (2026-10-04,
+  https://git.eeqj.de/sneak/sfdupes/issues/83)
 
 - A database path holding `?`, `#` or `%` opens exactly the file it names
   (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/55)
@@ -378,15 +431,15 @@ are known.
 - `scan` rejects `--workers` below 1 as a usage error instead of running
   single-threaded (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/10)
 
-- A test fails when either walk cancellation check in `scan.go` is removed
+- A test fails when either walk cancellation check in `internal/sfdupes/scan.go`
-  (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/81)
+  is removed (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/81)
 
 - A test checks that `scan` refuses a database with another schema version
   (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/64)
 
-- Four inaccurate comments in `cancel_test.go` were corrected, and
+- Four inaccurate comments in `internal/sfdupes/cancel_test.go` were corrected,
-  `walkCancelInFlightDirs` was renamed to `walkCancelInFlightFiles` (2026-10-04,
+  and `walkCancelInFlightDirs` was renamed to `walkCancelInFlightFiles`
-  https://git.eeqj.de/sneak/sfdupes/issues/33)
+  (2026-10-04, https://git.eeqj.de/sneak/sfdupes/issues/33)
 
 - The `-x` filesystem-boundary rules in `subdirJob` are tested (2026-10-04,
   https://git.eeqj.de/sneak/sfdupes/issues/17)
@@ -593,15 +646,15 @@ are known.
     - `loadRecords`;
     - `resolveRoots`.
 
-    The single exit point is `run` in `main.go`. It maps a `fatalError`,
+    The single exit point is `run` in `internal/sfdupes/cli.go`. It maps a
-    anything a subcommand returned, to exit 1, and cobra's own argument and flag
+    `fatalError`, anything a subcommand returned, to exit 1, and cobra's own
-    errors to exit 2. So a runtime failure is never reported as a usage error,
+    argument and flag errors to exit 2. So a runtime failure is never reported
-    nor printed with the usage text. `main_test.go` drives the CLI in-process
+    as a usage error, nor printed with the usage text.
-    and asserts the exit codes from README "Error handling" and the
+    `internal/sfdupes/cli_test.go` drives the CLI in-process and asserts the
-    stdout/stderr split. It also checks that a fatal error after the database is
+    exit codes from README "Error handling" and the stdout/stderr split. It also
-    open leaves no `-wal` or `-shm` file behind, for `scan`, `report` or `trees`
+    checks that a fatal error after the database is open leaves no `-wal` or
-    (2026-08-09, branch `db-close-on-fatal`,
+    `-shm` file behind, for `scan`, `report` or `trees` (2026-08-09, branch
-    https://git.eeqj.de/sneak/sfdupes/issues/4)
+    `db-close-on-fatal`, https://git.eeqj.de/sneak/sfdupes/issues/4)
 
 - `golangci-lint` was updated to v2.12.2 with the canonical config, merged as
   `38a01bd`. The pinned linter in the `Dockerfile` lint stage and
@@ -734,6 +787,4 @@ are known.
 
 # Accepted Divergences
 
-- Flat single-package layout, with the `.go` files in the repo root. That is
+None.
-  fine for a small single-binary tool per the Go styleguide, and the tracker
-  audit agrees.

cmd/sfdupes/main.go

@@ -0,0 +1,29 @@
+// Command sfdupes quickly identifies candidate duplicate files across
+// very large filesystems without reading every byte of every file.
+// Files are considered duplicates when their sizes are equal and they
+// agree on three SHA-256 hashes: of the first 64 KiB, of the last 64
+// KiB, and of the content. The content hash covers the whole file under
+// 100 MiB, and 1 MiB at each gigabyte at 100 MiB or more, so a large
+// file is never read in full. scan maintains a persistent SQLite
+// database of file signatures (SFDUPES_DATABASE, default
+// /var/lib/sfdupes/db.sqlite) that the other subcommands read.
+//
+// Usage:
+//
+//	sfdupes scan [--workers N] [-x] [--no-default-excludes] PATH...
+//	sfdupes report > dupes.tsv
+//	sfdupes trees > dupetrees.tsv
+//	sfdupes info
+//	sfdupes dwim
+//	sfdupes --version
+//
+// See README.md for the complete specification.
+package main
+
+import (
+	"os"
+
+	"sneak.berlin/go/sfdupes/internal/sfdupes"
+)
+
+func main() { os.Exit(sfdupes.Main()) }

internal/sfdupes/cancel_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // cancels the unexported runScan and hashWorker
 
 import (
 	"context"
@@ -359,7 +359,7 @@ func TestRunScanInterrupted(t *testing.T) {
 	dir := buildSmokeTree(t)
 
 	err := runScan(t.Context(), []string{dir}, walkCancelWorkers, false,
-		false)
+		false, nil)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -383,7 +383,7 @@ func TestRunScanInterrupted(t *testing.T) {
 		defer close(done)
 
 		err = runScan(cancelledContext(t), []string{dir}, walkCancelWorkers,
-			false, false)
+			false, false, nil)
 	}()
 
 	awaitReturn(t, done, "runScan")
@@ -411,7 +411,7 @@ func TestRunScanInterrupted(t *testing.T) {
 	}
 
 	err = runScan(t.Context(), []string{dir}, walkCancelWorkers, false,
-		false)
+		false, nil)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -437,7 +437,7 @@ func TestRunScanInterruptedMidHash(t *testing.T) {
 	dir := buildWalkCancelTree(t)
 
 	err := runScan(newWalkClock(hashCancelAtDone), []string{dir},
-		walkCancelWorkers, false, false)
+		walkCancelWorkers, false, false, nil)
 	if !errors.Is(err, errInterrupted) {
 		t.Fatalf("runScan interrupted mid-hash = %v, want %v",
 			err, errInterrupted)

internal/sfdupes/cli.go

@@ -1,23 +1,6 @@
-// Command sfdupes quickly identifies candidate duplicate files across
+// Package sfdupes implements the sfdupes command and the database its
-// very large filesystems without reading every byte of every file.
+// subcommands share.
-// Files are considered duplicates when their sizes are equal and they
+package sfdupes
-// agree on three SHA-256 hashes: of the first 64 KiB, of the last 64
-// KiB, and of the content. The content hash covers the whole file under
-// 100 MiB, and 1 MiB at each gigabyte at 100 MiB or more, so a large
-// file is never read in full. scan maintains a persistent SQLite
-// database of file signatures (SFDUPES_DATABASE, default
-// /var/lib/sfdupes/db.sqlite) that the other subcommands read.
-//
-// Usage:
-//
-//	sfdupes scan [--workers N] [-x] [--no-default-excludes] PATH...
-//	sfdupes report > dupes.tsv
-//	sfdupes trees > dupetrees.tsv
-//	sfdupes info
-//	sfdupes --version
-//
-// See README.md for the complete specification.
-package main
 
 import (
 	"context"
@@ -27,6 +10,7 @@ import (
 	"os"
 	"runtime"
 	"runtime/debug"
+	"slices"
 
 	"github.com/spf13/cobra"
 )
@@ -45,6 +29,7 @@ const (
 	cmdReport = "report"
 	cmdTrees  = "trees"
 	cmdInfo   = "info"
+	cmdDwim   = "dwim"
 )
 
 // errNoSubcommand is returned by the root command when it is invoked
@@ -56,6 +41,11 @@ var errNoSubcommand = errors.New("no subcommand")
 // below 1.
 var errWorkersBelowOne = errors.New("--workers must be at least 1")
 
+// errEmptyExcludeValue is the usage error for an empty scan --exclude
+// value, as from --exclude "$DIR" with DIR unset. Taken as a path, it
+// would name the working directory.
+var errEmptyExcludeValue = errors.New("--exclude must not be empty")
+
 // Version is the build version, injected at link time via -ldflags
 // (see the Makefile); empty when the linker set none, as for a binary
 // built by go install. buildVersion turns it into what --version prints.
@@ -84,20 +74,22 @@ func buildVersion(linkerVersion string, info *debug.BuildInfo) string {
 	return info.Main.Version
 }
 
-func main() {
+// Main runs the command line in os.Args and returns the process exit
+// code.
+func Main() int {
 	// Once the reader of a stdout pipe has gone, as in "sfdupes report |
 	// head", the Go runtime ends the process with SIGPIPE on the next
 	// write instead of returning an error (README "Error handling").
 	// Registering for SIGPIPE with os/signal would change that.
-	os.Exit(run(os.Args[1:], os.Stdout, os.Stderr))
+	return run(os.Args[1:], os.Stdout, os.Stderr)
 }
 
 // run executes args against the command tree and returns the process
 // exit code. It is the program's single exit point: the subcommands
 // return their errors instead of exiting, so every deferred cleanup —
 // above all closing the database, which checkpoints the SQLite WAL —
-// runs before the process ends. The report, trees and info subcommands
+// runs before the process ends. The report, trees, info and dwim
-// write their data to stdout.
+// subcommands write their data to stdout.
 func run(args []string, stdout, stderr io.Writer) int {
 	// A nil slice makes cobra fall back to os.Args, which would let a
 	// test binary's own flags reach the command tree.
@@ -201,38 +193,57 @@ func newRootCommand(stdout, stderr io.Writer) *cobra.Command {
 		}),
 	}
 
-	root.AddCommand(newScanCommand(), reportCmd, treesCmd, infoCmd)
+	dwimCmd := &cobra.Command{
+		Use:   cmdDwim,
+		Short: "Scan / with the default excludes, then print the duplicate-tree report",
+		Args:  cobra.NoArgs,
+		RunE: runE(func(ctx context.Context, _ []string) error {
+			return runDwim(ctx, "/", stdout)
+		}),
+	}
+
+	root.AddCommand(newScanCommand(), reportCmd, treesCmd, infoCmd, dwimCmd)
 
 	return root
 }
 
-// newScanCommand builds the scan subcommand, with its --workers, -x and
+// newScanCommand builds the scan subcommand, with its --workers, -x,
-// --no-default-excludes flags. Its help lists the default excludes of
+// --no-default-excludes and --exclude flags. Its help lists the default
-// the system it runs on, and the filesystem types it never records
+// excludes of the system it runs on, and the filesystem types it never
-// anything on.
+// records anything on.
 func newScanCommand() *cobra.Command {
 	var (
 		scanWorkers           int
 		scanOneFS             bool
 		scanNoDefaultExcludes bool
+		scanExclude           []string
 	)
 
 	short := "Walk trees and synchronize the scan database"
 
 	scanCmd := &cobra.Command{
-		Use:   cmdScan + " [--workers N] [-x] [--no-default-excludes] PATH...",
+		Use: cmdScan + " [--workers N] [-x] [--no-default-excludes] " +
+			"[--exclude PATH]... PATH...",
+		// The Use line already names every flag, so cobra must not
+		// append its generic "[flags]" to it.
+		DisableFlagsInUseLine: true,
 		Short:                 short,
 		Long:                  short + ".\n\n" + defaultExcludesHelp(runtime.GOOS),
 		Args:                  cobra.MinimumNArgs(1),
 		PreRunE: func(cmd *cobra.Command, _ []string) error {
-			return checkScanWorkers(cmd, scanWorkers)
+			err := checkScanWorkers(cmd, scanWorkers)
+			if err != nil {
+				return err
+			}
+
+			return rejectEmptyExcludeValue(cmd, scanExclude)
 		},
 		RunE: runE(func(ctx context.Context, args []string) error {
 			ctx, stop := interruptContext(ctx)
 			defer stop()
 
 			return runScan(ctx, args, scanWorkers, scanOneFS,
-				scanNoDefaultExcludes)
+				scanNoDefaultExcludes, scanExclude)
 		}),
 	}
 	scanCmd.Flags().IntVar(&scanWorkers, "workers", runtime.NumCPU(),
@@ -241,6 +252,10 @@ func newScanCommand() *cobra.Command {
 		"do not cross filesystem boundaries")
 	scanCmd.Flags().BoolVar(&scanNoDefaultExcludes, "no-default-excludes",
 		false, "enter the default excludes, the paths listed above")
+	// A string array, not a slice, so a comma in a path is not a
+	// separator.
+	scanCmd.Flags().StringArrayVar(&scanExclude, "exclude", nil,
+		"never enter or record `PATH` or anything below it; repeatable")
 
 	return scanCmd
 }
@@ -258,6 +273,18 @@ func checkScanWorkers(cmd *cobra.Command, workers int) error {
 	return fmt.Errorf("%w, got %d", errWorkersBelowOne, workers)
 }
 
+// rejectEmptyExcludeValue rejects an empty scan --exclude value. That
+// is a usage error reported in one line, as for --workers below 1.
+func rejectEmptyExcludeValue(cmd *cobra.Command, excludePaths []string) error {
+	if !slices.Contains(excludePaths, "") {
+		return nil
+	}
+
+	cmd.SilenceUsage = true
+
+	return errEmptyExcludeValue
+}
+
 // runE adapts a subcommand implementation, or the version print, to
 // cobra's RunE. Cobra prints the error and the command's usage text for
 // every error RunE returns, but a subcommand that ran and failed has no

internal/sfdupes/cli_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // drives the unexported run
 
 import (
 	"bytes"
@@ -14,6 +14,8 @@ import (
 	"strconv"
 	"strings"
 	"testing"
+
+	"golang.org/x/sys/unix"
 )
 
 // usageMarker is the first line of cobra's usage text, which a usage
@@ -568,7 +570,8 @@ func TestRunNamesDefaultDatabase(t *testing.T) {
 	}{
 		cmdScan: {
 			run: func() error {
-				return runScan(t.Context(), []string{missing}, 1, false, false)
+				return runScan(t.Context(), []string{missing}, 1, false, false,
+					nil)
 			},
 			want: scanWant,
 		},
@@ -620,6 +623,7 @@ func TestRunUsageErrors(t *testing.T) {
 		"report with args":   {[]string{cmdReport, "x"}, usageMarker},
 		"trees with args":    {[]string{cmdTrees, "x"}, usageMarker},
 		"info with args":     {[]string{cmdInfo, "x"}, usageMarker},
+		"dwim with args":     {[]string{cmdDwim, "x"}, usageMarker},
 		"unknown flag":       {[]string{cmdScan, "--nope", "/"}, usageMarker},
 		"unknown subcommand": {[]string{"nope"}, "unknown command"},
 	}
@@ -709,6 +713,45 @@ func TestRunHelp(t *testing.T) {
 				args, got)
 		}
 	}
+
+	// The scan help lists --exclude with its argument.
+	var stdout, stderr bytes.Buffer
+
+	run([]string{cmdScan, "-h"}, &stdout, &stderr)
+
+	if !strings.Contains(stderr.String(), excludeFlag+" PATH") {
+		t.Errorf("scan -h stderr = %q, want %s PATH", stderr.String(),
+			excludeFlag)
+	}
+}
+
+func TestRunScanUsageLine(t *testing.T) {
+	t.Parallel()
+
+	// README §Subcommands shows this usage line for scan, and the help
+	// and the usage error both print it.
+	const want = "\n  sfdupes scan [--workers N] [-x] [--no-default-excludes] " +
+		"[--exclude PATH]... PATH...\n"
+
+	cases := map[string][]string{
+		"help":          {cmdScan, "-h"},
+		"missing paths": {cmdScan},
+	}
+
+	for name, args := range cases {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			var stdout, stderr bytes.Buffer
+
+			run(args, &stdout, &stderr)
+
+			if !strings.Contains(stderr.String(), want) {
+				t.Errorf("run(%v) stderr = %q, want the usage line %q",
+					args, stderr.String(), want)
+			}
+		})
+	}
 }
 
 func TestRunVersion(t *testing.T) {
@@ -797,7 +840,7 @@ func TestRunVersionWriteFailureIsFatal(t *testing.T) {
 func scanFixture(t *testing.T) []string {
 	t.Helper()
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
 
 	dupes := []string{
 		writeFile(t, dir, "one/a.bin", pattern(1, 300)),
@@ -867,7 +910,7 @@ func TestRunScanSucceedsDespiteWarnings(t *testing.T) {
 	path := testDBPath(t)
 	t.Setenv(databaseEnv, path)
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
 	writeFile(t, dir, "a.bin", pattern(1, 300))
 
 	// Same size as a.bin, so the scan reads it, and the read fails.
@@ -895,58 +938,278 @@ func TestRunScanSucceedsDespiteWarnings(t *testing.T) {
 	assertNoSidecars(t, path)
 }
 
-func TestRunScanSkipsSymlinkOperand(t *testing.T) {
+// realTempDir returns a new temporary directory by its real path, so
-	path := testDBPath(t)
+// tests hold where the temporary directory is reached through a
-	t.Setenv(databaseEnv, path)
+// symlink.
+func realTempDir(t *testing.T) string {
+	t.Helper()
 
-	dir := t.TempDir()
+	dir, err := filepath.EvalSymlinks(t.TempDir())
-	writeFile(t, dir, "target/sub/f", pattern(1, 10))
-
-	link := filepath.Join(dir, "link")
-
-	err := os.Symlink(filepath.Join(dir, "target"), link)
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	// Scanning a directory through the symlink stores a record beneath
+	return dir
-	// the symlink's own path for a file beneath its target.
-	scanOK(t, filepath.Join(link, "sub"))
-
-	assertOperandSkipped(t, path, link, "symlink",
-		filepath.Join(link, "sub", "f"))
 }
 
+// symlink creates link pointing at target.
+func symlink(t *testing.T, target, link string) {
+	t.Helper()
+
+	err := os.Symlink(target, link)
+	if err != nil {
+		t.Fatal(err)
+	}
+}
+
+// TestRunScanFollowsSymlinkOperand scans link, a symlink to target.
+// The operand is scanned at its real path, and its file is recorded
+// there.
+func TestRunScanFollowsSymlinkOperand(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	f := writeFile(t, dir, "target/sub/f", pattern(1, 10))
+
+	link := filepath.Join(dir, "link")
+	symlink(t, filepath.Join(dir, "target"), link)
+
+	stderr := scanOK(t, link)
+
+	summary := "scan: 1 files seen (1 added, 0 updated, 0 removed, " +
+		"0 unchanged), 0 skipped\n"
+	if !strings.Contains(stderr, summary) {
+		t.Errorf("stderr = %q, want %q", stderr, summary)
+	}
+
+	if got := storedPaths(t, path); !slices.Equal(got, []string{f}) {
+		t.Errorf("records = %q, want %q", got, []string{f})
+	}
+}
+
+// TestRunScanWalksOperandUnderSymlinkOperand scans link and link/sub,
+// with link a symlink to target. Both resolve under target, so
+// target/sub is walked once, as part of target.
 func TestRunScanWalksOperandUnderSymlinkOperand(t *testing.T) {
 	path := testDBPath(t)
 	t.Setenv(databaseEnv, path)
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
-	writeFile(t, dir, "target/sub/f", pattern(1, 10))
+	f := writeFile(t, dir, "target/sub/f", pattern(1, 10))
 
 	link := filepath.Join(dir, "link")
+	symlink(t, filepath.Join(dir, "target"), link)
 
-	err := os.Symlink(filepath.Join(dir, "target"), link)
+	stderr := scanOK(t, link, filepath.Join(link, "sub"))
+
+	summary := "scan: 1 files seen (1 added, 0 updated, 0 removed, " +
+		"0 unchanged), 0 skipped\n"
+	if !strings.Contains(stderr, summary) {
+		t.Errorf("stderr = %q, want %q", stderr, summary)
+	}
+
+	if got := storedPaths(t, path); !slices.Equal(got, []string{f}) {
+		t.Errorf("records = %q, want %q", got, []string{f})
+	}
+}
+
+// TestRunScanWalksSymlinkedOperandBelowOperand is the case of
+// https://git.eeqj.de/sneak/sfdupes/issues/144: scan dir dir/link/sub,
+// with dir/link a symlink to a directory outside dir. The walk of dir
+// does not follow link, so dir/link/sub is walked on its own at its
+// real path, and the record an earlier scan stored there survives.
+func TestRunScanWalksSymlinkedOperandBelowOperand(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	a := writeFile(t, dir, "a", pattern(1, 10))
+
+	outside := realTempDir(t)
+	f := writeFile(t, outside, "sub/f", pattern(2, 10))
+
+	symlink(t, outside, filepath.Join(dir, "link"))
+
+	scanOK(t, filepath.Join(outside, "sub"))
+
+	stderr := scanOK(t, dir, filepath.Join(dir, "link", "sub"))
+
+	summary := "scan: 2 files seen (1 added, 0 updated, 0 removed, " +
+		"1 unchanged), 0 skipped\n"
+	if !strings.Contains(stderr, summary) {
+		t.Errorf("stderr = %q, want %q", stderr, summary)
+	}
+
+	want := []string{a, f}
+	slices.Sort(want)
+
+	if got := storedPaths(t, path); !slices.Equal(got, want) {
+		t.Errorf("records = %q, want %q", got, want)
+	}
+}
+
+// TestRunScanDropsSymlinkOperandInsideOperand scans dir and link, with
+// link, outside dir, a symlink to dir/real. link resolves inside dir,
+// so it is dropped as covered by dir, and its file is recorded once.
+func TestRunScanDropsSymlinkOperandInsideOperand(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	f := writeFile(t, dir, "real/f", pattern(1, 10))
+
+	link := filepath.Join(realTempDir(t), "link")
+	symlink(t, filepath.Join(dir, "real"), link)
+
+	stderr := scanOK(t, dir, link)
+
+	summary := "scan: 1 files seen (1 added, 0 updated, 0 removed, " +
+		"0 unchanged), 0 skipped\n"
+	if !strings.Contains(stderr, summary) {
+		t.Errorf("stderr = %q, want %q", stderr, summary)
+	}
+
+	if got := storedPaths(t, path); !slices.Equal(got, []string{f}) {
+		t.Errorf("records = %q, want %q", got, []string{f})
+	}
+}
+
+// TestRunScanDanglingSymlinkOperandIsFatal scans a valid directory and
+// a symlink whose target does not exist. The scan fails with exit 1 and
+// an error naming the symlink as given, before it walks anything, so no
+// database is created.
+func TestRunScanDanglingSymlinkOperandIsFatal(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	writeFile(t, dir, "f", pattern(1, 10))
+
+	dangling := filepath.Join(realTempDir(t), "dangling")
+	symlink(t, filepath.Join(dir, "missing"), dangling)
+
+	stdout := captureStdout(t)
+
+	var stderr bytes.Buffer
+
+	code := run([]string{cmdScan, dir, dangling}, os.Stdout, &stderr)
+	if code != exitFatal {
+		t.Errorf("run(scan %s %s) = %d, want %d", dir, dangling, code,
+			exitFatal)
+	}
+
+	assertFatalOutput(t, stderr.String(), stdout())
+
+	want := "sfdupes: resolve " + dangling + ": "
+	if !strings.Contains(stderr.String(), want) {
+		t.Errorf("stderr = %q, want %q", stderr.String(), want)
+	}
+
+	_, err := os.Stat(path)
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("database: stat error %v, want it missing", err)
+	}
+}
+
+// TestRunScanFromSymlinkedWorkingDirectoryRecordsRealPath runs scan .
+// after a cd through home/proj, a symlink to data/proj. The file is
+// recorded under data/proj, its real path.
+func TestRunScanFromSymlinkedWorkingDirectoryRecordsRealPath(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	f := writeFile(t, dir, "data/proj/f", pattern(1, 10))
+
+	err := os.Mkdir(filepath.Join(dir, "home"), 0o750)
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	// link is dropped as a symlink, but link/sub must still be scanned,
+	symlink(t, filepath.Join(dir, "data", "proj"),
-	// not dropped as lying under link.
+		filepath.Join(dir, "home", "proj"))
-	scanOK(t, link, filepath.Join(link, "sub"))
 
-	db, err := openDB(path, reportParams)
+	// t.Chdir also sets PWD to the path through the symlink, as a
+	// shell does after cd.
+	t.Chdir(filepath.Join(dir, "home", "proj"))
+
+	scanOK(t, ".")
+
+	if got := storedPaths(t, path); !slices.Equal(got, []string{f}) {
+		t.Errorf("records = %q, want %q", got, []string{f})
+	}
+}
+
+// TestRunScanFollowsSymlinkBeforeDotDot scans a path with a ".." after
+// home/proj, a symlink to data/proj: home/proj/.. from dir, and .. after
+// a cd through home/proj. As with realpath, the ".." counts from
+// data/proj, so data is scanned and home is not.
+func TestRunScanFollowsSymlinkBeforeDotDot(t *testing.T) {
+	for _, tc := range []struct{ wd, operand string }{
+		{"", "home/proj/.."},
+		{"home/proj", ".."},
+	} {
+		t.Run(tc.operand, func(t *testing.T) {
+			path := testDBPath(t)
+			t.Setenv(databaseEnv, path)
+
+			dir := realTempDir(t)
+			f := writeFile(t, dir, "data/proj/f", pattern(1, 10))
+			g := writeFile(t, dir, "data/g", pattern(2, 10))
+			writeFile(t, dir, "home/h", pattern(3, 10))
+			symlink(t, filepath.Join(dir, "data", "proj"),
+				filepath.Join(dir, "home", "proj"))
+
+			// t.Chdir also sets PWD to the path given, as a shell does
+			// after cd.
+			t.Chdir(filepath.Join(dir, tc.wd))
+
+			scanOK(t, tc.operand)
+
+			want := []string{f, g}
+			slices.Sort(want)
+
+			if got := storedPaths(t, path); !slices.Equal(got, want) {
+				t.Errorf("records = %q, want %q", got, want)
+			}
+		})
+	}
+}
+
+// TestRunScanSkipsSymlinkToFIFOOperand scans link, a symlink to a FIFO.
+// The scan warns about the FIFO by its real path, counts it as skipped,
+// and exits 0.
+func TestRunScanSkipsSymlinkToFIFOOperand(t *testing.T) {
+	t.Setenv(databaseEnv, testDBPath(t))
+
+	dir := realTempDir(t)
+	fifo := filepath.Join(dir, "fifo")
+
+	err := unix.Mkfifo(fifo, 0o600)
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	t.Cleanup(func() { _ = db.Close() })
+	link := filepath.Join(dir, "link")
+	symlink(t, fifo, link)
 
-	recordByPath(t, dbRecords(t, db), filepath.Join(link, "sub", "f"))
+	stderr := scanOK(t, link)
+
+	warning := "walk " + fifo + ": skipping FIFO operand\n"
+	if !strings.Contains(stderr, warning) {
+		t.Errorf("stderr = %q, want %q", stderr, warning)
+	}
+
+	summary := "scan: 0 files seen (0 added, 0 updated, 0 removed, " +
+		"0 unchanged), 1 skipped\n"
+	if !strings.Contains(stderr, summary) {
+		t.Errorf("stderr = %q, want %q", stderr, summary)
+	}
 }
 
 func TestRunScanSkipsZFSOperand(t *testing.T) {
-	zfs := filepath.Join(t.TempDir(), ".zfs")
+	zfs := filepath.Join(realTempDir(t), ".zfs")
 	f := writeFile(t, zfs, "snapshot/hourly/f", pattern(1, 10))
 
 	// No scan records a file inside a .zfs directory, so the record
@@ -997,7 +1260,7 @@ func TestRunScanSkipsOperandInsideZFS(t *testing.T) {
 	path := testDBPath(t)
 	t.Setenv(databaseEnv, path)
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
 	hourly := filepath.Join(dir, ".zfs", "snapshot", "hourly")
 	writeFile(t, hourly, "f", pattern(1, 10))
 	f := writeFile(t, dir, "f", pattern(2, 10))
@@ -1041,23 +1304,19 @@ func TestRunScanSkipsOperandInsideZFS(t *testing.T) {
 
 // TestRunScanSkipsOperandThroughSymlinkIntoZFS scans snaps/daily, where
 // snaps is a symlink to pool/.zfs/snapshot. The operand is warned about
-// by the path given, counted as skipped, and not walked.
+// by its real path, counted as skipped, and not walked.
 func TestRunScanSkipsOperandThroughSymlinkIntoZFS(t *testing.T) {
 	t.Setenv(databaseEnv, testDBPath(t))
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
 	writeFile(t, dir, "pool/.zfs/snapshot/daily/f", pattern(1, 10))
 
 	snaps := filepath.Join(dir, "snaps")
+	symlink(t, filepath.Join(dir, "pool", ".zfs", "snapshot"), snaps)
 
-	err := os.Symlink(filepath.Join(dir, "pool", ".zfs", "snapshot"), snaps)
+	stderr := scanOK(t, filepath.Join(snaps, "daily"))
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	daily := filepath.Join(snaps, "daily")
-	stderr := scanOK(t, daily)
 
+	daily := filepath.Join(dir, "pool", ".zfs", "snapshot", "daily")
 	warning := "walk " + daily + ": skipping operand inside a .zfs directory\n"
 	summary := "scan: 0 files seen (0 added, 0 updated, 0 removed, " +
 		"0 unchanged), 1 skipped\n"
@@ -1069,24 +1328,19 @@ func TestRunScanSkipsOperandThroughSymlinkIntoZFS(t *testing.T) {
 }
 
 // TestRunScanWalksOperandThroughSymlink scans data/daily, where data is
-// a symlink to a directory outside any .zfs directory. The operand is
+// a symlink to pool/data. The operand is scanned as usual, and its file
-// scanned as usual, and its file is recorded by the path given.
+// is recorded by its real path.
 func TestRunScanWalksOperandThroughSymlink(t *testing.T) {
 	path := testDBPath(t)
 	t.Setenv(databaseEnv, path)
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
-	writeFile(t, dir, "pool/data/daily/f", pattern(1, 10))
+	f := writeFile(t, dir, "pool/data/daily/f", pattern(1, 10))
 
 	data := filepath.Join(dir, "data")
+	symlink(t, filepath.Join(dir, "pool", "data"), data)
 
-	err := os.Symlink(filepath.Join(dir, "pool", "data"), data)
+	stderr := scanOK(t, filepath.Join(data, "daily"))
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	daily := filepath.Join(data, "daily")
-	stderr := scanOK(t, daily)
 
 	summary := "scan: 1 files seen (1 added, 0 updated, 0 removed, " +
 		"0 unchanged), 0 skipped\n"
@@ -1094,14 +1348,9 @@ func TestRunScanWalksOperandThroughSymlink(t *testing.T) {
 		t.Errorf("stderr = %q, want %q", stderr, summary)
 	}
 
-	db, err := openDB(path, reportParams)
+	if got := storedPaths(t, path); !slices.Equal(got, []string{f}) {
-	if err != nil {
+		t.Errorf("records = %q, want %q", got, []string{f})
-		t.Fatal(err)
 	}
-
-	t.Cleanup(func() { _ = db.Close() })
-
-	recordByPath(t, dbRecords(t, db), filepath.Join(daily, "f"))
 }
 
 func TestRunReportSucceeds(t *testing.T) {

internal/sfdupes/db.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes
 
 import (
 	"context"
@@ -146,33 +146,44 @@ var errHashSize = errors.New("hash is not 32 bytes")
 
 // databasePath resolves the database location: SFDUPES_DATABASE when
 // set and non-empty, the compiled-in default otherwise. It returns the
-// path, with the working directory, its symlinks resolved, put in front
+// path, made absolute by absolutePath, and where the path came from,
-// of it when it is relative, and where the path came from, "default" or
+// "default" or "from SFDUPES_DATABASE", for the line each subcommand
-// "from SFDUPES_DATABASE", for the line each subcommand prints first on
+// prints first on stderr.
-// stderr.
 func databasePath() (string, string, error) {
 	path, source := defaultDatabasePath, "default"
 	if p := os.Getenv(databaseEnv); p != "" {
 		path, source = p, "from "+databaseEnv
 	}
 
-	if filepath.IsAbs(path) {
+	abs, err := absolutePath(path)
-		return path, source, nil
-	}
-
-	// The working directory as the kernel reports it, with no symlink
-	// in it, as README "Database" says. os.Getwd may return PWD
-	// instead, which can name it through a symlink.
-	wd, err := unix.Getwd()
 	if err != nil {
 		return "", "", fmt.Errorf("resolve database path %s: %w", path, err)
 	}
 
-	// Not filepath.Abs or filepath.Join: they clean the path, which
+	return abs, source, nil
-	// drops each ".." together with the directory name before it. When
+}
-	// that directory is a symlink, the cleaned path names another file.
+
+// absolutePath returns path unchanged when it is absolute, and
+// otherwise the working directory, a "/" and path. The working
+// directory is the one the kernel reports, with no symlink in it;
+// os.Getwd may return PWD instead, which can name it through a symlink.
+// The result is not cleaned, as filepath.Abs and filepath.Join would
+// clean it. Cleaning drops each ".." together with the directory name
+// before it. When that name is a symlink, the cleaned path names
+// another file than the one the system reaches by following the symlink
+// first.
+func absolutePath(path string) (string, error) {
+	if filepath.IsAbs(path) {
+		return path, nil
+	}
+
+	wd, err := unix.Getwd()
+	if err != nil {
+		return "", err
+	}
+
 	// The working directory ends in "/" only when it is the root.
-	return strings.TrimSuffix(wd, "/") + "/" + path, source, nil
+	return strings.TrimSuffix(wd, "/") + "/" + path, nil
 }
 
 // createDatabaseDirectory creates the directory the database at path

internal/sfdupes/db_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // calls the unexported database openers
 
 import (
 	"context"

internal/sfdupes/dwim.go

@@ -0,0 +1,38 @@
+package sfdupes
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"os"
+	"runtime"
+)
+
+// runDwim implements the dwim subcommand: it runs what "sfdupes scan
+// root" runs with no flags, then what "sfdupes trees" runs, and ends
+// with a line on stderr saying how to print the report again. The dwim
+// command passes "/" as root. It catches SIGINT and SIGTERM as scan
+// does, but only during the scan. An error from the scan, an interrupt
+// included, ends it before the report. During the report, a signal
+// ends it at once, as it ends trees.
+func runDwim(ctx context.Context, root string, stdout io.Writer) error {
+	scanCtx, stop := interruptContext(ctx)
+	err := runScan(scanCtx, []string{root}, runtime.NumCPU(), false, false,
+		nil)
+
+	stop()
+
+	if err != nil {
+		return err
+	}
+
+	err = runTrees(ctx, stdout)
+	if err != nil {
+		return err
+	}
+
+	fmt.Fprintln(os.Stderr,
+		`dwim: run "sfdupes trees" to print this report again`)
+
+	return nil
+}

internal/sfdupes/dwim_test.go

@@ -0,0 +1,83 @@
+package sfdupes //nolint:testpackage // calls the unexported runDwim
+
+import (
+	"bytes"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestRunDwimPrintsLargestTreesFirst(t *testing.T) {
+	// dwim scans its root, then prints the trees report, the larger of
+	// two duplicate-tree pairs first, and ends by saying how to print
+	// the report again.
+	t.Setenv(databaseEnv, testDBPath(t))
+
+	root := realTempDir(t)
+
+	small := []string{
+		filepath.Dir(writeFile(t, root, "a1/f.bin", pattern(1, 300))),
+		filepath.Dir(writeFile(t, root, "a2/f.bin", pattern(1, 300))),
+	}
+	large := []string{
+		filepath.Dir(writeFile(t, root, "b1/f.bin", pattern(2, 5000))),
+		filepath.Dir(writeFile(t, root, "b2/f.bin", pattern(2, 5000))),
+	}
+
+	var stdout bytes.Buffer
+
+	stderr := captureStderr(t)
+
+	err := runDwim(t.Context(), root, &stdout)
+	if err != nil {
+		t.Fatalf("runDwim: %v; stderr: %s", err, stderr())
+	}
+
+	lines := strings.Split(strings.TrimSuffix(stdout.String(), "\n"), "\n")
+	if len(lines) != 3 {
+		t.Fatalf("stdout = %q, want a header and two groups",
+			stdout.String())
+	}
+
+	wantFirst := large[0] + "\t" + large[1] + "\t1\t5000\t"
+	if !strings.HasPrefix(lines[1], wantFirst) {
+		t.Errorf("first group = %q, want it to start %q", lines[1], wantFirst)
+	}
+
+	wantSecond := small[0] + "\t" + small[1] + "\t1\t300\t"
+	if !strings.HasPrefix(lines[2], wantSecond) {
+		t.Errorf("second group = %q, want it to start %q",
+			lines[2], wantSecond)
+	}
+
+	last := "\ndwim: run \"sfdupes trees\" to print this report again\n"
+	if got := stderr(); !strings.HasSuffix(got, last) {
+		t.Errorf("stderr = %q, want it to end %q", got, last)
+	}
+}
+
+func TestRunDwimScanErrorPrintsNoReport(t *testing.T) {
+	// A fatal scan error ends dwim before the report: a root that does
+	// not exist fails the scan as it fails scan itself.
+	t.Setenv(databaseEnv, testDBPath(t))
+
+	missing := filepath.Join(realTempDir(t), "missing")
+
+	var stdout bytes.Buffer
+
+	stderr := captureStderr(t)
+
+	err := runDwim(t.Context(), missing, &stdout)
+	if err == nil {
+		t.Fatal("runDwim on a missing root: error = nil, want it fatal")
+	}
+
+	if got := stdout.String(); got != "" {
+		t.Errorf("stdout = %q, want no report", got)
+	}
+
+	if got := stderr(); strings.Contains(got, "trees:") ||
+		strings.Contains(got, "dwim:") {
+		t.Errorf("stderr = %q, want nothing from trees or dwim", got)
+	}
+}

internal/sfdupes/exclude.go

@@ -1,8 +1,9 @@
-package main
+package sfdupes
 
-// This file holds two rules for the walk (README "Rules for the walk"):
+// This file holds three rules for the walk (README "Rules for the
+// walk"): the paths given with --exclude, which the scan never enters;
 // the default excludes, the paths a scan's walk does not enter below an
-// operand unless scan is given --no-default-excludes, and the
+// operand unless scan is given --no-default-excludes; and the
 // filesystems that are not persistent, on which the scan never records
 // or reads anything.
 
@@ -12,6 +13,7 @@ import (
 	"fmt"
 	"io"
 	"io/fs"
+	"os"
 	"path/filepath"
 	"slices"
 	"strings"
@@ -43,15 +45,23 @@ const typesPerHelpLine = 6
 // readMountInfo reads.
 var errMountInfoLine = errors.New("unexpected line")
 
+// errExcludedOperand reports an operand at or below a path given with
+// --exclude.
+var errExcludedOperand = errors.New("operand excluded by --exclude")
+
 // walkRules holds the rules for the walk that a scan sets up as it
-// starts. excludes are the default excludes, or none with
+// starts. excludePaths are the paths given with --exclude, each with
-// --no-default-excludes. nonPersistent maps the device ID of every
+// the device ID and inode number of the file it names (see
-// mounted filesystem that is not persistent to its type; the scan
+// excludePathInodes), which the walk neither enters nor records,
-// records and reads nothing on one of those, whatever the flags.
+// whatever path leads to them. defaultExcludes are the default
-// databaseFiles are the scan's own database files, which the walk
+// excludes, or none with --no-default-excludes. nonPersistent maps the
-// passes over (see databaseFiles in db.go).
+// device ID of every mounted filesystem that is not persistent to its
+// type; the scan records and reads nothing on one of those, whatever
+// the flags. databaseFiles are the scan's own database files, which the
+// walk passes over (see databaseFiles in db.go).
 type walkRules struct {
-	excludes      []string
+	excludePaths    []fileRec
+	defaultExcludes []string
 	nonPersistent   map[uint64]string
 	databaseFiles   []fileRec
 }
@@ -97,6 +107,88 @@ func isDefaultExclude(path string, excludes []string) bool {
 	return false
 }
 
+// excludePathInodes returns each of paths, the --exclude values, as its
+// real path (see realPaths) with the device ID and inode number of the
+// file there. The walk compares what it reaches with those, as it does
+// with the database files, so it also leaves out an --exclude path
+// spelled another way, such as in another letter case on macOS.
+func excludePathInodes(paths []string) ([]fileRec, error) {
+	resolved, err := realPaths(paths)
+	if err != nil {
+		return nil, err
+	}
+
+	files := make([]fileRec, 0, len(resolved))
+
+	for _, path := range resolved {
+		fi, err := os.Lstat(path)
+		if err != nil {
+			return nil, err
+		}
+
+		dev, ino := inodeOfInfo(fi)
+		files = append(files, fileRec{path: path, dev: dev, ino: ino})
+	}
+
+	return files, nil
+}
+
+// matchingExcludePath returns the path of the one of excludePaths with
+// the device ID and inode number of fi, an lstat result, or "" when
+// there is none.
+func matchingExcludePath(fi fs.FileInfo, excludePaths []fileRec) string {
+	dev, ino := inodeOfInfo(fi)
+	f := fileRec{dev: dev, ino: ino}
+
+	for _, exclude := range excludePaths {
+		if sameInode(exclude, f) {
+			return exclude.path
+		}
+	}
+
+	return ""
+}
+
+// checkOperandsNotExcluded returns an error when one of roots, the
+// operands, or a directory above it is one of excludePaths, the
+// --exclude paths, by device ID and inode number. The error names the
+// operand and the --exclude path. The roots are absolute, so the
+// directories above each one end at /. A path here that lstat cannot
+// read is an error too.
+func checkOperandsNotExcluded(roots []string, excludePaths []fileRec) error {
+	for _, root := range roots {
+		for path := root; ; path = filepath.Dir(path) {
+			fi, err := os.Lstat(path)
+			if err != nil {
+				return fmt.Errorf("scan %s: %w", root, err)
+			}
+
+			exclude := matchingExcludePath(fi, excludePaths)
+			if exclude != "" {
+				return fmt.Errorf("scan %s: %w %s", root, errExcludedOperand,
+					exclude)
+			}
+
+			if path == "/" {
+				break
+			}
+		}
+	}
+
+	return nil
+}
+
+// excludePathLine returns the line the scan prints for path when fi,
+// its lstat, makes it one of rules.excludePaths, and "" when it does
+// not.
+func (rules walkRules) excludePathLine(path string, fi fs.FileInfo) string {
+	if matchingExcludePath(fi, rules.excludePaths) == "" {
+		return ""
+	}
+
+	return fmt.Sprintf("scan: skipped %s (--exclude)", path)
+}
+
 // nonPersistentFilesystemTypes returns the filesystem types that are
 // not persistent: filesystems that the kernel makes up or that are kept
 // in memory. devfs is the macOS filesystem of /dev; the others are

internal/sfdupes/exclude_darwin.go

@@ -1,6 +1,6 @@
 //go:build darwin
 
-package main
+package sfdupes
 
 import (
 	"slices"

internal/sfdupes/exclude_other.go

@@ -1,6 +1,6 @@
 //go:build !darwin
 
-package main
+package sfdupes
 
 import (
 	"fmt"

internal/sfdupes/exclude_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // checks the unexported defaultExcludes
 
 import (
 	"bytes"
@@ -14,6 +14,9 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+// excludeFlag is the scan flag that names a path the walk never enters.
+const excludeFlag = "--exclude"
+
 func TestDefaultExcludes(t *testing.T) {
 	t.Parallel()
 
@@ -188,7 +191,7 @@ func TestScanDoesNotEnterDefaultExclude(t *testing.T) {
 	unseen := writeFile(t, excluded, "new", pattern(2, 10))
 
 	st, err := syncScan(t.Context(), db, []string{dir}, 1, false,
-		walkRules{excludes: []string{excluded}})
+		walkRules{defaultExcludes: []string{excluded}})
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -234,7 +237,7 @@ func TestScanWalksOperandInDefaultExclude(t *testing.T) {
 			db := openTestDB(t)
 
 			st, err := syncScan(t.Context(), db, roots, 1, false,
-				walkRules{excludes: []string{excluded}})
+				walkRules{defaultExcludes: []string{excluded}})
 			if err != nil {
 				t.Fatal(err)
 			}
@@ -262,7 +265,7 @@ func TestScanNamesNoDefaultExcludeThatIsAnOperand(t *testing.T) {
 	f := writeFile(t, excluded, "f", pattern(1, 10))
 
 	st, err := syncScan(t.Context(), db, []string{dir, excluded}, 1, false,
-		walkRules{excludes: []string{excluded}})
+		walkRules{defaultExcludes: []string{excluded}})
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -286,7 +289,7 @@ func TestRunScanTempDirOperand(t *testing.T) {
 	path := testDBPath(t)
 	t.Setenv(databaseEnv, path)
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
 	if !underAnyRoot(dir, defaultExcludes(runtime.GOOS)) {
 		t.Skip(dir + " is not inside a default exclude")
 	}
@@ -356,6 +359,217 @@ func TestRunScanNoDefaultExcludes(t *testing.T) {
 	assertRecorded(t, path, f, true)
 }
 
+// TestRunScanExcludeNested scans a tree through the command with an
+// --exclude path two levels below the operand. The walk records nothing
+// at or below it and says so. The record an earlier scan left there is
+// kept although its file is gone.
+func TestRunScanExcludeNested(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	excluded := filepath.Join(dir, "a", "b")
+	kept := writeFile(t, dir, "a/f", pattern(1, 10))
+	gone := writeFile(t, excluded, "gone", pattern(2, 10))
+
+	scanOK(t, dir)
+
+	err := os.Remove(gone)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	unseen := writeFile(t, excluded, "c/new", pattern(3, 10))
+
+	stderr := scanOK(t, excludeFlag, excluded, dir)
+
+	line := "scan: skipped " + excluded + " (--exclude)\n"
+	if !strings.Contains(stderr, line) {
+		t.Errorf("stderr = %q, want %q", stderr, line)
+	}
+
+	assertRecorded(t, path, kept, true)
+	assertRecorded(t, path, gone, true)
+	assertRecorded(t, path, unseen, false)
+}
+
+// TestRunScanExcludeRepeated gives --exclude twice, once for a
+// directory and once for a file, the file through a symlink to it. The
+// walk records neither, and records the file beside them.
+func TestRunScanExcludeRepeated(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	inDir := writeFile(t, dir, "skip/f", pattern(1, 10))
+	file := writeFile(t, dir, "skip-file", pattern(2, 10))
+	kept := writeFile(t, dir, "keep", pattern(3, 10))
+
+	link := filepath.Join(t.TempDir(), "link")
+	symlink(t, file, link)
+
+	scanOK(t, excludeFlag, filepath.Dir(inDir), excludeFlag, link, dir)
+
+	assertRecorded(t, path, inDir, false)
+	assertRecorded(t, path, file, false)
+	assertRecorded(t, path, kept, true)
+}
+
+// TestRunScanExcludeReachedByAnotherPath gives --exclude a directory
+// through a symlink and a file through "..", so the walk reaches
+// neither by the path given. The walk also reaches the file through a
+// hard link. It records none of these, prints a line naming each, and
+// records the file beside them.
+func TestRunScanExcludeReachedByAnotherPath(t *testing.T) {
+	path := testDBPath(t)
+	t.Setenv(databaseEnv, path)
+
+	dir := realTempDir(t)
+	inDir := writeFile(t, dir, "a/cache/f", pattern(1, 10))
+	file := writeFile(t, dir, "a/big", pattern(2, 10))
+	kept := writeFile(t, dir, "keep", pattern(3, 10))
+
+	hardLink := filepath.Join(dir, "big-link")
+
+	err := os.Link(file, hardLink)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	link := filepath.Join(t.TempDir(), "link")
+	symlink(t, filepath.Join(dir, "a"), link)
+
+	// Joined as text, since filepath.Join would drop the "..".
+	dotDot := dir + "/a/cache/../big"
+
+	stderr := scanOK(t, excludeFlag, filepath.Join(link, "cache"),
+		excludeFlag, dotDot, dir)
+
+	for _, skipped := range []string{filepath.Dir(inDir), file, hardLink} {
+		line := "scan: skipped " + skipped + " (--exclude)\n"
+		if !strings.Contains(stderr, line) {
+			t.Errorf("stderr = %q, want %q", stderr, line)
+		}
+	}
+
+	assertRecorded(t, path, inDir, false)
+	assertRecorded(t, path, file, false)
+	assertRecorded(t, path, hardLink, false)
+	assertRecorded(t, path, kept, true)
+}
+
+// TestRunScanExcludedOperand checks that an operand at or below an
+// --exclude path is a fatal error that names both, and so is an
+// --exclude path that does not exist. An operand that is a hard link to
+// an --exclude file is at it. Neither scan writes a database.
+func TestRunScanExcludedOperand(t *testing.T) {
+	dir := realTempDir(t)
+	sub := filepath.Join(dir, "sub")
+	f := writeFile(t, sub, "f", pattern(1, 10))
+
+	hardLink := filepath.Join(dir, "f-link")
+
+	err := os.Link(f, hardLink)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	cases := map[string]struct {
+		args []string
+		want string
+	}{
+		"at it": {
+			[]string{excludeFlag, sub, sub},
+			"scan " + sub + ": operand excluded by --exclude " + sub + "\n",
+		},
+		"below it": {
+			[]string{excludeFlag, dir, sub},
+			"scan " + sub + ": operand excluded by --exclude " + dir + "\n",
+		},
+		"after an operand it does not exclude": {
+			[]string{dir, excludeFlag, sub, sub},
+			"scan " + sub + ": operand excluded by --exclude " + sub + "\n",
+		},
+		"a hard link to it": {
+			[]string{excludeFlag, f, hardLink},
+			"scan " + hardLink + ": operand excluded by --exclude " + f + "\n",
+		},
+		"exclude missing": {
+			[]string{excludeFlag, filepath.Join(dir, "nope"), dir},
+			"--exclude: resolve " + filepath.Join(dir, "nope") + ": ",
+		},
+	}
+
+	for name, tc := range cases {
+		t.Run(name, func(t *testing.T) {
+			path := testDBPath(t)
+			t.Setenv(databaseEnv, path)
+
+			stdout := captureStdout(t)
+			stderr := captureStderr(t)
+
+			code := run(append([]string{cmdScan}, tc.args...), os.Stdout,
+				os.Stderr)
+			if code != exitFatal {
+				t.Errorf("run(scan %q) = %d, want %d", tc.args, code,
+					exitFatal)
+			}
+
+			assertFatalOutput(t, stderr(), stdout())
+
+			if !strings.Contains(stderr(), "sfdupes: "+tc.want) {
+				t.Errorf("stderr = %q, want %q", stderr(), tc.want)
+			}
+
+			_, err := os.Lstat(path)
+			if !errors.Is(err, os.ErrNotExist) {
+				t.Errorf("lstat %s: %v, want no database", path, err)
+			}
+		})
+	}
+}
+
+// TestRunScanRejectsEmptyExclude checks that an empty --exclude value,
+// given as its own argument, after "=" or after another --exclude, is
+// a usage error reported in one line before anything is scanned
+// (README "Settings").
+func TestRunScanRejectsEmptyExclude(t *testing.T) {
+	dir := t.TempDir()
+
+	cases := map[string][]string{
+		"separate":     {excludeFlag, "", dir},
+		"joined by =":  {excludeFlag + "=", dir},
+		"after a path": {excludeFlag, t.TempDir(), excludeFlag, "", dir},
+	}
+
+	for name, args := range cases {
+		t.Run(name, func(t *testing.T) {
+			path := testDBPath(t)
+			t.Setenv(databaseEnv, path)
+
+			var stdout, stderr bytes.Buffer
+
+			code := run(append([]string{cmdScan}, args...), &stdout,
+				&stderr)
+			if code != exitUsage {
+				t.Errorf("run(scan %q) = %d, want %d", args, code,
+					exitUsage)
+			}
+
+			want := "Error: --exclude must not be empty\n"
+			if stdout.String() != "" || stderr.String() != want {
+				t.Errorf("stdout, stderr = %q, %q, want \"\", %q",
+					stdout.String(), stderr.String(), want)
+			}
+
+			_, err := os.Lstat(path)
+			if !errors.Is(err, os.ErrNotExist) {
+				t.Errorf("lstat %s: %v, want no database", path, err)
+			}
+		})
+	}
+}
+
 // TestRunScanSkipsNonPersistentOperand scans, through the command, a
 // directory on a filesystem that is not persistent, with and without
 // --no-default-excludes: either way it is skipped, with a line saying

internal/sfdupes/info.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes
 
 import (
 	"context"

internal/sfdupes/info_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // seeds the database with unexported scanRec
 
 import (
 	"fmt"

internal/sfdupes/progress.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes
 
 import (
 	"fmt"

internal/sfdupes/progress_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // drives the unexported startProgress
 
 import (
 	"fmt"
@@ -642,9 +642,9 @@ func TestTerminalStalledLines(t *testing.T) {
 		{label: scanPhase("hash"), stalledOn: hashStalledOn},
 		{label: scanPhase("update"), stalledOn: "db"},
 		{label: treesPhase("read"), stalledOn: "db"},
-		{label: treesPhase("group"), stalledOn: "cpu"},
+		{label: treesPhase("group"), stalledOn: "cpu/disk"},
 		{label: treesPhase("write"), stalledOn: "out"},
-		{label: reportPhase("read"), stalledOn: "db/out"},
+		{label: reportPhase("read"), stalledOn: "db/disk/out"},
 	}
 
 	for _, params := range phases {

internal/sfdupes/report.go

@@ -1,14 +1,17 @@
-package main
+package sfdupes
 
 import (
 	"context"
 	"crypto/sha256"
 	"database/sql"
+	"errors"
 	"fmt"
 	"io"
+	"io/fs"
 	"os"
 	"slices"
 	"strings"
+	"syscall"
 	"time"
 )
 
@@ -41,10 +44,10 @@ func reportPhase(name string) string {
 }
 
 // runReport implements the report subcommand: it prints the file-level
-// duplicates report as TSV on stdout. It never touches the scanned
+// duplicates report as TSV on stdout. Of the scanned filesystem it
-// filesystem; its only I/O is the database, stdout, and stderr. It
+// only checks, with pathExists, that each path in a duplicate group is
-// names the database on stderr before it opens it. Any database
+// still there. It names the database on stderr before it opens it. Any
-// problem, including a missing database, is fatal.
+// database problem, including a missing database, is fatal.
 //
 // The work runs in two phases, each under a progress display of its
 // own on stderr: open (open the database) and read (read the number of
@@ -77,13 +80,14 @@ func runReport(ctx context.Context, stdout io.Writer) error {
 
 	defer func() { _ = db.Close() }()
 
-	// The read phase writes the report to stdout while it reads the
+	// The read phase checks the paths of each duplicate group on disk
-	// records, so it may wait on either.
+	// and writes the report to stdout while it reads the records, so it
+	// may wait on any of the three.
 	prog = newProgress(progressParams{
 		label: reportPhase("read"), unit: "records", total: -1,
-		waitingFor: countingRecords, stalledOn: "db/out",
+		waitingFor: countingRecords, stalledOn: "db/disk/out",
 	})
-	report, err := readReport(ctx, db, dbPath, stdout, prog)
+	report, err := readReport(ctx, db, dbPath, stdout, pathExists, prog)
 
 	prog.finish()
 
@@ -104,14 +108,15 @@ func runReport(ctx context.Context, stdout io.Writer) error {
 // dbPath, as its total, then reads every record in signature order,
 // counting each on prog, and writes the report to stdout: the header
 // line, then the rows of each size's duplicate groups once that size is
-// read. stdout may be the terminal prog is drawn on, so the lines go
+// read. A path for which exists returns false is left out of its group.
-// through prog's writer. It returns what it wrote, with the totals for
+// stdout may be the terminal prog is drawn on, so the lines go through
-// the summary line.
+// prog's writer. It returns what it wrote, with the totals for the
+// summary line.
 func readReport(ctx context.Context, db *sql.DB, dbPath string,
-	stdout io.Writer, prog *progress,
+	stdout io.Writer, exists func(path string) bool, prog *progress,
 ) (*reportWriter, error) {
 	out := prog.writer(stdout)
-	report := &reportWriter{out: out}
+	report := &reportWriter{out: out, exists: exists}
 
 	_, err := fmt.Fprintln(out, "first\tdupe\tsize")
 	if err != nil {
@@ -158,6 +163,10 @@ func readReport(ctx context.Context, db *sql.DB, dbPath string,
 type reportWriter struct {
 	out io.Writer
 
+	// exists reports whether a path is still on disk. A path that is
+	// gone is left out of its group.
+	exists func(path string) bool
+
 	sig    scanRec    // the signature being read; its path is not used
 	paths  []string   // the paths read so far with sig
 	groups [][]string // the duplicate groups of sig.size, each sorted
@@ -194,11 +203,20 @@ func (w *reportWriter) add(r scanRec) error {
 	return nil
 }
 
-// endGroup ends the paths read with w.sig: two or more are a duplicate
+// endGroup ends the paths read with w.sig: two or more that are still
-// group, held with its paths sorted until writeGroups writes it.
+// on disk are a duplicate group, held with its paths sorted until
+// writeGroups writes it. Empty files form no group, since removing one
+// frees nothing. Only the paths of a possible group are checked, never
+// a path with no duplicate or an empty file.
 func (w *reportWriter) endGroup() {
-	if len(w.paths) < minGroupSize {
+	if len(w.paths) >= minGroupSize && w.sig.size > 0 {
-		// A path with no duplicate is dropped, and its slice reused.
+		w.paths = slices.DeleteFunc(w.paths, func(p string) bool {
+			return !w.exists(p)
+		})
+	}
+
+	if len(w.paths) < minGroupSize || w.sig.size == 0 {
+		// The paths are dropped, and their slice reused.
 		w.paths = w.paths[:0]
 
 		return
@@ -245,6 +263,16 @@ func (w *reportWriter) finish() error {
 	return w.writeGroups()
 }
 
+// pathExists reports whether path is still on disk, for report and
+// trees. It is false only when lstat finds that the path, or a
+// directory above it, is gone. Any other error counts as present, so
+// a user who may not look inside a directory still sees its copies.
+func pathExists(path string) bool {
+	_, err := os.Lstat(path)
+
+	return !errors.Is(err, fs.ErrNotExist) && !errors.Is(err, syscall.ENOTDIR)
+}
+
 // escapePath returns a path as it is written in a report column (README
 // "Report output format"): a backslash, tab, newline or carriage return
 // becomes \\, \t, \n or \r, and every other byte is kept as it is.

internal/sfdupes/report_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // checks the unexported dupeGroup results
 
 import (
 	"bytes"
@@ -72,15 +72,37 @@ type dupeGroup struct {
 	paths []string
 }
 
+// everyPathExists stands in for pathExists where a test's records name
+// made-up paths that are not on disk.
+func everyPathExists(string) bool { return true }
+
+// recsOnDisk returns a new directory and recs with each path moved
+// under it, after creating an empty file at each moved path: report and
+// trees then find every path still on disk.
+func recsOnDisk(t *testing.T, recs []scanRec) (string, []scanRec) {
+	t.Helper()
+
+	dir := realTempDir(t)
+	moved := make([]scanRec, len(recs))
+
+	for i, r := range recs {
+		r.path = writeFile(t, dir, r.path, nil)
+		moved[i] = r
+	}
+
+	return dir, moved
+}
+
 // dupeGroups returns the duplicate groups report reads from db, in
 // report order, from the rows it writes: a group's rows come together,
-// each naming the group's first path.
+// each naming the group's first path. Every path counts as still on
+// disk.
 func dupeGroups(t *testing.T, db *sql.DB) []dupeGroup {
 	t.Helper()
 
 	var out bytes.Buffer
 
-	_, err := readReport(t.Context(), db, "", &out, nil)
+	_, err := readReport(t.Context(), db, "", &out, everyPathExists, nil)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -127,7 +149,8 @@ func dupeGroupsOf(t *testing.T, recs []scanRec) []dupeGroup {
 }
 
 func TestRunReportEscapesPaths(t *testing.T) {
-	t.Setenv(databaseEnv, seedDatabase(t, awkwardPairRecs()))
+	dir, recs := recsOnDisk(t, awkwardPairRecs())
+	t.Setenv(databaseEnv, seedDatabase(t, recs))
 
 	var stdout, stderr bytes.Buffer
 
@@ -138,7 +161,7 @@ func TestRunReportEscapesPaths(t *testing.T) {
 	}
 
 	want := "first\tdupe\tsize\n" +
-		`/d/\tone\ntwo\rthree\\four/f` + "\t/d/A/f\t5\n"
+		dir + `/d/\tone\ntwo\rthree\\four/f` + "\t" + dir + "/d/A/f\t5\n"
 	if got := stdout.String(); got != want {
 		t.Errorf("stdout = %q, want %q", got, want)
 	}
@@ -172,7 +195,8 @@ func TestReportStdoutFailsWhileReading(t *testing.T) {
 	// Counts the records read, and draws nothing.
 	prog := &progress{now: time.Now}
 
-	_, err = readReport(t.Context(), db, "", failingWriter{}, prog)
+	_, err = readReport(t.Context(), db, "", failingWriter{},
+		everyPathExists, prog)
 	if !errors.Is(err, errWriteFailed) ||
 		!strings.HasPrefix(err.Error(), "write stdout: ") {
 		t.Errorf("error = %v, want write stdout: %v", err, errWriteFailed)
@@ -203,7 +227,8 @@ const smokeReport = "first\tdupe\tsize\n" +
 // it ends; read counts the 6 records before it reads them. The summary
 // comes last, and stdout holds the report alone.
 func TestRunReportProgressNamesEachPhase(t *testing.T) {
-	path := seedDatabase(t, smokeTreeRecs())
+	dir, recs := recsOnDisk(t, smokeTreeRecs())
+	path := seedDatabase(t, recs)
 	t.Setenv(databaseEnv, path)
 
 	stderr := captureStderr(t)
@@ -216,8 +241,9 @@ func TestRunReportProgressNamesEachPhase(t *testing.T) {
 			code, exitOK, stderr())
 	}
 
-	if got := stdout.String(); got != smokeReport {
+	report := strings.ReplaceAll(smokeReport, "/d/", dir+"/d/")
-		t.Errorf("stdout = %q, want %q", got, smokeReport)
+	if got := stdout.String(); got != report {
+		t.Errorf("stdout = %q, want %q", got, report)
 	}
 
 	// How each line starts; the rate at the end of a phase varies.
@@ -266,7 +292,7 @@ func TestReadReportOnTerminal(t *testing.T) {
 		waitingFor: countingRecords,
 	}, terminalWidth, time.Hour, time.Now)
 
-	_, err = readReport(t.Context(), db, "", os.Stderr, p)
+	_, err = readReport(t.Context(), db, "", os.Stderr, everyPathExists, p)
 
 	p.finish()
 
@@ -301,6 +327,8 @@ func TestRunReportsIgnoreInsertionOrder(t *testing.T) {
 		scanRec{size: 50, head: a, tail: a, content: a, path: "/x/1"},
 	)
 
+	_, recs = recsOnDisk(t, recs)
+
 	reversed := slices.Clone(recs)
 	slices.Reverse(reversed)
 
@@ -434,6 +462,140 @@ func TestDupeGroups(t *testing.T) {
 	}
 }
 
+// TestDupeGroupsLeavesOutEmptyFiles checks that empty files form no
+// duplicate group, while 1-byte files still do.
+func TestDupeGroupsLeavesOutEmptyFiles(t *testing.T) {
+	t.Parallel()
+
+	e, x := sha256.Sum256(nil), testHash("x")
+	recs := []scanRec{
+		{size: 0, head: e, tail: e, content: e, path: "/a/.gitkeep"},
+		{size: 0, head: e, tail: e, content: e, path: "/b/.gitkeep"},
+		{size: 0, head: e, tail: e, content: e, path: "/c/__init__.py"},
+		{size: 1, head: x, tail: x, content: x, path: "/tiny1"},
+		{size: 1, head: x, tail: x, content: x, path: "/tiny2"},
+	}
+
+	groups := dupeGroupsOf(t, recs)
+	if len(groups) != 1 || groups[0].size != 1 ||
+		!slices.Equal(groups[0].paths, []string{"/tiny1", "/tiny2"}) {
+		t.Fatalf("groups = %+v, want only /tiny1 /tiny2 of size 1", groups)
+	}
+}
+
+// TestReadReportChecksOnlyPossibleDuplicates checks that report asks
+// whether a path is on disk only when its signature has two or more
+// records, and never for an empty file.
+func TestReadReportChecksOnlyPossibleDuplicates(t *testing.T) {
+	t.Parallel()
+
+	a, u, e := testHash("a"), testHash("u"), sha256.Sum256(nil)
+	recs := []scanRec{
+		{size: 9, head: a, tail: a, content: a, path: "/copy/1"},
+		{size: 9, head: a, tail: a, content: a, path: "/copy/2"},
+		{size: 9, head: a, tail: a, content: a, path: "/copy/3"},
+		{size: 9, head: u, tail: u, content: u, path: "/lonely"},
+		{size: 0, head: e, tail: e, content: e, path: "/empty1"},
+		{size: 0, head: e, tail: e, content: e, path: "/empty2"},
+	}
+
+	db := openTestDB(t)
+
+	err := applyChanges(t.Context(), db, recs, nil, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	var checked []string
+
+	exists := func(path string) bool {
+		checked = append(checked, path)
+
+		return true
+	}
+
+	_, err = readReport(t.Context(), db, "", io.Discard, exists, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	slices.Sort(checked)
+
+	want := []string{"/copy/1", "/copy/2", "/copy/3"}
+	if !slices.Equal(checked, want) {
+		t.Errorf("paths checked on disk = %q, want %q", checked, want)
+	}
+}
+
+// TestReadReportDropsGroupLeftWithOnePath checks that a group of two
+// paths, one of them gone, is no duplicate group. If kept, it would
+// write no row, but the summary line would count it.
+func TestReadReportDropsGroupLeftWithOnePath(t *testing.T) {
+	t.Parallel()
+
+	a := testHash("a")
+	recs := []scanRec{
+		{size: 9, head: a, tail: a, content: a, path: "/a/kept"},
+		{size: 9, head: a, tail: a, content: a, path: "/a/gone"},
+	}
+
+	db := openTestDB(t)
+
+	err := applyChanges(t.Context(), db, recs, nil, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	exists := func(path string) bool { return path != "/a/gone" }
+
+	report, err := readReport(t.Context(), db, "", io.Discard, exists, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if report.dupeGroups != 0 {
+		t.Errorf("duplicate groups = %d, want 0", report.dupeGroups)
+	}
+}
+
+// TestReadReportKeepsGroupLeftWithTwoPaths checks that a group of three
+// paths, one of them gone, is a duplicate group of the other two.
+func TestReadReportKeepsGroupLeftWithTwoPaths(t *testing.T) {
+	t.Parallel()
+
+	a := testHash("a")
+	recs := []scanRec{
+		{size: 9, head: a, tail: a, content: a, path: "/b/kept1"},
+		{size: 9, head: a, tail: a, content: a, path: "/b/gone"},
+		{size: 9, head: a, tail: a, content: a, path: "/b/kept2"},
+	}
+
+	db := openTestDB(t)
+
+	err := applyChanges(t.Context(), db, recs, nil, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	exists := func(path string) bool { return path != "/b/gone" }
+
+	var out bytes.Buffer
+
+	report, err := readReport(t.Context(), db, "", &out, exists, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if report.dupeGroups != 1 {
+		t.Errorf("duplicate groups = %d, want 1", report.dupeGroups)
+	}
+
+	want := "first\tdupe\tsize\n/b/kept1\t/b/kept2\t9\n"
+	if got := out.String(); got != want {
+		t.Errorf("stdout = %q, want %q", got, want)
+	}
+}
+
 func TestDupeGroupsContentSeparates(t *testing.T) {
 	t.Parallel()
 

internal/sfdupes/scan.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes
 
 import (
 	"cmp"
@@ -102,40 +102,44 @@ func scanPhase(name string) string {
 // scan command rejects anything less. Before the lock is taken, the
 // scan lists the filesystems that are not persistent, on which the walk
 // records nothing; unless noDefaultExcludes is set, the walk does not
-// enter the default excludes either. Once the database is open, the
+// enter the default excludes either. The walk neither enters nor
-// scan notes its files, which the walk passes over.
+// records the paths in excludePaths, the --exclude values; an operand
+// at or below one is a fatal error. Once the database is open, the scan
+// notes its files, which the walk passes over.
 func runScan(ctx context.Context, roots []string, workers int,
-	oneFS, noDefaultExcludes bool,
+	oneFS, noDefaultExcludes bool, excludePaths []string,
 ) error {
 	dbPath, source, err := databasePath()
 	if err != nil {
 		return err
 	}
 
-	// A database file missing now is created by lockScanDatabase below.
+	printDatabasePath(dbPath, source)
-	what := "database"
 
-	_, err = os.Stat(dbPath)
+	roots, err = realPaths(roots)
-	if errors.Is(err, fs.ErrNotExist) {
-		what = "new database"
-	}
-
-	fmt.Fprintf(os.Stderr, "scan: %s %s (%s)\n", what, escapePath(dbPath),
-		source)
-
-	roots, err = resolveRoots(roots)
 	if err != nil {
 		return err
 	}
 
-	nonPersistent, err := nonPersistentFilesystems()
+	var rules walkRules
+
+	rules.excludePaths, err = excludePathInodes(excludePaths)
+	if err != nil {
+		return fmt.Errorf("--exclude: %w", err)
+	}
+
+	err = checkOperandsNotExcluded(roots, rules.excludePaths)
+	if err != nil {
+		return err
+	}
+
+	rules.nonPersistent, err = nonPersistentFilesystems()
 	if err != nil {
 		return fmt.Errorf("list the mounted filesystems: %w", err)
 	}
 
-	rules := walkRules{nonPersistent: nonPersistent}
 	if !noDefaultExcludes {
-		rules.excludes = defaultExcludes(runtime.GOOS)
+		rules.defaultExcludes = defaultExcludes(runtime.GOOS)
 	}
 
 	lock, err := lockScanDatabase(dbPath)
@@ -181,6 +185,21 @@ func runScan(ctx context.Context, roots []string, workers int,
 	return nil
 }
 
+// printDatabasePath prints on stderr the line that names the database
+// at dbPath, with source, saying whether it is new. A database file
+// missing now is new: lockScanDatabase creates it.
+func printDatabasePath(dbPath, source string) {
+	what := "database"
+
+	_, err := os.Stat(dbPath)
+	if errors.Is(err, fs.ErrNotExist) {
+		what = "new database"
+	}
+
+	fmt.Fprintf(os.Stderr, "scan: %s %s (%s)\n", what, escapePath(dbPath),
+		source)
+}
+
 // interruptContext returns a copy of ctx that the first SIGINT or
 // SIGTERM cancels; the scan command runs the scan under it. stop
 // releases the signals.
@@ -209,29 +228,31 @@ func interrupted(walked int) error {
 	return errInterrupted
 }
 
-// resolveRoots converts each PATH operand to an absolute, lexically
+// realPaths returns the real path of each of paths, the PATH operands
-// cleaned path (symlinks are not resolved) and verifies that it
+// or the --exclude values: absolute, with every symlink in it followed.
-// exists. Database records are keyed by absolute path, so scan results
+// Database records are keyed by that path, so scan results depend
-// must not depend on the working directory.
+// neither on the working directory nor on the symlinks an operand was
-func resolveRoots(roots []string) ([]string, error) {
+// given through. A path that cannot be resolved, because it or a
-	abs := make([]string, 0, len(roots))
+// symlink's target does not exist or cannot be read, is a fatal error
+// before any scanning.
+func realPaths(paths []string) ([]string, error) {
+	resolved := make([]string, 0, len(paths))
 
-	for _, root := range roots {
+	for _, path := range paths {
-		a, err := filepath.Abs(root)
+		a, err := absolutePath(path)
 		if err != nil {
-			return nil, fmt.Errorf("resolve %s: %w", root, err)
+			return nil, fmt.Errorf("resolve %s: %w", path, err)
 		}
 
-		// A nonexistent operand is a fatal error before any scanning.
+		r, err := filepath.EvalSymlinks(a)
-		_, err = os.Lstat(a)
 		if err != nil {
-			return nil, err
+			return nil, fmt.Errorf("resolve %s: %w", path, err)
 		}
 
-		abs = append(abs, a)
+		resolved = append(resolved, r)
 	}
 
-	return abs, nil
+	return resolved, nil
 }
 
 // pruneRoots drops operands already covered by another operand:
@@ -294,7 +315,7 @@ func walkReaches(above, r string, oneFS bool, rules walkRules,
 	}
 
 	for ; dir != above; dir = filepath.Dir(dir) {
-		if isDefaultExclude(dir, rules.excludes) {
+		if isDefaultExclude(dir, rules.defaultExcludes) {
 			return false
 		}
 
@@ -417,14 +438,16 @@ func (s *scanState) runPhases(ctx context.Context, roots []string,
 
 // walkableRoots returns the operands the walk can start from: regular
 // files, and directories not named .zfs, none of them inside a
-// directory named .zfs. Every other operand is warned about, counted as
+// directory named .zfs. The operands are real paths from realPaths,
-// skipped, and dropped. A dropped operand is no longer a root, so the
+// so these checks apply to what a symlink operand points at. Every
-// records stored beneath it are treated as if it had not been given:
+// other operand is warned about, counted as skipped, and dropped. A
-// those under no other root count as outside the roots and are not
+// dropped operand is no longer a root, so the records stored beneath it
-// deleted as unverified. An operand on a filesystem in rules that is not
+// are treated as if it had not been given: those under no other root
-// persistent is dropped the same way, with a line saying so that is not
+// count as outside the roots and are not deleted as unverified. An
-// a warning, and is not counted as skipped. An operand that fails lstat
+// operand on a filesystem in rules that is not persistent is dropped
-// here is kept, and the walk warns about it unless it no longer exists.
+// the same way, with a line saying so that is not a warning, and is not
+// counted as skipped. An operand that fails lstat here is kept, and the
+// walk warns about it unless it no longer exists.
 func (s *scanState) walkableRoots(roots []string, rules walkRules) []string {
 	kept := make([]string, 0, len(roots))
 
@@ -490,10 +513,11 @@ func (s *scanState) loadIndex(ctx context.Context, roots []string) error {
 // as verified. When its disk_size differs from the record's, it is
 // queued for the update phase, which writes the new disk_size. A record
 // at or below a path the walk left out, because -x stopped at it, it is
-// a default exclude, or it is not on a persistent filesystem, counts as
+// an --exclude path or a default exclude, or it is not on a persistent
-// outside the roots: it is not deleted. That does not hold below a root
+// filesystem, counts as outside the roots: it is not deleted. That does
-// at or inside that path, since that root's own walk verifies the
+// not hold below a root at or inside that path, since that root's own
-// records below it; nor is a line printed for such a root.
+// walk verifies the records below it; nor is a line printed for such a
+// root.
 func (s *scanState) walkPhase(events <-chan walkEvent, roots []string) {
 	// The records the last scan left under the roots estimate how many
 	// files the walk will find. A first scan has none, and no estimate.
@@ -542,7 +566,7 @@ func (s *scanState) walkPhase(events <-chan walkEvent, roots []string) {
 		prog.increment()
 
 		old, ok := s.existing[ev.rec.path]
-		if !ok || old.size != ev.rec.size || mtimeAfter(ev.rec.mtime, old.mtime) {
+		if !ok || old.size != ev.rec.size || !ev.rec.mtime.Equal(old.mtime) {
 			s.toHash = append(s.toHash, ev.rec)
 
 			continue
@@ -554,7 +578,7 @@ func (s *scanState) walkPhase(events <-chan walkEvent, roots []string) {
 
 		// The space a file takes on disk can change while its size and
 		// mtime stay the same. ZFS, for one, counts a new file's blocks
-		// only seconds after the write.
+		// up to seconds after the write.
 		if old.diskSize != ev.rec.diskSize {
 			s.diskSizes = append(s.diskSizes, ev.rec)
 		}
@@ -777,16 +801,6 @@ func (s *scanState) updatePhase(ctx context.Context) error {
 	return applyChanges(ctx, s.db, nil, deletes, prog)
 }
 
-// mtimeAfter reports whether mtime a is later than mtime b.
-// Not a.After(b): time.Time wraps an mtime past year 292 billion; Unix() undoes it.
-func mtimeAfter(a, b time.Time) bool {
-	if a.Unix() != b.Unix() {
-		return a.Unix() > b.Unix()
-	}
-
-	return a.Nanosecond() > b.Nanosecond()
-}
-
 // underAnyRoot reports whether path is any of the roots or lies under
 // one of them.
 func underAnyRoot(path string, roots []string) bool {
@@ -841,9 +855,9 @@ type dirJob struct {
 // file's statted record, a warning when fail is set, or a path the walk
 // left out: when otherFSDir is set, a directory -x did not enter
 // because it is on another filesystem than its operand, and when
-// leftOut is set, a default exclude or a directory or file on a
+// leftOut is set, an --exclude path, a default exclude or a directory or
-// filesystem that is not persistent, with warn holding the line that
+// file on a filesystem that is not persistent, with warn holding the
-// says so.
+// line that says so.
 type walkEvent struct {
 	rec        fileRec
 	warn       string
@@ -894,19 +908,13 @@ func sendEvent(ctx context.Context, events chan<- walkEvent,
 // operandWarning returns the one-line warning for an operand the walk
 // does not start from, naming the path and what it is, or "" for one it
 // does: a regular file, or a directory not named .zfs, that is not
-// inside a directory named .zfs, neither as given nor with its symlinks
+// inside a directory named .zfs. realPaths has already followed the
-// resolved. Symlinks are never followed, including as operands.
+// operand's symlinks, so a symlink here is an operand replaced by one
+// after it was resolved, and is not followed.
 func operandWarning(root string, fi fs.FileInfo) string {
 	// The walk never enters a .zfs directory, so it starts from nothing
-	// inside one either, not even through a symlink. An operand whose
+	// inside one either.
-	// symlinks cannot be resolved is checked as given.
+	if slices.Contains(strings.Split(filepath.Dir(root), "/"), ".zfs") {
-	resolved, err := filepath.EvalSymlinks(root)
-	if err != nil {
-		resolved = root
-	}
-
-	if slices.Contains(strings.Split(filepath.Dir(root), "/"), ".zfs") ||
-		slices.Contains(strings.Split(filepath.Dir(resolved), "/"), ".zfs") {
 		return fmt.Sprintf("walk %s: skipping operand inside a .zfs directory",
 			root)
 	}
@@ -1131,8 +1139,9 @@ func walkOneDir(ctx context.Context, job dirJob, oneFS bool,
 // no longer exists, and one that has stopped being a regular file is
 // skipped. A file on a filesystem in rules that is not persistent, as
 // through a bind mount of one file, is not recorded: it gets an event
-// naming it instead. One of the scan's own database files in rules gets
+// naming it instead, and so does a file given with --exclude, reached
-// no event at all.
+// by any path, a hard link to it included. One of the scan's own
+// database files in rules gets no event at all.
 func emitFile(ctx context.Context, p string, e fs.DirEntry,
 	rules walkRules, events chan<- walkEvent,
 ) {
@@ -1156,6 +1165,10 @@ func emitFile(ctx context.Context, p string, e fs.DirEntry,
 	}
 
 	line := rules.nonPersistentLine(p, info)
+	if line == "" {
+		line = rules.excludePathLine(p, info)
+	}
+
 	if line != "" {
 		sendEvent(ctx, events, walkEvent{leftOut: p, warn: line})
 
@@ -1180,11 +1193,11 @@ func emitFile(ctx context.Context, p string, e fs.DirEntry,
 // subdirJob applies the descent rules to directory p: never enter
 // .zfs (ZFS snapshot pseudo-dirs would list every file once per
 // snapshot), never enter a directory on a filesystem in rules that is
-// not persistent, nor a default exclude in rules, and with -x never
+// not persistent, nor an --exclude path or a default exclude in rules,
-// enter a directory on a different filesystem than its operand. For the
+// and with -x never enter a directory on a different filesystem than
-// last three it sends an event that names the directory, so the scan
+// its operand. For all but .zfs it sends an event that names the
-// keeps the records below it, except those below an operand at or
+// directory, so the scan keeps the records below it, except those below
-// inside it.
+// an operand at or inside it.
 func subdirJob(ctx context.Context, p string, e fs.DirEntry,
 	parent dirJob, oneFS bool, rules walkRules,
 	events chan<- walkEvent,
@@ -1209,13 +1222,17 @@ func subdirJob(ctx context.Context, p string, e fs.DirEntry,
 	}
 
 	line := rules.nonPersistentLine(p, info)
+	if line == "" {
+		line = rules.excludePathLine(p, info)
+	}
+
 	if line != "" {
 		sendEvent(ctx, events, walkEvent{leftOut: p, warn: line})
 
 		return dirJob{}, false
 	}
 
-	if isDefaultExclude(p, rules.excludes) {
+	if isDefaultExclude(p, rules.defaultExcludes) {
 		sendEvent(ctx, events, walkEvent{
 			leftOut: p,
 			warn:    fmt.Sprintf("scan: skipped %s (default exclude)", p),

internal/sfdupes/scan_dev_darwin.go

@@ -1,6 +1,6 @@
 //go:build darwin
 
-package main
+package sfdupes
 
 import "syscall"
 

internal/sfdupes/scan_dev_other.go

@@ -1,6 +1,6 @@
 //go:build !darwin
 
-package main
+package sfdupes
 
 import "syscall"
 

internal/sfdupes/scan_test.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes //nolint:testpackage // drives the unexported syncTree
 
 import (
 	"bytes"
@@ -43,10 +43,16 @@ func writeFile(t *testing.T, dir, name string, data []byte) string {
 }
 
 // diskSizeOf returns the bytes the file at path takes on disk, from its
-// own lstat: st_blocks counts 512-byte units. ZFS counts a new file's
+// own lstat: st_blocks counts 512-byte units. ZFS counts a write's
-// blocks only seconds after the write, and fsync does not hasten that.
+// blocks only when it commits the write, up to seconds later, and one
-// A test therefore calls unix.Sync, which waits for it, before the scan
+// commit can take the file through several growing sizes. So this value
-// whose disk_size it compares with this.
+// can change at any moment. Neither fsync nor unix.Sync hastens that,
+// but a unix.Sync call often takes long enough that the blocks are
+// counted when it returns. A test that needs the value to change waits
+// for the change. A test that compares it with a scan's disk_size reads
+// it before and after the scan. The rewrite test rescans until the two
+// reads match, then compares exactly. The other two tests fail only on
+// a disk_size outside the two reads.
 func diskSizeOf(t *testing.T, path string) int64 {
 	t.Helper()
 
@@ -1088,7 +1094,7 @@ func TestScanOneFilesystemWalksNestedFileOperand(t *testing.T) {
 func buildSmokeTree(t *testing.T) string {
 	t.Helper()
 
-	dir := t.TempDir()
+	dir := realTempDir(t)
 	one := pattern(10, 2000)
 	f1 := pattern(30, 3000)
 	f2 := pattern(40, 100)
@@ -1179,11 +1185,12 @@ func assertSmokeDupeGroups(t *testing.T, dir string, db *sql.DB) {
 	t.Helper()
 
 	groups := dupeGroups(t, db)
-	if len(groups) != 5 {
+	if len(groups) != 4 {
-		t.Fatalf("len(groups) = %d, want 5", len(groups))
+		t.Fatalf("len(groups) = %d, want 4", len(groups))
 	}
 
-	wantSizes := []int64{3000, 2000, 100, 1, 0}
+	// The empty files empty1 and empty2 form no group.
+	wantSizes := []int64{3000, 2000, 100, 1}
 	for i, g := range groups {
 		if g.size != wantSizes[i] {
 			t.Errorf("groups[%d].size = %d, want %d",
@@ -1208,7 +1215,7 @@ func assertSmokeTreeGroups(t *testing.T, dir string, db *sql.DB) {
 
 	super, dirs := dbTree(t, db)
 
-	tg := collectTreeGroups(dirs, super, nil)
+	tg := collectTreeGroups(dirs, super, everyPathExists, nil)
 	if len(tg) != 1 {
 		t.Fatalf("len(tree groups) = %d, want 1", len(tg))
 	}
@@ -1377,6 +1384,47 @@ func TestSyncScanMtimeBump(t *testing.T) {
 	}
 }
 
+// TestSyncScanMtimeMovedBack replaces a file with other content of the
+// same size and an older mtime, as cp -p or a restore from backup does.
+// The next scan must hash the file again.
+func TestSyncScanMtimeMovedBack(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	db := openTestDB(t)
+	a := writeFile(t, dir, "a.bin", pattern(1, 500))
+
+	recorded := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+
+	err := os.Chtimes(a, recorded, recorded)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	syncTree(t, db, dir)
+
+	replaced := pattern(3, 500)
+	writeFile(t, dir, "a.bin", replaced)
+
+	older := recorded.Add(-time.Hour)
+
+	err = os.Chtimes(a, older, older)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	st := syncTree(t, db, dir)
+	if st != (scanStats{walked: 1, updated: 1}) {
+		t.Fatalf("rescan stats = %+v, want 1 updated", st)
+	}
+
+	assertWholeFileHashed(t, db, a, replaced)
+
+	if r := recordByPath(t, dbRecords(t, db), a); !r.mtime.Equal(older) {
+		t.Fatalf("mtime = %v, want %v", r.mtime, older)
+	}
+}
+
 // assertWholeFileHashed fails unless the record for path holds the
 // whole-file hash of data as its head, tail, and content.
 func assertWholeFileHashed(t *testing.T, db *sql.DB, path string,
@@ -1448,20 +1496,26 @@ func TestScanStoresDiskSize(t *testing.T) {
 
 	unix.Sync() // see diskSizeOf
 
-	want := diskSizeOf(t, p)
+	size := diskSizeOf(t, p)
-	if want == 0 || want >= length {
+	if size == 0 || size >= length {
 		t.Fatalf("the file takes %d bytes on disk, want more than none "+
-			"and fewer than its length %d", want, length)
+			"and fewer than its length %d", size, length)
 	}
 
 	for _, operand := range []string{filepath.Dir(p), p} {
 		db := openTestDB(t)
+
+		// ZFS may count the blocks during the scan; see diskSizeOf.
+		before := diskSizeOf(t, p)
+
 		syncTree(t, db, operand)
 
+		after := diskSizeOf(t, p)
+
 		got := recordByPath(t, dbRecords(t, db), p).diskSize
-		if got != want {
+		if got < before || got > after {
-			t.Errorf("scan of %s: disk_size = %d, want %d",
+			t.Errorf("scan of %s: disk_size = %d, want %d to %d",
-				operand, got, want)
+				operand, got, before, after)
 		}
 	}
 }
@@ -1479,7 +1533,7 @@ func TestSyncScanRewriteUpdatesDiskSize(t *testing.T) {
 
 	syncTree(t, db, dir)
 
-	before := recordByPath(t, dbRecords(t, db), p).diskSize
+	recorded := recordByPath(t, dbRecords(t, db), p).diskSize
 
 	pokeAt(t, p, 512*1024, pattern(7, 64*1024))
 
@@ -1492,22 +1546,53 @@ func TestSyncScanRewriteUpdatesDiskSize(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	unix.Sync() // see diskSizeOf
+	// Wait for ZFS to count the write; see diskSizeOf.
-
+	deadline := time.Now().Add(30 * time.Second)
-	want := diskSizeOf(t, p)
+	for diskSizeOf(t, p) == recorded {
-	if want == before {
+		if time.Now().After(deadline) {
-		t.Fatalf("the file still takes %d bytes on disk after the write "+
+			t.Fatalf("the file still takes %d bytes on disk 30 seconds "+
-			"into its hole", want)
+				"after the write into its hole", recorded)
 		}
 
+		unix.Sync()
+	}
+
+	// ZFS may still be counting, so the size may change during a rescan.
+	// Only a rescan with the same size on both sides of it can be compared
+	// exactly. A rescan that keeps the old size still fails, because the
+	// wait made before differ from it.
+	for {
+		before := diskSizeOf(t, p)
+
 		st := syncTree(t, db, dir)
+
+		after := diskSizeOf(t, p)
+
 		if st != (scanStats{walked: 1, updated: 1}) {
 			t.Fatalf("rescan stats = %+v, want 1 updated", st)
 		}
 
+		if before == after {
 			got := recordByPath(t, dbRecords(t, db), p).diskSize
-	if got != want {
+			if got != after {
-		t.Errorf("disk_size = %d, want %d", got, want)
+				t.Errorf("disk_size = %d, want %d", got, after)
+			}
+
+			return
+		}
+
+		if time.Now().After(deadline) {
+			t.Fatalf("the size on disk was still changing 30 seconds after "+
+				"the write into its hole: %d, then %d", before, after)
+		}
+
+		// A later mtime makes the next rescan read the file again.
+		later = later.Add(time.Second)
+
+		err = os.Chtimes(p, later, later)
+		if err != nil {
+			t.Fatal(err)
+		}
 	}
 }
 
@@ -1533,15 +1618,21 @@ func TestSyncScanUnchangedUpdatesDiskSize(t *testing.T) {
 
 	unix.Sync() // see diskSizeOf
 
+	// ZFS may count the blocks during the rescan; see diskSizeOf.
+	before := diskSizeOf(t, p)
+
 	// A file read again would count as updated.
 	st := syncTree(t, db, dir)
+
+	after := diskSizeOf(t, p)
+
 	if st != (scanStats{walked: 1, unchanged: 1}) {
 		t.Fatalf("rescan stats = %+v, want 1 unchanged", st)
 	}
 
 	got := recordByPath(t, dbRecords(t, db), p).diskSize
-	if want := diskSizeOf(t, p); got != want {
+	if got < before || got > after {
-		t.Errorf("disk_size = %d, want %d", got, want)
+		t.Errorf("disk_size = %d, want %d to %d", got, before, after)
 	}
 }
 
@@ -2198,7 +2289,7 @@ func TestPruneRoots(t *testing.T) {
 
 	// The walk of /p does not enter /p/q when it is a default exclude, so
 	// /p/q is kept, without -x too; /p/q/r lies under it and is dropped.
-	got = pruneRoots(nested, false, walkRules{excludes: []string{pqRoot}},
+	got = pruneRoots(nested, false, walkRules{defaultExcludes: []string{pqRoot}},
 		device)
 	if !slices.Equal(got, want) {
 		t.Fatalf("pruneRoots excluding %s = %q, want %q", pqRoot, got, want)
@@ -2216,23 +2307,86 @@ func TestPruneRoots(t *testing.T) {
 	}
 }
 
-func TestReportsNeverTouchFilesystem(t *testing.T) {
+// TestRunReportsLeaveOutRemovedCopies scans the smoke tree, then
-	t.Parallel()
+// removes copies the reports name. report leaves out each removed file,
+// and a group left with one path. trees checks only the root of each
+// duplicate tree: with a file under t2 removed it still names t2, and
+// with t2 removed it names nothing.
+func TestRunReportsLeaveOutRemovedCopies(t *testing.T) {
+	t.Setenv(databaseEnv, testDBPath(t))
 
 	dir := buildSmokeTree(t)
-	db := openTestDB(t)
+	p := func(name string) string { return filepath.Join(dir, name) }
 
-	syncTree(t, db, dir)
+	scanOK(t, dir)
 
-	// Remove the scanned tree entirely; the analysis must be
+	for _, name := range []string{"b/copy2.bin", "tiny2", "t2/sub/f2"} {
-	// unaffected because it reads the database alone.
+		err := os.Remove(p(name))
-	err := os.RemoveAll(dir)
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	want := "first\tdupe\tsize\n" +
+		p("t1/f1") + "\t" + p("t2/f1") + "\t3000\n" +
+		p("t1/f1") + "\t" + p("t3/f1") + "\t3000\n" +
+		p("a/one.bin") + "\t" + p("b/copy.bin") + "\t2000\n" +
+		p("t1/sub/f2") + "\t" + p("t3/sub/f2renamed") + "\t100\n"
+	if got := runStdout(t, cmdReport); got != want {
+		t.Errorf("report stdout = %q, want %q", got, want)
+	}
+
+	// disk_size, the last column, depends on the filesystem.
+	header := "first\tdupe\tfiles\tsize\tdisk_size\n"
+
+	want = header + p("t1") + "\t" + p("t2") + "\t2\t3100\t"
+	if got := runStdout(t, cmdTrees); !strings.HasPrefix(got, want) ||
+		strings.Count(got, "\n") != 2 {
+		t.Errorf("trees stdout = %q, want one row starting %q", got, want)
+	}
+
+	err := os.RemoveAll(p("t2"))
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	assertSmokeDupeGroups(t, dir, db)
+	if got := runStdout(t, cmdTrees); got != header {
-	assertSmokeTreeGroups(t, dir, db)
+		t.Errorf("trees stdout after removing t2 = %q, want %q", got, header)
+	}
+}
+
+func TestPathExists(t *testing.T) {
+	t.Parallel()
+
+	dir := t.TempDir()
+	file := writeFile(t, dir, "file", nil)
+	locked := writeFile(t, dir, "locked/file", nil)
+
+	// Without search permission on its directory, a file cannot be
+	// looked at by anyone but root.
+	err := os.Chmod(filepath.Dir(locked), 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Runs before t.TempDir's own cleanup, which must delete the files.
+	t.Cleanup(func() {
+		//nolint:gosec // removing the directory needs its search bit back
+		_ = os.Chmod(filepath.Dir(locked), 0o700)
+	})
+
+	cases := map[string]bool{
+		dir:                                 true,
+		file:                                true,
+		locked:                              true,
+		filepath.Join(dir, "gone"):          false,
+		filepath.Join(file, "under-a-file"): false,
+	}
+	for path, want := range cases {
+		if got := pathExists(path); got != want {
+			t.Errorf("pathExists(%q) = %v, want %v", path, got, want)
+		}
+	}
 }
 
 func TestUnderRoot(t *testing.T) {
@@ -2261,33 +2415,24 @@ func TestUnderRoot(t *testing.T) {
 }
 
 // TestRunScanPassesOverDatabase scans a tree that holds the scan's own
-// database, reached directly and through a symlinked directory. The
+// database, and in one case also a hard link to the database file. The
-// walk passes over the database file, its -wal and -shm files and its
+// walk knows the database's files by device ID and inode number, so it
-// .lock file with no line and no count, and the record an earlier scan
+// passes over the database file, its -wal and -shm files, its .lock
-// left for the .lock file is deleted. A rescan of the unchanged tree
+// file and the hard link with no line and no count. The record an
-// updates nothing, and report lists none of the database's files.
+// earlier scan left for the .lock file is deleted. A rescan of the
+// unchanged tree updates nothing, and report lists none of the
+// database's files.
 func TestRunScanPassesOverDatabase(t *testing.T) {
-	for _, viaSymlink := range []bool{false, true} {
+	for _, viaHardLink := range []bool{false, true} {
-		t.Run(fmt.Sprintf("via symlink %v", viaSymlink), func(t *testing.T) {
+		t.Run(fmt.Sprintf("via hard link %v", viaHardLink), func(t *testing.T) {
-			base := t.TempDir()
+			tree := realTempDir(t)
-			tree := filepath.Join(base, "real", "tree")
+			hardLink := filepath.Join(tree, "hardlink")
-			operand := tree
 
 			writeFile(t, tree, "a", pattern(1, 100))
 			writeFile(t, tree, "b", pattern(1, 100))
 			writeFile(t, tree, "empty1", nil)
 			writeFile(t, tree, "empty2", nil)
 
-			if viaSymlink {
-				err := os.Symlink(filepath.Join(base, "real"),
-					filepath.Join(base, "link"))
-				if err != nil {
-					t.Fatal(err)
-				}
-
-				operand = filepath.Join(base, "link", "tree")
-			}
-
 			dbPath := filepath.Join(tree, "db.sqlite")
 			t.Setenv(databaseEnv, dbPath)
 
@@ -2297,7 +2442,7 @@ func TestRunScanPassesOverDatabase(t *testing.T) {
 			}
 
 			err = applyChanges(t.Context(), db, []scanRec{
-				{path: filepath.Join(operand, "db.sqlite.lock")},
+				{path: filepath.Join(tree, "db.sqlite.lock")},
 			}, nil, nil)
 			if err != nil {
 				t.Fatal(err)
@@ -2305,17 +2450,30 @@ func TestRunScanPassesOverDatabase(t *testing.T) {
 
 			closeScanDatabase(t.Context(), db, dbPath)
 
-			assertScanSummary(t,
+			if viaHardLink {
+				err = os.Link(dbPath, hardLink)
+				if err != nil {
+					t.Fatal(err)
+				}
+			}
+
+			stderr := assertScanSummary(t,
 				"4 files seen (4 added, 0 updated, 1 removed, 0 unchanged)",
-				operand)
+				tree)
-			assertScanSummary(t,
+			stderr += assertScanSummary(t,
 				"4 files seen (0 added, 0 updated, 0 removed, 4 unchanged)",
-				operand)
+				tree)
+
+			if strings.Contains(stderr, hardLink) {
+				t.Errorf("scan stderr = %q, want no line naming %s",
+					stderr, hardLink)
+			}
 
 			report := runStdout(t, cmdReport)
 			if strings.Contains(report, "db.sqlite") ||
-				!strings.Contains(report, filepath.Join(operand, "empty2")) {
+				strings.Contains(report, hardLink) ||
-				t.Errorf("report = %q, want the empty files and none of "+
+				!strings.Contains(report, filepath.Join(tree, "b")) {
+				t.Errorf("report = %q, want a and b and none of "+
 					"the database's files", report)
 			}
 		})
@@ -2339,8 +2497,9 @@ func TestRunScanPassesOverDatabaseOperands(t *testing.T) {
 
 // assertScanSummary scans operands and fails the test unless the
 // summary line gives counts with nothing skipped, and the line naming
-// the database is the only one that names one of its files.
+// the database is the only one that names one of its files. It returns
-func assertScanSummary(t *testing.T, counts string, operands ...string) {
+// everything the scan printed to stderr.
+func assertScanSummary(t *testing.T, counts string, operands ...string) string {
 	t.Helper()
 
 	stderr := scanOK(t, operands...)
@@ -2354,4 +2513,6 @@ func assertScanSummary(t *testing.T, counts string, operands ...string) {
 		t.Errorf("scan %q: stderr = %q, want only the line naming the "+
 			"database to name one of its files", operands, stderr)
 	}
+
+	return stderr
 }

internal/sfdupes/trees.go

@@ -1,4 +1,4 @@
-package main
+package sfdupes
 
 import (
 	"context"
@@ -38,10 +38,11 @@ func treesPhase(name string) string {
 // runTrees implements the trees subcommand: it reads every record from
 // the database in path order, reconstructs the directory hierarchy from
 // the record paths, computes a Merkle-style digest per directory, and
-// prints maximal duplicate-tree groups as TSV on stdout. It never
+// prints maximal duplicate-tree groups as TSV on stdout. Of the scanned
-// touches the scanned filesystem; its only I/O is the database, stdout,
+// filesystem it only checks, with pathExists, that the root of each
-// and stderr. It names the database on stderr before it opens it. Any
+// duplicate tree is still there, never the files below it. It names the
-// database problem, including a missing database, is fatal.
+// database on stderr before it opens it. Any database problem,
+// including a missing database, is fatal.
 //
 // The work runs in four phases, each under a progress display of its
 // own on stderr: open (open the database), read (read the number of
@@ -86,13 +87,13 @@ func runTrees(ctx context.Context, stdout io.Writer) error {
 		return fmt.Errorf("database %s: %w", dbPath, err)
 	}
 
-	// Grouping works in memory, so it waits on nothing but the
+	// Grouping works in memory, then checks the root of each duplicate
-	// processor.
+	// tree on disk, so it waits on the processor or the disk.
 	prog = newProgress(progressParams{
 		label: treesPhase("group"), unit: "directories",
-		total: int64(len(allDirs)), stalledOn: "cpu",
+		total: int64(len(allDirs)), stalledOn: "cpu/disk",
 	})
-	dupes := collectTreeGroups(allDirs, super, prog)
+	dupes := collectTreeGroups(allDirs, super, pathExists, prog)
 
 	prog.finish()
 
@@ -300,10 +301,14 @@ func fileEntry(name string, r scanRec) string {
 
 // collectTreeGroups groups directories by digest, counting each one on
 // prog as it is grouped, and returns every maximal group with two or
-// more members, each group's members sorted by path, groups ordered by
+// more members for which exists returns true, each group's members
-// tree size descending then by first path ascending.
+// sorted by path, groups ordered by tree size descending then by first
+// path ascending. A group whose trees hold 0 bytes is left out, since
+// removing one frees nothing. The groups it implies hold 0 bytes too,
+// so leaving it out hides nothing. exists is called only on the members
+// of maximal groups.
 func collectTreeGroups(allDirs []*treeNode, super *treeNode,
-	prog *progress,
+	exists func(path string) bool, prog *progress,
 ) [][]*treeNode {
 	groups := make(map[[sha256.Size]byte][]*treeNode)
 	for _, d := range allDirs {
@@ -315,7 +320,19 @@ func collectTreeGroups(allDirs []*treeNode, super *treeNode,
 	var dupes [][]*treeNode
 
 	for _, g := range groups {
-		if len(g) < minGroupSize || suppressed(g, super) {
+		// The members share a digest, so they share a total size.
+		if len(g) < minGroupSize || g[0].totalSize == 0 ||
+			suppressed(g, super) {
+			continue
+		}
+
+		// A member that is gone leaves the group. A suppressed group is
+		// still implied: its members still on disk have parents still
+		// on disk, and those parents' group is checked the same way.
+		g = slices.DeleteFunc(g, func(n *treeNode) bool {
+			return !exists(n.path)
+		})
+		if len(g) < minGroupSize {
 			continue
 		}
 

internal/sfdupes/trees_test.go

@@ -1,10 +1,12 @@
-package main
+package sfdupes //nolint:testpackage // checks the unexported treeNode results
 
 import (
 	"bytes"
+	"crypto/sha256"
 	"database/sql"
 	"os"
 	"slices"
+	"strconv"
 	"strings"
 	"testing"
 	"time"
@@ -178,7 +180,7 @@ func TestTreeNamesSortingBeforeSlash(t *testing.T) {
 		t.Fatalf("directory paths = %q, want %q", got, want)
 	}
 
-	groups := collectTreeGroups(dirs, super, nil)
+	groups := collectTreeGroups(dirs, super, everyPathExists, nil)
 
 	gotGroups := groupPaths(groups)
 	wantGroups := [][]string{{"/a", "/c"}}
@@ -194,7 +196,8 @@ func TestTreeNamesSortingBeforeSlash(t *testing.T) {
 }
 
 func TestRunTreesEscapesPaths(t *testing.T) {
-	t.Setenv(databaseEnv, seedDatabase(t, awkwardPairRecs()))
+	dir, recs := recsOnDisk(t, awkwardPairRecs())
+	t.Setenv(databaseEnv, seedDatabase(t, recs))
 
 	var stdout, stderr bytes.Buffer
 
@@ -205,7 +208,7 @@ func TestRunTreesEscapesPaths(t *testing.T) {
 	}
 
 	want := "first\tdupe\tfiles\tsize\tdisk_size\n" +
-		`/d/\tone\ntwo\rthree\\four` + "\t/d/A\t1\t5\t4096\n"
+		dir + `/d/\tone\ntwo\rthree\\four` + "\t" + dir + "/d/A\t1\t5\t4096\n"
 	if got := stdout.String(); got != want {
 		t.Errorf("stdout = %q, want %q", got, want)
 	}
@@ -223,9 +226,11 @@ func TestRunTreesDiskSize(t *testing.T) {
 		{size: 20, diskSize: 1024, head: g, tail: g, content: g, path: "/b/sub/g"},
 	}
 
+	dir, recs := recsOnDisk(t, recs)
 	t.Setenv(databaseEnv, seedDatabase(t, recs))
 
-	want := "first\tdupe\tfiles\tsize\tdisk_size\n/a\t/b\t2\t30\t1536\n"
+	want := "first\tdupe\tfiles\tsize\tdisk_size\n" +
+		dir + "/a\t" + dir + "/b\t2\t30\t1536\n"
 	if got := runStdout(t, cmdTrees); got != want {
 		t.Errorf("stdout = %q, want %q", got, want)
 	}
@@ -236,10 +241,11 @@ func TestRunTreesDiskSize(t *testing.T) {
 // there. After the line naming the database, each phase prints a line
 // naming it with its place among the four as it starts and another as
 // it ends. read counts the 6 records before it reads them, group counts
-// the 8 directories, and write the 1 group. The summary comes last, and
+// the directories, and write the 1 group. The summary comes last, and
 // stdout holds the report alone.
 func TestRunTreesProgressNamesEachPhase(t *testing.T) {
-	path := seedDatabase(t, smokeTreeRecs())
+	dir, recs := recsOnDisk(t, smokeTreeRecs())
+	path := seedDatabase(t, recs)
 	t.Setenv(databaseEnv, path)
 
 	stderr := captureStderr(t)
@@ -253,11 +259,15 @@ func TestRunTreesProgressNamesEachPhase(t *testing.T) {
 	}
 
 	report := "first\tdupe\tfiles\tsize\tdisk_size\n" +
-		"/d/t1\t/d/t2\t2\t3100\t8192\n"
+		dir + "/d/t1\t" + dir + "/d/t2\t2\t3100\t8192\n"
 	if got := stdout.String(); got != report {
 		t.Errorf("stdout = %q, want %q", got, report)
 	}
 
+	// The smoke-test records name 8 directories, / included. Moved under
+	// dir, they name one more for each slash in dir.
+	dirs := strconv.Itoa(8 + strings.Count(dir, "/"))
+
 	// How each line starts; the rate at the end of a phase varies.
 	want := []string{
 		"trees: database " + path + " (from " + databaseEnv + ")",
@@ -265,8 +275,9 @@ func TestRunTreesProgressNamesEachPhase(t *testing.T) {
 		treesOpenLine,
 		"[2/4] read: counting the records, elapsed 0s, remain unknown",
 		"[2/4] read: [6/6] 100%, ",
-		"[3/4] group: [0/8] 0%, 0 directories/s, elapsed 0s, remain unknown",
+		"[3/4] group: [0/" + dirs + "] 0%, 0 directories/s, elapsed 0s, " +
-		"[3/4] group: [8/8] 100%, ",
+			"remain unknown",
+		"[3/4] group: [" + dirs + "/" + dirs + "] 100%, ",
 		"[4/4] write: [0/1] 0%, 0 groups/s, elapsed 0s, remain unknown",
 		"[4/4] write: [1/1] 100%, ",
 		"trees: 6 records read, 1 duplicate tree groups, 1 dupe trees, ",
@@ -295,7 +306,7 @@ func TestRunTreesProgressNamesEachPhase(t *testing.T) {
 func TestWriteTreeGroupsOnTerminal(t *testing.T) {
 	stderr := captureStderr(t)
 	super, dirs := treeOf(t, smokeTreeRecs())
-	dupes := collectTreeGroups(dirs, super, nil)
+	dupes := collectTreeGroups(dirs, super, everyPathExists, nil)
 
 	// No timed draws: the display is drawn as it starts and as it ends.
 	p := startProgress(progressParams{
@@ -376,7 +387,7 @@ func TestCollectTreeGroupsMaximal(t *testing.T) {
 
 	super, dirs := treeOf(t, smokeTreeRecs())
 
-	groups := collectTreeGroups(dirs, super, nil)
+	groups := collectTreeGroups(dirs, super, everyPathExists, nil)
 
 	want := [][]string{{"/d/t1", "/d/t2"}}
 	if got := groupPaths(groups); !slices.EqualFunc(got, want,
@@ -400,14 +411,16 @@ func TestCollectTreeGroupsDeterministic(t *testing.T) {
 
 	super, dirs := treeOf(t, recs)
 
-	forward := groupPaths(collectTreeGroups(dirs, super, nil))
+	forward := groupPaths(collectTreeGroups(dirs, super, everyPathExists,
+		nil))
 
 	reversed := slices.Clone(recs)
 	slices.Reverse(reversed)
 
 	superR, dirsR := treeOf(t, reversed)
 
-	backward := groupPaths(collectTreeGroups(dirsR, superR, nil))
+	backward := groupPaths(collectTreeGroups(dirsR, superR, everyPathExists,
+		nil))
 	if !slices.EqualFunc(forward, backward, slices.Equal) {
 		t.Fatalf("output depends on record order: %v vs %v",
 			forward, backward)
@@ -427,7 +440,36 @@ func TestCollectTreeGroupsSiblings(t *testing.T) {
 
 	super, dirs := treeOf(t, recs)
 
-	got := groupPaths(collectTreeGroups(dirs, super, nil))
+	got := groupPaths(collectTreeGroups(dirs, super, everyPathExists, nil))
+
+	want := [][]string{{"/p/x1", "/p/x2"}}
+	if !slices.EqualFunc(got, want, slices.Equal) {
+		t.Fatalf("groups = %v, want %v", got, want)
+	}
+}
+
+// TestCollectTreeGroupsLeavesOutEmptyTrees checks that directories
+// holding only empty files form no group, and that an empty file still
+// counts toward the digest of a tree that holds other files.
+func TestCollectTreeGroupsLeavesOutEmptyTrees(t *testing.T) {
+	t.Parallel()
+
+	h, tl, c, e := testHash("h"), testHash("t"), testHash("c"),
+		sha256.Sum256(nil)
+	recs := []scanRec{
+		{size: 0, head: e, tail: e, content: e, path: "/e/a/.gitkeep"},
+		{size: 0, head: e, tail: e, content: e, path: "/e/b/.gitkeep"},
+		{size: 10, head: h, tail: tl, content: c, path: "/p/x1/f"},
+		{size: 0, head: e, tail: e, content: e, path: "/p/x1/.gitkeep"},
+		{size: 10, head: h, tail: tl, content: c, path: "/p/x2/f"},
+		{size: 0, head: e, tail: e, content: e, path: "/p/x2/.gitkeep"},
+		// Without the empty file, /p/x3 is no copy of /p/x1.
+		{size: 10, head: h, tail: tl, content: c, path: "/p/x3/f"},
+	}
+
+	super, dirs := treeOf(t, recs)
+
+	got := groupPaths(collectTreeGroups(dirs, super, everyPathExists, nil))
 
 	want := [][]string{{"/p/x1", "/p/x2"}}
 	if !slices.EqualFunc(got, want, slices.Equal) {
@@ -450,10 +492,98 @@ func TestCollectTreeGroupsDifferingParents(t *testing.T) {
 
 	super, dirs := treeOf(t, recs)
 
-	got := groupPaths(collectTreeGroups(dirs, super, nil))
+	got := groupPaths(collectTreeGroups(dirs, super, everyPathExists, nil))
 
 	want := [][]string{{"/p/a/x", "/q/b/x"}}
 	if !slices.EqualFunc(got, want, slices.Equal) {
 		t.Fatalf("groups = %v, want %v", got, want)
 	}
 }
+
+// TestCollectTreeGroupsChecksOnlyReportedCopies checks that trees asks
+// whether a directory is on disk only for the members of the groups it
+// returns. It never asks about /a/proj/sub, which lies inside a copy
+// and whose group the group of /a/proj implies.
+func TestCollectTreeGroupsChecksOnlyReportedCopies(t *testing.T) {
+	t.Parallel()
+
+	f, g, u := testHash("f"), testHash("g"), testHash("u")
+	recs := []scanRec{
+		{size: 10, head: f, tail: f, content: f, path: "/a/proj/f"},
+		{size: 20, head: g, tail: g, content: g, path: "/a/proj/sub/g"},
+		{size: 10, head: f, tail: f, content: f, path: "/b/proj/f"},
+		{size: 20, head: g, tail: g, content: g, path: "/b/proj/sub/g"},
+		// /a lacks this file, so /a and /b are no copies of each other.
+		{size: 30, head: u, tail: u, content: u, path: "/b/lonely"},
+	}
+
+	super, dirs := treeOf(t, recs)
+
+	var checked []string
+
+	exists := func(path string) bool {
+		checked = append(checked, path)
+
+		return true
+	}
+
+	got := groupPaths(collectTreeGroups(dirs, super, exists, nil))
+
+	want := [][]string{{"/a/proj", "/b/proj"}}
+	if !slices.EqualFunc(got, want, slices.Equal) {
+		t.Fatalf("groups = %v, want %v", got, want)
+	}
+
+	slices.Sort(checked)
+
+	if !slices.Equal(checked, want[0]) {
+		t.Errorf("directories checked on disk = %q, want %q",
+			checked, want[0])
+	}
+}
+
+// TestCollectTreeGroupsDropsGroupLeftWithOneCopy checks that a group of
+// two copies, one of them gone, is not returned. If kept, it would
+// write no row, but the summary line would count it.
+func TestCollectTreeGroupsDropsGroupLeftWithOneCopy(t *testing.T) {
+	t.Parallel()
+
+	f := testHash("f")
+	recs := []scanRec{
+		{size: 10, head: f, tail: f, content: f, path: "/c/kept/f"},
+		{size: 10, head: f, tail: f, content: f, path: "/c/gone/f"},
+	}
+
+	super, dirs := treeOf(t, recs)
+
+	exists := func(path string) bool { return path != "/c/gone" }
+
+	got := groupPaths(collectTreeGroups(dirs, super, exists, nil))
+	if len(got) != 0 {
+		t.Errorf("groups = %v, want none", got)
+	}
+}
+
+// TestCollectTreeGroupsKeepsGroupLeftWithTwoCopies checks that a group
+// of three copies, one of them gone, is returned with the other two.
+func TestCollectTreeGroupsKeepsGroupLeftWithTwoCopies(t *testing.T) {
+	t.Parallel()
+
+	f := testHash("f")
+	recs := []scanRec{
+		{size: 10, head: f, tail: f, content: f, path: "/d/kept1/f"},
+		{size: 10, head: f, tail: f, content: f, path: "/d/gone/f"},
+		{size: 10, head: f, tail: f, content: f, path: "/d/kept2/f"},
+	}
+
+	super, dirs := treeOf(t, recs)
+
+	exists := func(path string) bool { return path != "/d/gone" }
+
+	got := groupPaths(collectTreeGroups(dirs, super, exists, nil))
+
+	want := [][]string{{"/d/kept1", "/d/kept2"}}
+	if !slices.EqualFunc(got, want, slices.Equal) {
+		t.Errorf("groups = %v, want %v", got, want)
+	}
+}