fetch downloads each file to a temp name beside it and first removes
whatever is there. A manifest listing both a.txt and .a.txt.tmp had
fetch delete the second while fetching the first, then exit 0 with a
tree check rejects.
The refusal of a manifest that lists the saved manifest's own name or
temp name now covers this too: a listed file, or a directory a listed
file is in, may not sit at any name fetch writes besides the listed
files themselves. Temp names come from tempPathFor, names are compared
ignoring case as before, and the refusal still happens before the
destination is created or any file requested.
Model: opus-5-5
The Makefile is now one shim per script/ entrypoint, in the order of
the canonical model Makefile, plus build, generate and fuzz. The new
script/build writes bin/mfer with the urfave_cli_no_docs tag and the
git describe revision that script/docker stamps. script/generate now
adds the Go bin directory to PATH itself, and the Docker lint stage
calls script/gofumpt --check directly. Removed: the tarball-caching
rules and their ignore entries, godoc, ci, fixme, clean, run, default,
fmt-check-go, fmt-check-md, and bin/gitrev.sh, which only the
Makefile called.
Model: opus-5-5
fetch takes --dest (default .) and writes every file there through the
existing symlink and hard-link guards, which now work relative to that
directory. A file already there with the listed size and hash is
skipped; a leftover temp file is still replaced. Once every file
verifies, the manifest is saved as index.mf through the same temp file
and rename, so check runs on the result; a manifest that lists index.mf
or its temp name at the top of the tree is refused, since saving would
replace that file. --require-signature is shared with check and
enforced through verifyRequiredSigner. Both refusals come before any
file is downloaded or anything is written.
Model: opus-5-5
The Go format check ran plain gofmt, which accepts code that gofumpt,
the formatter script/fmt runs, rewrites; and the two covered different
files. Both now go through script/gofumpt, which runs one pinned
gofumpt version over every Go file in --write or --check mode, as
script/prettier does for Markdown. It replaces script/fmt-check-go;
make fmt-check-go and the Docker lint stage call it. go run builds the
pinned version, so nothing has to install gofumpt, and the check fails
when gofumpt cannot run instead of passing. Generated mfer/mf.pb.go
passes: gofumpt holds generated files to gofmt's rules. The check is
not redundant: .golangci.yml enables no golangci-lint formatters.
Model: opus-5-5
check now always looks under the base directory for files the manifest
does not list, hidden files and directories included, and prints one
warning per file. The result still depends only on the listed files;
--no-extra-files turns each unlisted file into a failure, and --quiet
hides the warnings but not the failures. A directory that cannot be
listed is reported the same way, and the search goes on past it.
The manifest itself is left out by file identity, as gen and freshen
do; a symlink under the base is compared by what it points to. A base
directory named through a symlink is resolved before the search.
Model: opus-5-5
FORMAT.md moves to docs/ and every reference follows; its two relative
links to mfer/mf.proto now point up one directory, its text is otherwise
unchanged. contrib/usage.sh moves to bin/, not docs/: it is a script you
run (it builds mfer, then generates and checks a manifest of the repo),
not reading material. Both scripts now use #!/usr/bin/env bash,
set -euo pipefail and a main function. bin/gitrev.sh still exits non-zero
outside a git checkout, so the Makefile still falls back to unknown.
.gitignore now ignores only the built bin/mfer rather than all of bin/,
which holds tracked scripts.
Model: opus-5-5
The freshen tests built a fixture and never ran freshen. They now run
gen and freshen through the command entry point on a temp dir: an
unchanged tree leaves the manifest's entries as they were, and a
modified file (also one edited without changing its size), a new or a
deleted file gives a manifest that lists exactly the tree, with sizes,
hashes and mtimes from disk, on which check passes.
New fetch tests run the command against an httptest server: a nested
tree lands complete, a hash mismatch exits non-zero and leaves no file,
and a partly filled destination gets every listed file downloaded again
and replaced, while a file the manifest does not list is left alone.
Model: opus-5-5
internal/log now logs through log/slog. After Init, records go to a small
slog.Handler that prints the same lines as before; before Init they go to
slog.Default(). The helpers and their call sites are unchanged. With
NO_COLOR set on a terminal, log lines are no longer colored.
pterm is dropped: it only printed progress lines, which fmt now writes
directly, outside slog. apex/log, pterm and their indirect dependencies
leave go.mod; golang.org/x/term becomes direct.
simplelog is not imported: importing it replaces slog's default handler in
every program that imports package mfer. The logger stays process-global,
since injecting it would change package mfer's API.
Progress and log lines now share the logger's lock, so the check progress
test runs check ten times to catch a missing wait.
Model: opus-5-5
mf.proto named git.eeqj.de/sneak/mfer/mfer as its Go package, which
disagrees with the sneak.berlin/go/mfer module path go.mod declares.
go_package is now sneak.berlin/go/mfer/mfer, the import path of the
mfer/ package. mf.pb.go is regenerated with make generate, which
changes only the go_package bytes in its embedded descriptor, and
mf.proto.sha256 records the new mf.proto hash.
Model: opus-5-5
fetch now makes every request through an http.Client with a time
limit, ten minutes by default and set with --timeout, which must be
greater than zero. A connection error, a timeout, or a 5xx or 429
response is retried, up to five tries in all, after a random wait
whose limit doubles from one second, or after the wait the server's
Retry-After asks for, up to one minute. Each try of a file starts a
new temp file, so a retry never keeps a partial file; the size and
hash checks are unchanged. Manifest and file URLs are built with
URL.JoinPath, so trailing slashes, query strings and names that need
escaping all work.
Model: opus-5-5
With --progress, runCheck started the progress goroutine but waited
only for the results goroutine, so check could log its summary, or
return, before the last progress line was written and cleared. The
progress goroutine now signals a sync.WaitGroup when it finishes, and
runCheck waits on it as soon as Check returns, as generate does.
The new test sends stdout and stderr to one buffer and slows down
stdout writes, so without the wait the progress output lands after
the summary.
Model: opus-5-5
AddFileWithHash took any non-empty bytes as a hash, so the builder could
write a manifest that mfer refuses to load. It now decodes the hash with
go-multihash and also requires a digest of at least 32 bytes, the SHA-256
length the reader's decoding-cost limit assumes: a valid but shorter
multihash, such as an empty identity hash or SHA-1, still makes a
manifest of one-character paths too costly to load. When mfer freshen
meets such a hash in an existing manifest, its error names the manifest
entry and says to regenerate the manifest with mfer generate. Test
fixtures whose stand-in hashes were not valid multihashes now use a
SHA-256 multihash.
Model: opus-5-5
mfer gen now writes index.mf instead of .index.mf, the name fetch
requests and the README calls the standard filename. Given a
directory, check, freshen, list and export look only for index.mf;
.index.mf is no longer recognized. Because index.mf is not hidden, gen
and freshen leave out of their own listing the manifest they write and
a temp file an interrupted run left beside it, matched by file
identity (os.SameFile) however the path is spelled. The scanner takes
these as a plain ExcludePaths list; manifestTempPath alone names the
temp file, for both writes, both skips and the tests.
Model: opus-5-5
The test and format scripts regenerated mfer/mf.pb.go whenever
mfer/mf.proto looked newer by mtime, which a fresh checkout often causes,
so make check could rewrite a committed file and needed protoc. Nothing
regenerates it any more except make generate (script/generate), which
refuses to run unless protoc 33.4 and protoc-gen-go v1.36.11, the
versions that wrote the committed file, are on PATH, and records the
hash of mf.proto in mfer/mf.proto.sha256. A Go test compares that hash
with mf.proto and fails, naming make generate, when they differ. The
mtime rule, the unused protoc-gen-go v1.28.1 install rule, make clean's
deletion of mf.pb.go and the Dockerfile's touch workarounds are removed.
Model: opus-5-5
NewChecker now takes *CheckerOptions (ManifestPath, BasePath, Fs), named
like ScannerOptions. A nil Fs still means the OS filesystem, as before and
as in ScannerOptions; nil options or an empty path return an error naming
the missing path.
Audit of the other exported constructors in mfer: NewManifestFromFile took
a filesystem and a path positionally; it now takes
*ManifestFromFileOptions (Path, Fs) with the same nil and empty rules.
NewBuilder and NewScanner take no arguments, NewScannerWithOptions already
takes options, and NewManifestFromReader takes one reader, which the style
guide exempts; these are unchanged.
Model: opus-5-5
Before decoding the manifest, the parser adds up what decoding sets
aside for each file entry, hash, timestamp and MIME type, however short
its encoding, and refuses the manifest once that passes 8 times the
decompressed size; manifests mfer writes come to at most about 7.15
times. Empty entries decoded to about 50 times their size: under 1 KB of
manifest allocated about 500 MB. Fields the decoder does not know are
dropped; kept, they took up to 5 times more. A test refuses entries
counted just over 8 times and loads them just under. The fuzz ceiling
rises from 16 to 20 times the input and decompressed data; seeds of
empty entries and of empty hashes fail it without the fix.
Model: opus-5-5
urfave/cli before v2.25.5 counted a flag given by its alias twice, so
one -v or --verbose already gave debug output. Bump it to v2.27.7,
which counts it once: one -v gives verbose output, two give debug.
The -v help text now says -v -v instead of -vv, which stays refused:
urfave/cli's option for combined short flags would let a flag that
takes a value read the next letter as its value.
-v and --verbose together stay refused: urfave/cli v2 refuses a flag
given under two of its names, and one flag with an alias keeps help and
parsing simple.
The bump changes some help output; generate and fetch now name their
arguments. Tests start each run at the default log level.
Model: opus-5-5
The final stage is scratch, which has no CA certificates, so fetch from
an HTTPS URL failed to verify any server. Copy the CA bundle from the
pinned builder image into the final stage.
Model: opus-5-5
The two status-send tests now call the send directly on a channel nobody
receives from, so a blocking send hangs the test into its timeout instead
of racing a 100 ms timer that a loaded host can miss.
The gpg test for a child holding gpg's output had the same problem: its
100 ms deadline could fire before the fake gpg wrote the PID of sleep,
and the cleanup then failed. The fake gpg now writes that PID to a named
pipe, and the test cancels only after reading it. It then waits up to
10 s for the call, so a call that waits for sleep fails the test and the
cleanup still kills sleep.
Model: opus-5-5
script/lint now builds only the lint stage of the main Dockerfile
(docker build --no-cache --target lint), whose build runs the linter, so
a successful build is a clean lint. It is uncached because a cached
build runs no linter, and a trap removes the image it tagged; the tag
carries the process ID so concurrent runs do not collide. The lint stage
calls golangci-lint directly, since make lint now needs Docker. Nothing
installs or runs golangci-lint on the host any more: bootstrap and the
Makefile drop the install, and script/fmt drops golangci-lint run --fix.
Model: opus-5-5
The final stage is scratch, which has no C library, but the builder
compiled mfer with cgo on (the golang image's default). The binary
imports net (mfer's own HTTP code does, and so does google/uuid), so
with cgo on it came out dynamically linked and the image could not
start. The image's go build now sets CGO_ENABLED=0; nothing in mfer
needs cgo. A new builder step runs ldd on the binary and fails the
build unless it reports a static executable.
Model: opus-5-5
The Description first line now names the project, purpose, category,
WTFPL license, and author. A new Getting Started section gives a
copy-pasteable build-from-source block and gen/check/fetch usage. Problem
Statement and Proposed Solution move under a new Rationale heading, the
prose kept. A new Design section documents the package layout: the mfer/
library with its committed protobuf code, the internal/cli commands,
internal/log and internal/bork, and the cmd/mfer entrypoint. Authors is
renamed Author with the canonical link.
Getting Started uses go build because the make target that builds the
binary runs protoc, which script/bootstrap does not install.
Model: opus-4-8 (implementation); opus-5-5 (rebase, rework)
urfave/cli's built-in version flag claimed -v, so "mfer -v --version"
failed to parse, and -v meant version at the root but verbose on
generate, check, freshen and fetch. Verbose is the more common meaning,
so the root now takes -v and -q too, and the version flag takes -V. A
-v or -q before generate, check, freshen, fetch or export applies to
that subcommand; list keeps its fixed quiet logging.
User-visible changes: -v no longer prints the version; use -V or
--version. "mfer version" prints "mfer version 0.1.0 (...)", the same
line as --version, instead of the bare "0.1.0 (...)".
Tests: runCLI now points the logger at io.Discard after each run, so
other tests' log lines cannot land in a finished run's output.
Model: opus-4-8 (implementation); opus-5-5 (rework)
FuzzNewManifestFromReader fails when the parser returns both or neither
of a manifest and an error, or allocates more than a fixed multiple of
its input and the decompressed data it may read, plus room for the
decoder's window buffers. make test runs the seed corpus; make fuzz
fuzzes for one minute.
Parser bug: MaxDecompressedSize did not bound decompression; the zstd
decoder decoded a payload under 128 KiB in full before the LimitReader
read any of it. It now decodes only what the LimitReader reads and
refuses windows over the 8 MiB mfer writes with. Seeds: a frame claiming
8 GiB, two frames together over the limit, empty frames with growing
windows.
Model: opus-5-5
Every gpg run now has a one-minute deadline (gpgTimeout) on top of its
caller's context and is killed when either ends. A timeout is reported
as "gpg timed out" under the failing operation instead of "signal:
killed". Only gpg itself is killed; WaitDelay (one second) stops the run
from waiting on a process gpg left behind that still holds its output,
such as a wrapper script that does not exec the real gpg.
Builder.Build and Checker.ExtractEmbeddedSigningKeyFP take a context, so
ToManifest's context now reaches signing and the contextcheck
suppression calling signing non-cancellable is gone. Manifest loading
takes no context, so its signature check is bounded by the timeout
alone.
Model: opus-5-5
TODO.md is deleted and the README TODO section now only points to the
issue tracker, where open work and open design questions live from now
on. AGENTS.md says so too, and says this overrides the shared policy's
README todo list for this repo. The README Open Questions section
becomes Original Design Questions, each noting where it now stands.
Every open item in both lists was already done or already owned by an
issue, apart from one, now #121,
and the chunk checksum question, now on
#81.
Model: opus-5-5
errmsg_test.go now calls the function that emits each user-visible CLI
error message and asserts on the full text it returns, instead of
rendering format strings copied from production, so rewording any pinned
message fails the suite. The unknown-command message is driven through
the root command's action.
The signer-mismatch case signs a manifest with a throwaway gpg key and
is skipped where gpg is absent, like the repo's other signing tests.
The freshen mtime-presence test gives the scanned file an epoch mtime,
so it fails if recordEntry reads an absent manifest mtime as the epoch.
Model: opus-4-8 (first implementation); opus-5-5 (completion, this message)
A bare `docker build .` served the Dockerfile's check steps from the
layer cache on an unchanged tree and exited 0 without running them.
script/cibuild now computes the version on its own line and runs the
same build command as script/docker and the shared policy, --no-cache
included, so every CI build runs the checks. README.md describes
script/cibuild accordingly.
Model: opus-5-5
downloadFile opened the temp file with os.Create, which opens and
empties a file already at that name. If that file was a hard link to a
file outside the destination directory, fetch overwrote the outside
file. It now removes whatever is at the temp name, which removes only
that name, and creates the temp file with O_EXCL, so the create fails if
anything is still or again there. A leftover temp file from an
interrupted run is still replaced. The new test puts a hard link at the
temp name and checks that fetch succeeds and the outside file is
unchanged. The command name is now the constant cmdFetch, like the other
command names, because lint requires it once a third test uses it.
Model: opus-5-5
sanitizePath checks manifest paths only as text, so a symlink already
inside the destination directory could send fetch's writes outside it.
checkNoSymlinks now looks at each existing part of a path with os.Lstat
and refuses the path if any part is a symlink, wherever it points.
fetch runs it immediately before each write: creating the parent
directories, creating the temp file, and renaming it into place. The new
test puts such a symlink at each of those three places, and once inside
a plain directory, and checks that the fetch fails and nothing outside
changes. The G304 comment now states what holds. A symlink swapped in
between a check and its write is not caught; os.Root closes that once
the Go version is raised.
Model: opus-5-5
.dockerignore now sends .git but not .git/config, which can hold a
credential and which git describe does not need. The build stage stamps
main.Gitrev from the VERSION build argument when one is given, otherwise
from git describe --tags --always, and fails if .git is present and no
version comes out. git there trusts /src whoever owns it, since a
context sent as a tar archive keeps its files' owners. script/docker is
replaced by the canonical copy, which passes VERSION; bin/gitrev.sh uses
--tags too, so every entrypoint stamps the same value for a clean
commit.
Model: opus-5-5
Untrusted .mf files were parsed with no path validation, so an entry like ../../etc/passwd reached filepath.Join against the checker's base path and mfer check could stat and read outside it. ValidatePath ran only when building a manifest. It now runs on every entry as the manifest loads, so every consumer is covered. The whole manifest is rejected on the first bad entry instead of dropping it, which could hide files from a check; the error wraps errInvalidManifestPath and names the path.
Disclosure: a path that is not valid UTF-8 is refused earlier, by the protobuf string decoder, so that error does not name the path.
Model: opus-4-8 (implementation); fable-5-1 (summary)
script/test now runs go test -timeout 30s -race -cover ./..., quiet on success and rerunning verbose on failure. -race exposed a data race on the process-global apex/log logger: Init reconfigured it on every CLI run while other goroutines logged through it. internal/log now mutates the global under the write lock and reads it under the read lock; WithError is dropped because its Entry logged outside that lock, and run() reports a failed command's error via log.Errorf instead (still shown under -q). The corruption fuzz test is scaled to 1500 files / 100 iterations to fit the budget under -race; it pins the same claims at reduced breadth. Independent review ran the Docker lint gate uncached.
Model: opus-4-8 (implementation, review)
model: claude-fable-5
Add the canonical .editorconfig from sneak/prompts verbatim. Broaden .gitignore to cover secrets (.env, .env.*, *.key, *.pem), OS files, editor files, and Go artifacts while keeping every repo-specific entry; no tracked file becomes ignored, and bin/gitrev.sh stays tracked despite the /bin/ rule. Remove the stale .drone.yml ignore entry and the DRONE_COMMIT_SHA branch from bin/gitrev.sh, left from the Gitea Actions migration; the GITREV and git-describe paths still work.
A reader may trip over .editorconfig saying four spaces for every file while Go files are tab-indented: gofmt governs Go, editorconfig does not, and the file is taken verbatim by policy.
Disclosure: the worker's cold docker build hit the 10s test timeout in the gpg signature test; this change has no Go code, and that timeout is tracked by issues 67 and 62.
Model: opus-4-8 (implementation); fable-5-1 (merge)
Adds .prettierrc and .prettierignore, a single script/prettier entrypoint shared by fmt and fmt-check so the two cannot drift, and prettier 3.9.6 pinned by yarn.lock integrity hash.
Markdown formatting is now gated in the authoritative Docker build via a new mdfmt stage, since the golangci-lint image has no node. REPO_POLICIES.md is ignored so local tooling cannot drift it from upstream.
Removes the || true that made the previous prettier invocation unable to fail.
Adopts golangci-lint v2.12.2 and the canonical .golangci.yml (default: all), and fixes all resulting findings across the tree.
Two intended behavior changes: absent MFFilePath.Mtime is handled explicitly in freshen, list and export rather than dereferenced (main panicked); gpg positional key IDs now follow an explicit -- end-of-options marker.
All twelve reworded user-visible error messages restored to byte-identical parity with main and pinned by tests.
## Summary
Performs a design and status review of the codebase and adds a comprehensive TODO section to `README.md` listing remaining work for a 1.0 release.
### What changed
- **README.md**: Added a `TODO: Remaining Work for 1.0` section covering:
- **7 design questions** requiring @sneak's input before implementation (manifest type export, Go module path, GPG vs pure-Go crypto, format framing, etc.) — each with an answer field for inline decisions
- **Implementation tasks** organized by category: repo infrastructure, format & correctness, library, CLI, testing & robustness, documentation, and release checklist
- Updated build status section (removed stale Drone CI badge, replaced with description of current Docker-based CI)
- **TODO.md**: Removed — items integrated into README TODO section
- **AGENTS.md**: Updated reference from `TODO.md` to README TODO section
### Design review findings
**What works well:**
- Core library (Builder, Scanner, Checker) is solid with good test coverage
- Format specification is well-designed (protobuf + zstd, multihash, deterministic serialization)
- CLI covers all major operations (gen, check, list, export, freshen, fetch)
- Test suite is thorough — builder, scanner, checker, GPG, CLI integration, corruption detection
- afero abstraction enables clean testing without filesystem side effects
**Key gaps for 1.0:**
- Missing repo infrastructure (`.golangci.yml`, `.editorconfig`, CI workflow)
- `manifest` type is unexported — consumers can't use it in their own type declarations
- GPG signing shells out to `gpg` subprocess — fragile and may not be installed
- Go module path inconsistency between `go.mod` and proto `go_package`
- `fetch` command lacks retry logic and has no HTTP timeout
- Missing fuzz tests for untrusted input deserialization
- Freshen CLI command has incomplete integration test coverage
closes#47closes#50
Co-authored-by: user <user@Mac.lan guest wan>
Co-authored-by: clawbot <clawbot@noreply.git.eeqj.de>
Co-authored-by: Jeffrey Paul <sneak@noreply.example.org>
Reviewed-on: #54
Co-authored-by: clawbot <clawbot@noreply.example.org>
Co-committed-by: clawbot <clawbot@noreply.example.org>
Follow-up to [PR #51](#51) — addresses sneak's review comment requesting Gitea Actions to run the checks.
## Changes
Adds `.gitea/workflows/check.yml` that runs `docker build .` on every push, matching the standard CI pattern used across all `sneak/*` repos (identical to [chat](https://git.eeqj.de/sneak/chat/src/branch/main/.gitea/workflows/check.yml) and [dnswatcher](https://git.eeqj.de/sneak/dnswatcher/src/branch/main/.gitea/workflows/check.yml)).
Since the Dockerfile already runs `make check` (fmt-check, lint, test), a successful build implies all checks pass.
The `actions/checkout` action is pinned by SHA (`@11bd71901bbe5b1630ceea73d27597364c9af683` = v4.2.2) per REPO_POLICIES.
## Verification
`docker build .` passes clean.
Reviewed-on: #53
Co-authored-by: clawbot <sneak+clawbot@sneak.cloud>
Co-committed-by: clawbot <sneak+clawbot@sneak.cloud>
Closes #49
`.index.mf` is a generated manifest file produced at runtime by `mfer gen`. It was committed to the repo but should not be tracked in version control.
Changes:
- `git rm .index.mf` to remove it from tracking
- Added `.index.mf` to `.gitignore` to prevent accidental re-commits
Co-authored-by: user <user@Mac.lan guest wan>
Reviewed-on: #52
Co-authored-by: clawbot <clawbot@noreply.example.org>
Co-committed-by: clawbot <clawbot@noreply.example.org>
Closes#39
Splits the Dockerfile into a dedicated lint stage using the `golangci/golangci-lint` image directly, rather than copying the binary into the builder stage.
## Changes
### Dockerfile
- **Lint stage** (`AS lint`): Uses the pre-built `golangci/golangci-lint` image (pinned by sha256) to run `make fmt-check` and `make lint`. This is a self-contained stage with its own `go mod download` and source copy.
- **Builder stage** (`AS builder`): Runs only `make test` and the final binary build. No longer needs golangci-lint installed.
- **Stage dependency**: `COPY --from=lint /src/go.sum /dev/null` forces BuildKit to always execute the lint stage (without this, unused stages are silently skipped).
- Both stages touch `mfer/mf.pb.go` to prevent make from trying to regenerate via protoc.
With BuildKit, the lint and builder stages run in parallel after their shared `go mod download` layers complete, so lint/formatting failures surface much faster without blocking on test execution.
### Makefile
- Added `lint` to the `check` target prereqs: `check: test lint fmt-check` (was `check: test fmt-check`), matching the [REPO_POLICIES](https://git.eeqj.de/sneak/prompts/raw/branch/main/prompts/REPO_POLICIES.md) requirement.
Co-authored-by: clawbot <clawbot@noreply.git.eeqj.de>
Reviewed-on: #45
Co-authored-by: clawbot <clawbot@noreply.example.org>
Co-committed-by: clawbot <clawbot@noreply.example.org>
sneak
clawbot
