Merges main into next as a two-parent merge, so main becomes an ancestor of next and the later merge of next into main does not conflict. That later merge is what lands the script/ scaffold from #16 on main.
From main this brings golangci-lint v2.12.2 with .golangci.yml, and the fix that makes every handler emit slog attributes.
Only README.md and TODO.md conflicted, and both keep everything from both sides. README.md has main's Attribute output section, then next's Entrypoints section. TODO.md lists both sides' completed steps, newest first; this merge finishes no item in it, so nothing else moved. The other files merged cleanly: the Makefile is still next's @script/* shims, the Dockerfile has main's v2.12.2 lint image, and .golangci.yml is main's, unchanged.
The manager merges this with Gitea's merge-commit style, not squash, to keep that ancestry.
Deviation: one merge commit instead of one squashed commit, for the ancestry above.
Judgement call: the Dockerfile keeps its separate make fmt-check, make lint and make test steps instead of one make check step (item 4 of the issue's definition of done); they are the three checks script/check runs.
Judgement call: lines that were already out of date on both sides (README.md Current Status, TODO.md Status and Next Step) are left as they were, since this merge did not make them false.
Unverified: make check was not run on the host, because it runs the linter outside Docker.
Model: opus-5-5
Merges `main` into `next` as a two-parent merge, so `main` becomes an ancestor of `next` and the later merge of `next` into `main` does not conflict. That later merge is what lands the `script/` scaffold from https://git.eeqj.de/sneak/simplelog/issues/16 on `main`.
From `main` this brings golangci-lint v2.12.2 with `.golangci.yml`, and the fix that makes every handler emit `slog` attributes.
Only `README.md` and `TODO.md` conflicted, and both keep everything from both sides. `README.md` has `main`'s Attribute output section, then `next`'s Entrypoints section. `TODO.md` lists both sides' completed steps, newest first; this merge finishes no item in it, so nothing else moved. The other files merged cleanly: the `Makefile` is still `next`'s `@script/*` shims, the `Dockerfile` has `main`'s v2.12.2 lint image, and `.golangci.yml` is `main`'s, unchanged.
The manager merges this with Gitea's merge-commit style, not squash, to keep that ancestry.
- Deviation: one merge commit instead of one squashed commit, for the ancestry above.
- Judgement call: the `Dockerfile` keeps its separate `make fmt-check`, `make lint` and `make test` steps instead of one `make check` step (item 4 of the issue's definition of done); they are the three checks `script/check` runs.
- Judgement call: lines that were already out of date on both sides (`README.md` Current Status, `TODO.md` Status and Next Step) are left as they were, since this merge did not make them false.
- Unverified: `make check` was not run on the host, because it runs the linter outside Docker.
Model: opus-5-5
Add the canonical .golangci.yml (v2 schema, all linters enabled with a
small documented disable list) and pin the Dockerfile lint stage to
golangci/golangci-lint:v2.12.2 by tag and digest, replacing the old
v1.64.8 digest-only pin.
Fix all findings surfaced by the v1 to v2 jump without changing any
exported signatures or behavior:
- add package and exported-symbol doc comments (revive)
- rename unused handler parameters to underscore (revive)
- check or explicitly discard error returns (errcheck, errchkjson)
- wrap errors with %w instead of %v (err113)
- use http.NewRequestWithContext instead of http.Post (noctx)
- replace fmt.Println with fmt.Fprintln(os.Stdout, ...) (forbidigo)
- name magic numbers as constants (mnd)
- add explicit slog.LevelDebug case (exhaustive)
- interface{} to any (modernize)
- move tests to the simplelog_test package (testpackage) and add
t.Parallel() (paralleltest)
- move NewWebhookHandler above its methods (funcorder)
- whitespace, line-length, and blank-line fixes (wsl_v5, whitespace,
nlreturn, lll, embeddedstructfieldcheck)
- nolint with justification for the intentional init/global design
(gochecknoinits, gochecknoglobals) and interface-returning
constructor (ireturn)
Every handler drops slog attributes: Handle never reads the record's
attributes, and WithAttrs and WithGroup return the receiver unchanged. So
slog.Info("casting", "device", d) loses its field.
The test checks the bytes each handler writes, directly and through
MultiplexHandler: record attributes, WithAttrs, WithGroup, slog.Group nesting
and LogValuer resolution. It also pins what a fix must not break: logged values
are never modified, durations are nanoseconds in JSON and "3s" on the console,
and the record's own field names win a key collision. Console keys and values
are quoted as slog.NewTextHandler quotes them, invalid UTF-8 included.
This commit adds only the test, and it fails; the fix follows.
Rule suppressed: paralleltest; the tests swap the process-wide os.Stdout.
Refs: #19
Model: opus-5-5
Handle never read the record's attributes, and WithAttrs and WithGroup
returned the receiver unchanged. The JSON and webhook handlers also marshaled
slog.Record itself, whose attributes are unexported.
Each handler now carries a handlerAttrs value, copied rather than mutated, so
sibling loggers cannot leak attributes into each other. JSON and webhook
output nest groups as objects; the console appends key=value pairs with dotted
group keys, quoted as slog.NewTextHandler quotes them, invalid UTF-8
included. Logged values are never written to. In JSON the record's own fields
win a key collision, and a repeated key keeps its last value unless both are
groups, which merge; the README documents both.
Deviation: DEL is quoted on the console; the stdlib leaves it bare.
Model: opus-5-5
Merges main into next as a two-parent merge, so main becomes an ancestor
of next and the later merge of next into main does not conflict. This
brings in golangci-lint v2.12.2 with .golangci.yml, and the fix that
makes every handler emit slog attributes.
README.md and TODO.md conflicted. README.md keeps main's Attribute
output section, followed by next's Entrypoints section. TODO.md keeps
the completed steps from both sides, newest first. The other files
merged cleanly: the Makefile is still next's script/ shims, and the
Dockerfile has main's v2.12.2 lint image.
Model: opus-5-5
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Merges
mainintonextas a two-parent merge, somainbecomes an ancestor ofnextand the later merge ofnextintomaindoes not conflict. That later merge is what lands thescript/scaffold from #16 onmain.From
mainthis brings golangci-lint v2.12.2 with.golangci.yml, and the fix that makes every handler emitslogattributes.Only
README.mdandTODO.mdconflicted, and both keep everything from both sides.README.mdhasmain's Attribute output section, thennext's Entrypoints section.TODO.mdlists both sides' completed steps, newest first; this merge finishes no item in it, so nothing else moved. The other files merged cleanly: theMakefileis stillnext's@script/*shims, theDockerfilehasmain's v2.12.2 lint image, and.golangci.ymlismain's, unchanged.The manager merges this with Gitea's merge-commit style, not squash, to keep that ancestry.
Dockerfilekeeps its separatemake fmt-check,make lintandmake teststeps instead of onemake checkstep (item 4 of the issue's definition of done); they are the three checksscript/checkruns.README.mdCurrent Status,TODO.mdStatus and Next Step) are left as they were, since this merge did not make them false.make checkwas not run on the host, because it runs the linter outside Docker.Model: opus-5-5
Add the canonical .golangci.yml (v2 schema, all linters enabled with a small documented disable list) and pin the Dockerfile lint stage to golangci/golangci-lint:v2.12.2 by tag and digest, replacing the old v1.64.8 digest-only pin. Fix all findings surfaced by the v1 to v2 jump without changing any exported signatures or behavior: - add package and exported-symbol doc comments (revive) - rename unused handler parameters to underscore (revive) - check or explicitly discard error returns (errcheck, errchkjson) - wrap errors with %w instead of %v (err113) - use http.NewRequestWithContext instead of http.Post (noctx) - replace fmt.Println with fmt.Fprintln(os.Stdout, ...) (forbidigo) - name magic numbers as constants (mnd) - add explicit slog.LevelDebug case (exhaustive) - interface{} to any (modernize) - move tests to the simplelog_test package (testpackage) and add t.Parallel() (paralleltest) - move NewWebhookHandler above its methods (funcorder) - whitespace, line-length, and blank-line fixes (wsl_v5, whitespace, nlreturn, lll, embeddedstructfieldcheck) - nolint with justification for the intentional init/global design (gochecknoinits, gochecknoglobals) and interface-returning constructor (ireturn)Every handler drops slog attributes: Handle never reads the record's attributes, and WithAttrs and WithGroup return the receiver unchanged. So slog.Info("casting", "device", d) loses its field. The test checks the bytes each handler writes, directly and through MultiplexHandler: record attributes, WithAttrs, WithGroup, slog.Group nesting and LogValuer resolution. It also pins what a fix must not break: logged values are never modified, durations are nanoseconds in JSON and "3s" on the console, and the record's own field names win a key collision. Console keys and values are quoted as slog.NewTextHandler quotes them, invalid UTF-8 included. This commit adds only the test, and it fails; the fix follows. Rule suppressed: paralleltest; the tests swap the process-wide os.Stdout. Refs: #19 Model: opus-5-5Review passed.
Model: opus-5-5