[youdie006]

What type of PR is this?

• bug

What this PR does / why we need it:

A TimestampFlag whose layout carries a month and day but no year returns today's date instead of the parsed date, but only when the value is January 1.

Config: TimestampConfig{Layouts: []string{"01-02"}, Timezone: time.UTC}

--time 01-01   before: 2026-09-08T00:00:00Z   after: 2026-01-01T00:00:00Z
--time 01-02   2026-01-02T00:00:00Z  (both)
--time 03-04   2026-03-04T00:00:00Z  (both)
--time 12-31   2026-12-31T00:00:00Z  (both)

timestampValue.Set decides "this layout has no date" by comparing the parsed instant against defaultTS, a probe parsed from time.TimeOnly, which lands on year-0 January 1 (flag_timestamp.go:90-93). A layout that does carry a date parses 01-01 onto that exact same day, so the two cases are indistinguishable and the date-less branch fires, overwriting month and day with time.Now().

The fix asks the layout rather than the parsed value. layoutHasDate round-trips a reference date that is deliberately not January 1 (0002-03-04) through Format/Parse: a date-less layout drops it and comes back as January 1, a layout with a date returns March 4.

• flag_timestamp.go — capture the layout that matched in the parse loop, add the layoutHasDate guard to the first branch, add the helper.
• flag_test.go — TestTimestampFlagApply_YearlessLayoutJanuaryFirst, three rows.

The one-sentence version: the very next branch (flag_timestamp.go:104, else if timestamp.Year() == 0) already handles "layout has a date but no year" correctly — January 1 is the single input that never reaches it.

matchedLayout is load-bearing: the loop otherwise discards the layout and the check cannot be made without it. The helper is unexported, so make generate produces no change to godoc-current.txt.

Which issue(s) this PR fixes:

None — found by reading flag_timestamp.go. I did not open one first since #2400 and #2401 landed the same way, but happy to file one if you would prefer.

Special notes for your reviewer:

time.Stamp-style layouts ("Jan _2 15:04:05", i.e. syslog and journal timestamps) are the canonical reason to use a year-less layout, and they are affected too — that row is in the test.

One honest limitation: flag_timestamp.go:92 calls time.Now() directly, so on January 1 itself the buggy code returns the right answer by coincidence and the two January rows would pass against the bug. That day is unobservable by construction, not a gap in the test. The March 4 control row and the mutation checks below still discriminate on every other day.

I also considered detecting date components by scanning the layout string for 1/01/Jan/2/02/_2 tokens, and rejected it — the round-trip probe is shorter and does not have to enumerate Go's layout vocabulary. If a layout is lossy enough that Format/Parse does not round-trip, the err == nil guard makes layoutHasDate return false, which is exactly today's behaviour, so an exotic layout degrades to the status quo rather than regressing.

Testing

TestTimestampFlagApply_YearlessLayoutJanuaryFirst drives the production chain — cmd.Run → Action → cmd.Timestamp("time") — with two January 1 rows and a March 4 control.

• On main the two January rows fail (expected: 2026-01-01, actual: 2026-09-08) and the control passes.
• Mutation-checked both directions, failing disjoint tests: forcing layoutHasDate to false (restoring current behaviour) fails only the new test, and forcing it to true fails only the pre-existing TestTimestampFlagApply_ShortenedLayouts. Both sides of the boundary are pinned.
• make test passes (-race, coverage 99.3%). make vet, make lint, make ensure-goimports, make generate + make diffcheck, make v3diff, make check-binary-size (1.8MB) all clean, and golangci-lint run ./... reports 0 issues.
• Existing timestamp coverage still passes, including time.Kitchen, time.Stamp, StampMilli/Micro/Nano, TimeOnly, "15:04" and time.RFC3339.

Release Notes

Fixed `TimestampFlag` returning the current date instead of the parsed date when the layout has no year and the value is January 1.

---

AI assistance disclosure: this patch was found and written with Claude Code. Every number above is verbatim from running it against main and against this branch.

[dearchap]

Approving. Verified on the PR branch: the new test fails on main (bug reproduced) and passes with the fix; full go test -race ., go vet, gofmt, and make v3diff are clean. The round-trip probe (layoutHasDate) is exact and only changes behavior in the formerly buggy case — edge probing (year-only, year-month, day-of-year, time-only, multi-layout layouts) shows no regressions. Minor non-blocking notes: Hoist layoutHasDate out of the per-Set path if you want to avoid the extra Format/Parse on each parse; the test's time.Now().UTC().Year() has a narrow NYE UTC rollover race, consistent with existing tests.