Skip to content

chore: GH-208 delete three dead things and correct a signal name - #280

Open
joaodinissf wants to merge 3 commits into
mainfrom
chore/sweep-triage
Open

chore: GH-208 delete three dead things and correct a signal name#280
joaodinissf wants to merge 3 commits into
mainfrom
chore/sweep-triage

Conversation

@joaodinissf

Copy link
Copy Markdown
Collaborator

Three deletions and a correction from a codebase sweep. No behaviour changes.

OutputCaptureOutputCapture, ActionOutput, NewOutputCapture, Add
and Get had no reference outside their own file; the package captures output
through outputs.go. Not to be confused with the protobuf frontseat.ActionOutput,
which is a different and very much live type.

The discarded local-action countPublishBuildFinished named its seventh
parameter _, so countLocalResults computed a number that could not reach
OperationFinished, which has no field for it. With the parameter gone,
publishBuildFinished has no remaining use for its results map either.
ActionState.Local goes the same way: written on every completed and failed
action, read by nothing.

The signal name — three comments claimed SIGTERM where the code sends
os.Interrupt, which is SIGINT. The comments moved to the code rather than the
reverse: dev servers install Ctrl-C handlers, and syscall.SIGTERM does not
compile on Windows.

ActionResult.Local itself is deliberately untouched. An earlier reading of it
as never-set was wrong — sixteen in-daemon sites set it (conformance, plugin
sync, mise install, fix and generate tasks), and it drives ExecutedRemotely,
which frontseat.proto specifies as false for in-daemon actions.

Verifying that turned up a real defect, filed separately rather than fixed here:
Cached and Local are set by disjoint code paths, so ACTION_STATUS_CACHED
is unreachable and the CLI's "%d remote, %d local" split has a permanently
zero local bucket. The unreachable branch is left in place as the only surviving
evidence of the intent.

Relates to #208.

Published with assistance from Claude.

joaodinissf and others added 3 commits August 1, 2026 22:24
Nothing outside output_capture.go referenced OutputCapture, ActionOutput,
NewOutputCapture, Add or Get. The package captures action output through
outputs.go instead.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UdUyW1UwSfbg5tT7PYn5KD
PublishBuildFinished named its seventh parameter _, so countLocalResults
fed a value that could not reach OperationFinished, which has no field
for it. Removing the parameter leaves publishBuildFinished with no use
for its results map either.

ActionState.Local goes the same way: written on every completed and
failed action, read by nothing.

ActionResult.Local itself stays. It is set by the sixteen in-daemon
sites and drives ExecutedRemotely, which frontseat.proto specifies as
false for in-daemon actions.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UdUyW1UwSfbg5tT7PYn5KD
Three comments claimed SIGTERM; the code sends os.Interrupt, which is
SIGINT. SIGINT is the better choice here — dev servers install Ctrl-C
handlers, and syscall.SIGTERM does not compile on Windows — so the
comments move to the code rather than the reverse.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UdUyW1UwSfbg5tT7PYn5KD
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant