tomerfiliba / plumbum / 29844480577

88%

push

github

fix(cli): don't change the global SIGPIPE disposition (#834)

* fix(cli): don't change the global SIGPIPE disposition

Setting SIGPIPE to SIG_DFL in Application.run() made any later socket
send() to a closed peer kill the whole process; rpyc servers died this
way when a client disconnected. Catch BrokenPipeError around the run
body instead, as the Python signal docs recommend, keeping the piped-
help fix from #727.

Assisted-by: ClaudeCode:claude-fable-5

* refactor(cli): flatten the nested try in Application.run

Move the parse/dispatch body into _parse_and_dispatch() so run() has a
single try for BrokenPipeError. Also drops a type-ignore: the subapp
result now assigns to a plain Application.

Assisted-by: ClaudeCode:claude-fable-5

* fix(cli): flush stdout before exit so a late EPIPE is caught

Help output can sit in the stdout buffer after head exits without any
write failing; the EPIPE then only surfaced at interpreter shutdown,
printing 'Exception ignored ... BrokenPipeError' to stderr (seen on
Linux CI). Flush inside the guarded region so the handler catches it,
and add a deterministic regression test for the race.

Assisted-by: ClaudeCode:claude-fable-5

1328 of 1782 branches covered (74.52%)

38 of 42 new or added lines in 1 file covered. (90.48%)

5504 of 6291 relevant lines covered (87.49%)

12.41 hits per line