build: 9c and 9l can report again

Both filter the compiler's output through $egrep, and plan9port assigns
that variable nowhere -- six scripts read it, one unrelated developer
script sets it.  The pipeline dies on the empty command and takes the
whole diagnostic with it, so -Wall -Wextra was decorative and a build
that failed to compile printed nothing but an exit status.  That is how
the tests' own dialect bug hid until it was hunted with a patched 9c.

Both scripts source $PLAN9/config, which the distribution does not ship,
so one line in it is the fix the scripts themselves ask for; it needs no
patching of a packaged file and covers 9c and 9l together.  The daemon it
builds is byte-identical, and with the diagnostics visible the whole tree
and every test object compile without a warning.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-08-17 20:04:07 +09:00
parent ad2f2fc5eb
commit f6dd056ffb

View File

@@ -42,4 +42,9 @@ RUN dbus-uuidgen --ensure=/etc/machine-id
ENV PLAN9=/usr/lib/plan9 ENV PLAN9=/usr/lib/plan9
ENV PATH=$PATH:/usr/lib/plan9/bin ENV PATH=$PATH:/usr/lib/plan9/bin
# 9c and 9l filter their own output through $egrep, which plan9port never
# assigns, so the pipeline dies and takes every warning and error with it.
# Both source $PLAN9/config, which the distribution does not ship.
RUN echo 'egrep="grep -E"' > $PLAN9/config
WORKDIR /src WORKDIR /src