Re: [BUG REPORT] gnulib Android NDK and/or Cygwin build failure regressi

From:

osm0sis

Subject:

Re: [BUG REPORT] gnulib Android NDK and/or Cygwin build failure regression after 0c8a563f

Date:

Sat, 23 Apr 2022 21:53:56 +0000

The differences between the invoking commands in the build outputs I shared had a good clue, which was the new " - -Wno-unsuffixed-float-constants" in the broken one that was added by the breaking commit, but there should have been more options than that and the lone " - " definitely appeared problematic.

Comparing the Makefiles produced before/after the breaking commit showed the following new section:

GL_CFLAG_GNULIB_WARNINGS =

-Wno-cast-qual

-Wno-conversion

-Wno-float-equal

-Wimplicit-fallthrough

-Wno-sign-compare

-Wno-undef

-Wno-unused-function

-Wno-unused-parameter

-Wno-float-conversion

-Wno-pedantic

-Wno-sign-conversion

-Wno-type-limits

-Wno-unsuffixed-float-constants

But reading the script in m4/gnulib-common.m4 those should have ended up on one line.

Examined a copy of the conftest.out file the script parses those from and noticed it had CR+LF due to being from Windows Android NDK and/or on Cygwin, so realized the solution was to strip all the CRs out so that it can be parsed correctly, and the CRs in the extra lines were why the build output errors were weirdly truncated/appended mid-sentence.

I'm attaching my proposed patch to resolve the issue, and in the future contributors would just need to be mindful that any compiler output on Windows being parsed might have CRs.

Thanks for your consideration!

Chris