Hi Paul,
> + Don’t assume CHAR_BIT == 8.
I can agree with that...
> * tests/test-largefile.c, tests/test-year2038.c:
> Prefer compile-time to run-time tests.
... but not with that. I wrote in the comments:
This test fails if AC_SYS_YEAR2038 could not arrange for a time_t
that supports values > 2147483647.
This can happen
- on systems that are not year 2038 safe, or
- if the user specified --disable-year2038 at configure time.
It is intended that this test fails in these circumstances.
It is not intended that this test *doesn't compile* in these circumstances.
Because when I am running a testdir on some platforms, I am interested
in _all_ the test failures. When a single test not compile, I have to
1. edit the Makefile, remove the first two mentions of the particular test,
2. then run "gmake check 2>&1 | tee log3" again.
This is all painful and a waste of time if it occurs regularly. And yes,
I do want to continue to run testdirs on GNU/Hurd/x86, GNU/kFreeBSD/x86,
AIX/powerpc, Solaris 11.4/x86, Cygwin 2.9.0, Haiku regularly.
This change thus makes my regular testing work more tedious, without real
advantage. We need to revert that.
Bruno
