[gdal-dev] Slightly-off failing tests on i686

[yosh]

> we do have a CI configuration for i386 using clang in debug mode. Maybe
> your use of gcc in -O2 yields to slightly different results in some cases.

After encountering some wild GCC results when adding -ffloat-store, I switched
C(C|XX) to clang and it fared much better on almost all fronts. Thanks
for that!

Alas, there seems to be three remaining problems I'm encountering on i686
tests:

1. There's a few lingering multithreaded tiff test fails:
   - test_tiff_read_multi_threaded[False-False-100-100-3-1-creation_options11]
   - test_tiff_read_multi_threaded[False-False-100-100-3-1-creation_options12]
2. During test_tiff_write_17, a zombie `stat` process gets forked and lingers,
   locking up the test suite until I kill the `pytest` process. If I edit the
   target to use "-vv" on pytest, it /doesn't/ lock up the test, but rather it
   fails from a vague "MemoryError". This test isn't skipped by the
   SKIP_MEM_INTENSIVE_TEST environment variable (which I'm already setting
   to "YES")
3. Two asserts in gdrivers/test_gpkg_26 are off by a factor of 2. It's quite
   strange.

Here's a debug log: https://0x0.st/KLSG.log

Any ideas on what might be happening?

yosh

(P.S. apologies for the double-email even; forgot to :reply -a in aerc for the
mailing list x.x)