Following private discussion, this version of the patch enables
-fno-strict-aliasing selectively by architecture, to avoid possibly
slowing down those that don't need it.
(I included sparc, which currently works, because the big-endian code is
undefined as opposed to guaranteed wrong, and henc
The test_cppbind issue appears to be caused by using a plain char
(unsigned on affected platforms) to hold -1: it can be triggered on
amd64 by the -funsigned-char compiler flag, and is there fixed by the
attached patch.
I can't reproduce the binobj bug, but enabling -fno-strict-aliasing may
f
2 matches
Mail list logo
