Bootstrap comparison failures, again

Zack Weinberg zack@wolery.cumb.org
Sat Jan 22 22:58:00 GMT 2000


On Sat, Jan 22, 2000 at 10:42:03PM -0800, Mark Mitchell wrote:
> 
> Folks --
> 
>   I've confirmed that building without -j allows the bootstrap to
> succeed.  So, there is definitely a parallel make bug lurking in the
> fixincludes stuff.
> 
>   I tested this by doing brand-new builds with and without -j.

Based on the difference you posted, I'm guessing the trouble is not
with fixincludes itself but all the gunk surrounding it.  Your -j
build saw two different versions of stddef.h, which is not fixed, it's
copied from src/gcc/ginclude.  Look at stmp-int-headers and its
dependencies.

My understanding of the sequencing here is that build/include is
supposed to be created once, after the stage1 cc1 and xgcc are built
but before they are used to build anything.  It sounds like it is
getting recreated part way through stage2, which would be wrong.

zw



More information about the Gcc-bugs mailing list