This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] buiild machinery: fix the profiledbootstrap breakage.


On Wed, Mar 12, 2008 at 11:01 PM, Paolo Bonzini <bonzini@gnu.org> wrote:
> Seongbae Park (박성배, 朴成培) wrote:
>  > Hi build machinery maintainers,
>  >
>  > Paolo's patch on Feb 20th broke the profiledbootstrap
>  > by building crt* with -fprofile-generate during stageprofile,
>  > which makes the compiler from stageprofile unable to link any executable
>  > (unless you pass -fprofile-generate, that is).
>  >
>  > The below patch fixed the problem (along with my other pending patch
>  > on libiberty/cplus-dem.c). Tested by doing a profiledbootstrap on
>  > i686.
>  > Ok for mainline ?
>
>  Why is it necessary for all-target-gcc?  If it's not really necessary,
>  maybe it's better to remove the flag in libgcc's makefile (see audit
>  trail of PR32009 for an example).

Now that I look, I seem to have lost the log file that shows
why all-stageprofile-gcc also needs this change and my memory fails me too,
but I do remember that I first tried dropping -fprofile-generate from
all-stageprofile-libgcc only and that still didn't fix the problem.
I'll try again and see if I can figure out why.

While I think my patch is slightly better than filtering in libgcc makefile
since this is a bootstrap effect (rather than generic "don't use this flag
for building a library" like PR32009),
I don't mind how we fix this. If you think filtering in libgcc makefile
is the right approach, I'll produce a patch for that.

>  Sorry for the breakage.
>
>  Paolo

Obviously many people don't care about profiledbootstrap, so no worries :)

Seongbae

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]