Bug 31340 – Dangerous C-flag for python

• Product: Gentoo Linux
• Status: RESOLVED WONTFIX

[Description Ervin Németh 2003-10-17 03:26:13 UTC]

With sys-devel/gcc-3.3.1-r5 building python does a segfault.

CFLAGS="-march=athlon-xp -mmmx -msse -m3dnow -mfpmath=sse,387 -O3
-maccumulate-outgoing-args -fomit-frame-pointer -funroll-loops
-fmove-all-movables -freduce-all-givs -fprefetch-loop-arrays -ffunction-sections
-fforce-addr -pipe"

Removing "-fforce-addr" solves the problem.


Reproducible: Always
Steps to Reproduce:

[Comment 1 Ervin Németh 2003-10-17 03:28:38 UTC]

Created attachment 19350 [details, diff]
Patch for dev-lang/python-2.2.3-r3

Here is my proposed patch to solve the problem.  I really don't know if the
bug
is CPU and/or gcc-verion dependent...

[Comment 2 Alastair Tse (RETIRED) 2003-10-17 07:21:20 UTC]

we can't possibly cater for all insane optimisations. it is a responsibility
of the user to not go crazy with CFLAGS.

[Comment 3 Ervin Németh 2003-10-18 02:20:59 UTC]

I don't see *your* problem here. I've found a conflict and supplied a patch.

Anyway is there a sharp and clean border between user and developer in Open
Source?

And please, don't insult your users, it's not nice.

[Comment 4 Alastair Tse (RETIRED) 2003-10-18 05:40:28 UTC]

i'm not delibrately insulting you. i'm just saying that i don't feel obliged
to filter out gcc optimisations that are just not standard. Gentoo allows
people to set their own CFLAGS to give them power to tweak, but when you
set them at unreasonably dangerous levels, it is up to the user (here, and
in my previous comment, user means "Gentoo User", not the user/dev thing,
because i'm a "Gentoo User" too) to be responible and know the consequences
of those CFLAGS.

http://www.gentoo.org/doc/en/faq.xml has something about over-optimizations.
although i don't even agree with the -frerun-*, but i'm very conservative
about that.

you can also think of it the other way, giving users more power to tweak,
we can't always stop them shooting themselves in the foot. and when we restrict
the cflags too much, people will start submitting bugs about how restrictive
we are and how gentoo is about choice. we have to walk a fine line between
giving users choice and restricting the choice. all in all, it is also possible
that the -fforce-addr flag you talk about is dependent on gcc version and
-march=, so without more datapoints, it is an isolated occurance.

even though i'm not going to accept this patch, i thank you for submitting
this bug and investigating which of those flags you have that breaks for
you. it maybe useful in the future when someone else tries those CFLAGS.