1 |
-----BEGIN PGP SIGNED MESSAGE----- |
2 |
Hash: SHA1 |
3 |
|
4 |
Simon Stelling wrote: |
5 |
> And we all like X freezing: |
6 |
> |
7 |
> http://bugs.gentoo.org/show_bug.cgi?id=80329 |
8 |
> |
9 |
> In short, yes they can break your system. Pretty easily even, if you |
10 |
> don't know what you're doing. |
11 |
> |
12 |
|
13 |
Well, I didn't see much that looked dangerous in the CFLAGS in the last |
14 |
example. Obviously -ffast-math is going to cause major issues, but if a |
15 |
package won't build correctly with something like -O3 it really needs to |
16 |
be filtered by the ebuild. Granted, in this case it was more obscure |
17 |
than just -O3 but there wasn't anything there which set off warning |
18 |
bells when I looked at it. |
19 |
|
20 |
I'm sure there was either some kind of GCC bug, or some sensitive aspect |
21 |
of the software itself (making assumptions about data structures in |
22 |
memory/etc) that caused the problem. Gentoo obviously can't track down |
23 |
and fix these, but unless we're talking about an unusually dangerous |
24 |
CFLAGS entry the solution is probably to add filters to ebuilds, and not |
25 |
expect users to use nothing but -O2 and maybe a few others just in case. |
26 |
|
27 |
Don't get me wrong - I'm not suggesting having devs or ATs build |
28 |
everything with 47 different CFLAGS before stabilizing either. I think |
29 |
that users with unusual CFLAGS should be prepared to be trailblazers, |
30 |
but if they can figure out which flag is killing the software and can |
31 |
file a bug report it should probably be added to the ebuild as a filter. |
32 |
|
33 |
Then again, that is probably close to what is happening now anyway - I'm |
34 |
sure that if a user filed a simple bug titled "please add |
35 |
- -ftree-vectorize to filter-flags for xyz/abc-1.23" it would probably get |
36 |
closed pretty quickly without much dev complaint. I do agree that devs |
37 |
have better things to do than to rebuild the package 47 times themselves |
38 |
to figure out what flag is killing things - and the bug submitter most |
39 |
likely can handle this on their own. |
40 |
-----BEGIN PGP SIGNATURE----- |
41 |
Version: GnuPG v1.4.5 (GNU/Linux) |
42 |
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org |
43 |
|
44 |
iD8DBQFFFStkG4/rWKZmVWkRAmsrAKCkg6pUjQbZXksxXycF10pxbFHwGACgwmYI |
45 |
Hm0bvKsGClbu3GWxPDxdqj8= |
46 |
=o1tf |
47 |
-----END PGP SIGNATURE----- |