Gentoo Websites Logo
Go to: Gentoo Home Documentation Forums Lists Bugs Planet Store Wiki Get Gentoo!
Bug 258337 - sqlite should filter -ffast-math
Summary: sqlite should filter -ffast-math
Status: RESOLVED INVALID
Alias: None
Product: Gentoo Linux
Classification: Unclassified
Component: New packages (show other bugs)
Hardware: x86 Linux
: High enhancement (vote)
Assignee: Petteri Räty (RETIRED)
URL:
Whiteboard:
Keywords:
: 282076 (view as bug list)
Depends on:
Blocks:
 
Reported: 2009-02-09 15:42 UTC by ultimA
Modified: 2009-08-20 09:17 UTC (History)
2 users (show)

See Also:
Package list:
Runtime testing required: ---


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description ultimA 2009-02-09 15:42:50 UTC
When compiling dev-db/sqlite-3.6.6.2, util.c throws compiler error "sqllite cannot be built with -ffast-math" - or something like that, I cannot remember the exact wording. The emerge fails in consequence.

sqlite ebuilds should filter the -ffast-math option.

Reproducible: Always
Comment 1 Jeremy Olexa (darkside) (RETIRED) archtester gentoo-dev Security 2009-02-09 15:57:36 UTC
Yes, there are known issues with this flag. If we filter it for one package, then we will have to filter it for half the tree. That doesn't seem sensible since the gcc man page says that it can cause broken code.

gcc man page:
This option should never be turned on by any -O option since it can
result in incorrect output for programs which depend on an exact     
implementation of IEEE or ISO rules/specifications for math functions.  

So, this "bug" won't be fixed. You can feel free to inquire upstream and see if they will choose to support this flag.
Comment 2 ultimA 2009-02-09 18:36:11 UTC
I know it's unsafe to use this compiler option, but that's not my point. You also should *not* filter this flag for "half of the tree", only for this single package, and here is my reasoning:

I perfectly know that you should not enable this option globally, but there are crazy people like me who like to experiment. I will NOT report instability problems for packages since I know I've done something I shouldn't have. However, I take my chances, the problems might not surface for me, or the broken code might not be broken with the next compiler version. The problem is that even though I started it on purpose, my unattended emerge process (that could last several days) will fail miserably under all circumstances.

The reason you should filter this flag in sqlite's ebuild is not because it might/will result in broken code, but because there is no chance the emerge will end successfully, thus could result in losing a day's cpu work if a long emerge is left unattended.

Sry for reopening this bug, but I feel you initially misunderstood my point. If you still think it's an invalid report, I will accept your decision.
Comment 3 Jeremy Olexa (darkside) (RETIRED) archtester gentoo-dev Security 2009-02-09 18:41:58 UTC
I still think it is invalid because "more than a few" of the flags that gcc provides will result in the build process dying because they are not safe to use. We can't filter them all..

Regardless, at this point, it is the maintainer's decision because you are persistant ;)
Comment 4 Petteri Räty (RETIRED) gentoo-dev 2009-02-09 23:50:44 UTC
Use emerge --keep-going.
Comment 5 Petteri Räty (RETIRED) gentoo-dev 2009-08-20 09:17:27 UTC
*** Bug 282076 has been marked as a duplicate of this bug. ***