Bug 581022 – dev-qt/qtscript - error: undefined reference to 'cti_vm_throw'

• Status: RESOLVED UPSTREAM
• Product: Gentoo Linux

[Description Maxim Sheviakov 2016-04-24 09:41:23 UTC]

Hello.

qtscript-5.6.0 doesn't compile because of absent openvg.h file. I'm unsure, but that may be related to Mesa. Maybe a simple patch for that is needed?

[Comment 1 Maxim Sheviakov 2016-04-24 09:41:43 UTC]

Created attachment 431780 [details]
emerge --info output

[Comment 2 Maxim Sheviakov 2016-04-24 09:42:48 UTC]

Created attachment 431782 [details]
build.log of qtscript (compressed)

[Comment 3 Davide Pesavento (RETIRED) 2016-04-24 21:24:10 UTC]

openvg has nothing to do with this, you're looking at the wrong error.

The actual error line is:
/var/tmp/portage/dev-qt/qtscript-5.6.0/temp/ccC9F0pi.ltrans0.ltrans.o(.text+0x4c): error: undefined reference to 'cti_vm_throw'

However I cannot make sense of your build.log due to '-s' in your MAKEOPTS. Please attach a new build log without '-s'.

Does it build without LTO?

[Comment 4 Maxim Sheviakov 2016-04-25 12:40:43 UTC]

Well, yeah - stuff compiles without LTO. It's upstream bug, then, as other QT stuff compiles and runs okay with LTO enabled.

[Comment 5 Davide Pesavento (RETIRED) 2016-04-25 14:29:55 UTC]

*** Bug 581024 has been marked as a duplicate of this bug. ***

[Comment 6 Petross404(Petros S) 2017-04-22 13:49:17 UTC]

(In reply to Maxim Sheviakov from comment #4)
> Well, yeah - stuff compiles without LTO. It's upstream bug, then, as other
> QT stuff compiles and runs okay with LTO enabled.

I can confirm that =dev-qt/qtscript-5.7.1 continues to fail with :
error: undefined reference to 'cti_vm_throw'

[Comment 7 Petross404(Petros S) 2017-04-22 13:50:30 UTC]

Created attachment 470642 [details]
build.log (qtscript-5.7.1)

[Comment 8 Davide Pesavento (RETIRED) 2017-04-23 13:44:07 UTC]

*** Bug 616100 has been marked as a duplicate of this bug. ***

[Comment 9 Petross404(Petros S) 2017-05-14 20:35:49 UTC]

Created attachment 472626 [details, diff]
fix-cti_vm_throw.patch

This fixed the link problem for qtscript-5.7.1

[Comment 10 Davide Pesavento (RETIRED) 2017-05-14 20:38:09 UTC]

(In reply to Petross404(Petros S) from comment #9)
> Created attachment 472626 [details, diff] [details, diff]
> fix-cti_vm_throw.patch
> 
> This fixed the link problem for qtscript-5.7.1

is this an upstream patch?

[Comment 11 Petross404(Petros S) 2017-05-14 20:43:00 UTC]

(In reply to Davide Pesavento from comment #10)
> (In reply to Petross404(Petros S) from comment #9)
> is this an upstream patch?

No, it's just a patch I created after reading comments in various forums.

https://bugs.webkit.org/show_bug.cgi?id=56088
http://markmail.org/thread/cgmnxgm3ftjqtxa5

Oh, I forgot to mention that I used bfd. Do you want me to test with gold and report as well?

[Comment 12 Davide Pesavento (RETIRED) 2017-05-14 21:17:29 UTC]

(In reply to Petross404(Petros S) from comment #11)
> Oh, I forgot to mention that I used bfd. Do you want me to test with gold
> and report as well?

You're most likely already using gold, qt5 uses it by default when available.
I'm more concerned about clang... I don't think it supports __attribute__((used))...

[Comment 13 Petross404(Petros S) 2017-05-14 21:27:56 UTC]

(In reply to Davide Pesavento from comment #12)
> You're most likely already using gold, qt5 uses it by default when available.
> I'm more concerned about clang... I don't think it supports
> __attribute__((used))...

One step at the time. The ebuild can conditionally apply the patch and fix the issue for the GCC users (I hope it workes for others too). 

As for the Clang users, well upstream must support LTO at some point for all compilers.