Processed: gdc: FTBFS: ld: final link failed: Bad value

[Iain Buclaw]

On 4 December 2016 at 10:15, Iain Buclaw <ibuclaw at gdcproject.org> wrote:
> On 3 December 2016 at 22:29, Matthias Klose <doko at debian.org> wrote:
>> On 03.12.2016 22:13, Markus Koschany wrote:
>>> On 03.12.2016 22:00, Matthias Klose wrote:
>>>> reassign 846669 gdc-6
>>>> forcemerge 845377 846669
>>>> thanks
>>>>
>>>> please see #845377.
>>>>
>>>> This build is configured with --enable-default-pie.  The linker is now called
>>>> with -pie, but cc1d isn't passed either -fPIE, or -fPIC.  The work around is to
>>>> pass either -fPIC, or -no-pie.  Also passing -fPIE doesn't have an effect.
>>>
>>> Hi,
>>>
>>> sorry I'm not a D expert. Do you intend to fix #845377 before the Freeze or are all
>>> reverse-dependencies supposed to work around this issue by passing either -fPIC or -no-pie to configure?
>>
>> Please give Iain some time to have a look at this.
>>
>
> I'll be spending today on this.  Hopefully not all. :-)
>
> Regards

It looks like the pre-emptive check for "is this symbol external?" is
not right.  The codegen pass thinks it should be static, but it is
infact not, and so it emits the call without @PLT.  This only affects
D templates.

Incidentally, I recently cleaned up the area where this is done,
making it a small fix in upstream.  Will have a look at debian's
source package to prepare the relevant patch.

Regards
Iain