g-ir-scanner fails with python-3.8

[Jon Turney]

On 23/05/2021 12:50, Lemures Lemniscati via Cygwin-apps wrote:
> On Sun, 23 May 2021 00:00:25 -0600, Brian Inglis
>> On 2021-05-22 17:44, Lemures Lemniscati via Cygwin-apps wrote:
>>> On Wed, 19 May 2021 22:09:59 +0100, Jon Turney
>>>> On 19/05/2021 20:32, Ken Brown via Cygwin-apps wrote:
>>>>> Trying to build harfbuzz, I get the following python failure with
>>>>> python-3.8
>>>>>
[...]

Thanks for looking into this.

>> I question the definition successful when both logs contain GICOMP syntax errors
>> which appear non-fatal and non-terminal:
>>
>> i686:
>> <https://ci.appveyor.com/project/cygwin/scallywag/builds/39277866/job/8y08y0qcu3k4y69x#L1360>
>> GICOMP   /cygdrive/c/projects/playground/gobject-introspection-1.54.1-4.i686/src/gobject-introspection-1.54.1/gir/GL-1.0.gir
>> /usr/lib/gcc/i686-pc-cygwin/10/include/stddef.h:424: syntax error, unexpected identifier in '  __float128 __max_align_f128 __attribute__((__aligned__(__alignof(__float128))));' at '__float128'

This is 'expected', since (this version of?) g-ir-compiler doesn't know 
about __float128 as a built-in type.

There's no input which can legitimately make gcc terminate with segv, so 
trying to blame this on gobject-introspection is incorrect. There is a 
something going wrong in gcc, or (more probably) cygwin.

 From previous reports [1], [2] it seems to be a problem which only 
manifests itself on WOW64.

[1] https://cygwin.com/pipermail/cygwin-apps/2020-October/040548.html
[2] https://sourceware.org/legacy-ml/cygwin/2014-10/msg00552.html

> Oh, I overlooked them.
> 
> 
> I looked into a log corresponding to the current version 1.54.1-3 [1].
> 
> [1]: https://cygwin.com/cgi-bin2/jobs.cgi?srcpkg=gobject-introspection&id=2763
> 
> 
> * i686: No syntax error related to '__float128' [2]
>    at gobject-introspection-1.54.1/gir/GL-1.0.gir.
>    So, it is newly introduced

I think it just didn't get far enough, the same error is seen when 
building locally (which is how I had to build the package, because of 
the 'signal 11' problem)