CI system cryptic error

[Marco Atzeri]

On 21.01.2022 17:03, Hamish McIntyre-Bhatty wrote:
> On 21/01/2022 14:06, Jon Turney wrote:
>> On 20/01/2022 15:50, Hamish McIntyre-Bhatty wrote:
>>> Hi there,
>>>
>>> Recently, I created a test package for python-imaging, and the CI 
>>> system gave a build error that I didn't see locally:
>>>
>>> *** ERROR: unknown wheel filename.
>>>
>>> This only occurred for the Python 3.8 build (3.6 and 3.7 are 
>>> unaffected). Considering some of the library name changes between 
>>> these versions, is it possible that this is a bug in the CI tool 
>>> setup or in cygport?
>>
>> Weird.
>>
>> I assume it's this job: 
>> https://github.com/cygwin/scallywag/actions/runs/1592256829
> 
> Yep, sorry I forgot to add the link.
> 
>>
>> In which case further note it only failed with this error on i686, 
>> x86_64 seems to have built ok.  That might explain why you don't see 
>> it locally, at least, but it's still puzzling that there's a difference.
> 
> No I did a local i686 build too, and it was fine...
> 
>>
>> This error comes from here:
>>
>> https://cygwin.com/git/?p=cygwin-apps/cygport.git;a=blob;f=cygclass/python-wheel.cygclass;h=90b0faac71eea6005401b4232e5b334bdc81b756;hb=HEAD#l190 
>>
>>
>> and means something more like "couldn't find the wheel file for python 
>> $ver"
>>
>> I downloaded the builddir archive, and inspecting that it seems that a 
>> wheel for python3.8 (and 3.9!) isn't being generated.
>>
>>> $ tar -tf builddir.tar.xz | grep whl
>>> python-imaging-8.4.0-1.i686/build/dist/Pillow-8.4.0-cp36-cp36m-cygwin_3_3_3_i686.whl 
>>>
>>> python-imaging-8.4.0-1.i686/build/dist/Pillow-8.4.0-cp37-cp37m-cygwin_3_3_3_i686.whl 
>>>
>>
>> There seems to be completely different output between x86_64 and i686 
>> when building the python wheel, but I have no idea why!
>>
> 
> Hmm, very strange. Maybe just a freak error? I'll try building again, 
> seeing as I have a new version now anyway, and see what happens. I guess 
> seeing as 64-bit builds are the future for Cygwin, I'll just ignore this 
> issue if it happens again, as long as I check the 32-bit builds I built 
> work.
> 
> Hamish
> 

I saw something similar but I do not remember the details.

Cygport is looking for something that has changed name or it is not 
available anymore. Probably the old name is still used by Python2 and
it is the reason why the error seldom show up in local build of
maintainers.

cygport --debug should provide some hints