need help making src:fuzzylite reproducible

[Vagrant Cascadian]

On 2020-10-28, Johannes Schauer wrote:
> Quoting Vagrant Cascadian (2020-10-28 16:46:11)
>> >> Unfortunately, the reprotest job of my salsa CI pipeline still fails:
>> >> 
>> >> https://salsa.debian.org/debian/fuzzylite/-/pipelines/189461
>> >> 
>> >> I do not understand why. You can also use above link to download the artifacts
>> >> and investigate yourself. The html output of diffoscope is in the first link
>> >> above. The only odd thing that I can see is this extra string '\k#pk'. I do not
>> >> know where that comes from and why it's only part of one build and not the
>> >> other.
>> >
>> > I now uploaded the new Debian revision despite the unreproducibility problem
>> > above. The package has now been processed by reproducible.debian.net:
>> >
>> > https://tests.reproducible-builds.org/debian/rb-pkg/unstable/amd64/diffoscope-results/fuzzylite.html
>> >
>> > Looking at the diff, I don't understand where the differences come from.
>> >
>> > Can anybody help?
>> 
>> A somewhat educated guess is it has successfully stripped out the build
>> path from some things, but because the length of the build path is
>> different between the two builds, the padded area where the build path
>> would have been stored is still longer. :/
>
> I made a test and patched away all instances of __FILE__ from the upstream
> source and then did a build with fixfilepath disabled by setting
> reproducible=-fixfilepath in debian/rules.
>
> There are still some weird differences in length that I cannot explain. See
> salsa CI output
>
> https://salsa.debian.org/debian/fuzzylite/-/jobs/1114548
>
> or the attached diffoscope output. Maybe this is a completely different issue,
> but I would've thought that with __FILE__ gone from the upstream source, there
> should be no differences left?

Just to rule out the path-length issue. Try doing two builds with
different paths, of the same length, such as:

  /build/package-version/1234 and /build/package-version/5678

Both our test infrastructure and reprotest vary the path-length, and
I've seen this get embedded into builds regardless of -ffile-prefix-map.
We should probably make a new issue for this...


If that's not it, I'm not sure, unfortunately.


live well,
  vagrant
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 227 bytes
Desc: not available
URL: <http://alioth-lists.debian.net/pipermail/reproducible-builds/attachments/20201028/adfcf8e3/attachment.sig>