Hi
On 06/01/2025 18:15, Paul Gevers wrote:
Hi,
On 06-01-2025 12:38, Peter B wrote:
ISTM that the "exactly equal" version equality of the dependency
cannot itself cause the package
to be rebuilt, as the package will now FTBFS because the exact
version of the dependency will no longer be available.
The version should be calculated during the build, such that a rebuild
will update it and it will always be in sync. This is a field of a
binary, not of the source.
so the package name only goes into the source, and then the package name
plus version gets into the binary?
(It's not a Depends, it's a new field).
I get that, but the value of the field must be a build depends for it to
be "Built-Using"
The package will need updating to change the dependency version.
Slightly more work than just
a rebuild, and manual intervention is still needed.
That would be a PITA and wrong.
Does any know for sure what causes these checksums to change?
I think so. fp-fix-timestamps (which is a Debian helper to get
reproducible FreePascal packages in Debian, even if we patch upstream
source) is made to ease it a bit.
https://wiki.debian.org/ReproducibleBuilds/TimestampsInPPUGeneratedByFP
has more info.
But surely, timestamps &b checksums are different things?
Anyway, my understanding is the issue relates to packages that ship ppu
files, which would be lazarus & cge.
@Abou,
I can't see any purpose in having this bug assigned to winff, as winff
does not provide or consume any ppus from other packages.
I'm thinking adding to the workflow notes for fpc
https://salsa.debian.org/pascal-team/fpc/-/wikis/Debian%20FPC%20Packaging
something like...
"After an upload of fpc, check that lazarus builds, and if there are
checksum errors, giveback or upload a new version of lazarus.
Also check that castle-model-viewer builds, and if there are checksum
errors, giveback or upload a new version of castle-game-engine.
Regards,
Peter
