rpmspec: Use NEVRA for binary packages queries #2995
Conversation
|
This wont fix it for other potential users of archsuffix though. |
|
May be headerIsSource() should be more correct... I'll look into it |
Needed for the next patch
|
Now to something completely different... What about just setting the SOURCERPM tag for binary packages that are parsed for querying? One could argue that we should move more of this kind of initialization to the parse stage so we get a more complete result earlier. But that is a crusade I am not willing to embark on for this kind of bug. There is also the question if that is really a smart move, given that with dynamic specs features we' rather set things late. So this is probably the cleanest solution for now that won't affect anything else. |
Avoid packages being detected as srpms erroneously Resolves: rpm-software-management#2819
|
Yeah, (looking into) populating SOURCERPM early was what I meant with fixing at the source. Based on a quick look, you could just move SOURCERPM insertion from processBinaryFiles() to, say, finalizeSpec() and be done with it with equal amount of work, without adding duplicate special paths. |
|
OK, I looked into this and there are road blocks everywhere. Let's just stick to the PR as is. I generally agree that this is not the way to do this but the build code is an entangled mess and moving stuff round at this point is something we just should not do. Looking at all the hidden assumptions and internal dependencies in the code is a story for another time and goes far beyond this issue. |
| runroot rpmspec --rpms \ | ||
| -q /data/SPECS/hello.spec | grep src | ||
| runroot rpmspec --rpms \ | ||
| -q /data/SPECS/hello.spec | cut -d. -f-2 |
There was a problem hiding this comment.
Put these two checks into separate RPMTEST_CHECK() blocks, it makes the tests far more reliable when you don't need to guess which part output what. There are a bunch of tests that do this before we learned that you can have many checks inside a setup/clean pair 😅 but it's a bad practise we should not do in new tests.
Also, these are not writing anything so you don't need RPMDB_INIT (which has non-zero cost).
| ], | ||
| [0], | ||
| [hello-1.0-1.src | ||
| ], |
There was a problem hiding this comment.
I'd put this in the same SETUP/CLEANUP group as the --rpms tests, just in its own RPMTEST_CHECK() block. These have a not-entirely-trivial cost especially now in the containerized setup. This doesn't write anything either so RPMDB_INIT is unncessary (and lack of which actually guards against things unexpectedly starting to write stuff too).
|
The build code is a mess for sure, but adding hacks on top of hacks only makes it worse. |
|
Took a quick look at my own suggestion in the earlier comment and it brings out some truly WTF failures 😆 |
|
Here's a simple and straightforward way source headers are always indentified as such right after parse: #3012 |
|
Aaargh, except that the issue here was not positively identifying source headers but binaries 🤦 |
|
Added a second commit there to deal with RPMTAG_SOURCERPM too: |
|
Merged #3012 instead |
This fixes a regression that was present since 4.18 that printed them out as source/nosource packages as the headers are no proper binary packages. Setting the default query to "%{NEVRA}\n" forces the right output.
Resolves: #2819