On Wed, Dec 16, 2020 at 11:06:45AM -0800, Adam Williamson wrote:
See I thought that too at first, and was going to cite it, but then I
thought, wait. The problem isn't that the update *actually broke the
ABI*, right? The problem is that it *unnecessarily bumped the soname*.
I think abidiff's job is to catch the *opposite* problem, isn't it?
Where the ABI changes but the soname isn't bumped.
I would need to check, but I suspect possibly in this case abidiff just
wouldn't do anything at all, because what it would seem to make sense
to do is run it only on pairs of shared libraries with identical
sonames from the two package builds. When the soname is bumped, it
wouldn't make sense to run abidiff, because you'd *expect* the ABI to
change in that case.
Ah, indeed. That could well be the case.
I wonder if we shouldn't setup a 'soname bump test', make it gating for
everything and require waiving it. It would be a extra step and more
hassle, but it would prevent unintended soname bumps from landing, it
would make it a deliberate choice.
kevin