On 14/04/2025 22:39, Brian Inglis via Cygwin-apps wrote:
On 2025-04-14 11:06, Jon Turney wrote:
On 13/04/2025 18:13, Brian Inglis via Cygwin-apps wrote:
Hi folks,

Just got an update to libvpx 1.15.1 which bumps ABI from 9 to 11,
because previous release 1.15.0 should have bumped ABI from 9 to 10,
but was marked as ABI compatible to prior release 1.14.1,
incompatibility found upstream in January,
but only released now in April on Google source repo!

Can I provide an override.hint with 1.15.1,
which builds new ABI package libvpx11 cygvpx-11.dll,
to replace version 1.15.0 libvpx9 cygvpx9.dll with 1.14.1,
and/or do I need to do some juggling with
1.15.1 OBSOLETES 1.15.0 libvpx9 cygvpx-9.dll,
or some nastier fudging of
null 1.15.0 packages uploaded manually?

The approach I would take is:

* remove the ABI 10 (1.15.0) version of the package

Thanks Jon,

So I just vault-ed libvpx-1.15.0-1 and got:
vault: package 'libvpx' version '1.15.0-1' marked as expirable

and that release has disappeared from libvpx-utils/-devel/9!

That was the previous version which was thought to be ABI compatible and was packaged and released as 9!
[...]
* add 1.15.0 to replace-versions: in override.hint for all the subpackages, instructing setup to downgrade to ABI 9 versions of those packages.

That is 1.14.1 - so I should add to 1.15.1 PKG_HINTS=override, and CYGWIN_FILES=override.hint, containing only replace-versions: 1.15.0, or should I also add keep: 1.14.1, or do I need to release 1.14.1-2 with that override?

Uh, no. override.hint files don't belong in source packages, because they aren't tied to a particular version.

I *think* you can hand-write the override.hint file and manually upload them, but maybe that doesn't work and I'll need to do that for you...

* package the ABI 11 (1.15.1) version of the package

* examine things which you've recently rebuilt which erroneously depend on libvpx9, when they should depend on libvpx10. Remove those versions, and rebuild with a release bump, depending on libvpx11.

Only the other subpackages are affected according to cygcheck-dep.

I think that should make everything consistent?
Hopefully for users!


Reply via email to