Conversation
baberlevi
left a comment
There was a problem hiding this comment.
I feel like this approach is consistent with what's been done for other packages, e.g. singularity-legacy
|
Interesting. At first glance, I'm not a huge fan of "legacy" packages, but it does make life a lot easier whenever a package changes its build system. @tgamblin @alalazo how do we feel about this being the official approach to handling a build system change? Possible caveats:
|
I think at some point - maybe when the release process will be more stable / frequent? - we should start deprecating the oldest versions of packages (meaning removing the versions and cleaning recipes from anything that is specific to them).
There could be another approach based on repositories. If, instead of naming the old package
I'm not aware of other package managers doing that, but I can have a look around and report here what found 🙂 |
|
Rather than just resurrect the previous package, I think better would be to create a
This does happen often enough. Most package managers I've seen use an unversioned name for the current package and a versioned name for the previous version when the break occurred. For example, for package I can very quickly pull one together that builds the same as the new meson one with the same variants and options. I'll push it up shortly. |
|
#11553, I believe is likely a better solution since it uses the same variants and configurations as the new meson build, just with the autools configuration. |
|
I really like @alalazo's builtin-legacy suggestion! |
See discussion in #11537, #11542 -- migrating the
mesapackage to the meson build system broke many packages requiring python2.x. This PR introduces a new packagemesa-legacywhich is a snapshot ofmesafrom commit a698ac9, directly before the migration to meson.