autopublish: Increment minor version instead of patch version

[davidlattimore]

The workflow is currently failing because it's trying to publish 0.0.16, while the last version published was 0.0.149.

[davidlattimore]

Happy to add an offset to github.run_number if you prefer

[lnicola]

Somewhere else we're publishing only on changes to the version number, that might be another option.

[davidlattimore]

Somewhere else we're publishing only on changes to the version number, that might be another option.

That's true, although assuming I'm understanding correctly, that would mean a PR each time we wanted to release to crates.io.

[lnicola]

It would, but we don't usually release without changes, and we can ask contributors to bump the version.

The bigger problem is that xaction is looking at the tags instead of querying crates.io.

[davidlattimore]

If someone made a change to a lower-level crate, such as ra_ap_stdx and they're thus bumping its version, would they then need to bump the version of ra_ap_stdx used by all the crates that depend on it? It seems like that might require a bit of work on the part of people making a change. It'd also perhaps result in more merge conflicts. I'd gotten the impression that publishing rust-analyzer to crates.io was somewhat of a secondary concern. Just autopublishing everything once per week has the nice outcome that nobody really needs to think about it (except when it breaks).

[davidlattimore]

I also posted #14827 to just apply an offset, in case that's preferable to using setting the minor version number.

[Veykril]

So if i see this right the cause was us renaming the workflow right? In that case I'd say we should just go with #14827 for now, that way we are back to the old behevior if I understand this correctly

[davidlattimore]

So if i see this right the cause was us renaming the workflow right? In that case I'd say we should just go with #14827 for now, that way we are back to the old behevior if I understand this correctly

That's my understanding. It wasn't obvious initially because autopublishing was broken for a different reason - specifically the switch to use workspace.dependencies for local dependencies (bed4db3) broke publishing because cargo-workspaces didn't support automatic renaming of workspace dependencies. I fixed that in cargo-workspaces and that's when this problem showed up instead.

I'll close this PR is favour of #14827.