fix(crates): Resume workspace publish

[jan-auer]

The crates target is not able to resume a workspace publish because already published crates cannot be republished. This means a partially successful publish cannot be retried.

To work around this, the crates target can skip publishing a crate if it already exists in this version. The downside of this is that we would silence errors when publishing the wrong version.

See getsentry/publish#1070 (comment)

[BYK]

I'd argue a better way to handle this is to have multiple crates targets per package.

[BYK]

Or at least not make this behavior the default and hide behind a flag.

[jan-auer]

Having multiple crates targets is pretty much not an option here. In workspaces, you frequently have dozens of crates which you would then have to order topological by their dependency graph. That's too much manual work every time, unfortunately.

We could also add this as a flag, @chadwhitacre is there an easy way to pass a "this is a retry" flag to craft already?

[BYK]

Having multiple crates targets is pretty much not an option here. In workspaces, you frequently have dozens of crates which you would then have to order topological by their dependency graph. That's too much manual work every time, unfortunately.

I never said "manual" 😁 I think we can expand the publish state file format to have this extra metadata.

[jan-auer]

I see, yes that makes much more sense. If the crates target could remember the crates it already published, it could just pick up where it left off. This PR here is more meant as a quick fix, although I'm happy to wait a little longer for something proper.

[chadwhitacre]

I say we merge here and reticket @BYK's idea for running this through the state file.

[chadwhitacre]

Maybe with an additional comment linking to the new ticket.

[kamilogorek]

👀