Cargo workspace2

[xclaesse]

No description provided.

[xclaesse]

@bonzini This is my very, very WIP for workspace support. The code is a total mess right now, but it finally passed configuration of the whole gst-plugins-rs tree! Over 400 subprojects generated.... https://paste.centos.org/view/6920dc55

[bonzini]

Thanks, I can try to integrate workspace parsing in my PR (just based on unit tests) and the interpreter code should be the same.

[xclaesse]

It doesnt't have unit tests yet, my test case is simple:

project('rs')
subproject('gst-plugins-rs')

[wrap-git]
url=https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs.git
revision=main
method=cargo

[bonzini]

I can add some ;)

[bonzini]

Ok, I added workspace inheritance to #14672 and this diff is the rebase of your wip patch on top of it.

[xclaesse]

@bonzini I started to clean this PR commits and integrate some of your ideas I liked. I'm not yet totally satisfied with the result, but it's starting to look good IMHO. Now all dataclasses are created using _raw_to_dataclass() that can validate/convert attributes from raw to dataclasses.

[bonzini]

Ok I will also do the same and apply some of your commits, so that the WIP parts are up to date and conflicts are not a problem. Hopefully we start getting some review soon. :)

[bonzini]

Updated my branch(es). Among other things I renamed the conversion function to _raw_to_dataclass, switched the computed Meson version to lazy property.

I also added release notes for "cargo: Use -rs suffix only for rust ABI".

I would also like to reinforce that without unit tests for the dataclass conversion, this is not going anywhere. Some bugs that you have in your tree:

• changing the default feature to use default_factory is wrong, because you're dropping the feature if any other feature is defined

• as mentioned earlier, the version of the Cargo.lock file is an integer and not a string

Without tests it's also unclear what changes will need release notes. For example I suspect that "cargo: When loading wraps multiple packages can have the same URL" does require them, but I am not sure.

I really would prefer it if you switched to https://github.com/bonzini/meson/commits/cargo-workspace2-for-xclaesse, since it also takes into account more of the review comments that you got for #12983. In any case I am very interested in moving this work forward, so I'll keep rebasing my branches and making them follow what you do as closely as possible.

[xclaesse]

I would also like to reinforce that without unit tests for the dataclass conversion, this is not going anywhere.

We do have tests cases already, but I agree not enough.

• changing the default feature to use default_factory is wrong, because you're dropping the feature if any other feature is defined

Thanks, reverted that change.

• as mentioned earlier, the version of the Cargo.lock file is an integer and not a string

I missed that you mentioned that before, fixed.

cargo: When loading wraps multiple packages can have the same URL

AFAIK, that's specific to workspace. Without workspace 1 subproject is 1 crate. For release notes, that's part of adding support for workspace.

I really would prefer it if you switched to https://github.com/bonzini/meson/commits/cargo-workspace2-for-xclaesse

Could you rebase that branch on top of mine so I can more easily see what you're adding? I'm still very strongly against bringing back TypedDict.

[bonzini]

Could you rebase that branch on top of mine so I can more easily see what you're adding? I'm still very strongly against bringing back TypedDict.

I am adding nothing except rebasing on top of #14672 (which I am keeping green). I understand you don't want to bring back TypedDict but you already got pushback from 2.5 people (I am the 0.5).