Host resources wit-bindgen code generation #6886
Conversation
There was a problem hiding this comment.
Thanks again for this! I'm running out of time today but wanted to give some quick feedback on this. There's a few comments below and I have more thoughts on the generated code but I'll try to flesh that out more tomorrow (I'll probably send a PR-to-your-PR in the morning)
In the meantime though this'll need to have some tests as well, both at crates/component-macro/tests/codegen/*.wit and additionally one in tests/all/component_model/*. You'll find some other bindgen-related tests in that directory too.
github-actions
bot
added
the
wasmtime:api
Subscribe to Label Actioncc @peterhuene
This issue or pull request has been labeled: "wasmtime:api"
Thus the following users have been cc'd because of the following labels:
To subscribe or unsubscribe from this label, edit the |
silesmo
force-pushed
the
resources-basic
branch
from
4532eac
to
17f0a87
Compare
Avoid collecting resources-as-we-go in favor of doing that more declaratively elsewhere. Additionally handle imported-vs-exported resource via the order that interfaces are visited.
* Use `&mut self` instead of `StoreContextMut` * Make resource traits as supertraits of the `Host` trait generated for each resource. * More uniformly handle types/returns with resource methods. * Fix derivations of `Clone` and `Copy` for where handles are contained.
|
Ok I've worked on this a bit this morning and I've pushed up Nor2-io#1 which contains another round of comments for me (and some fixes for the deny/vet-related things I think). As I mentioned over there I'm happy to go through anything if you'd like clarifications or explanations! |
Updates to host resources and `bindgen!` resources in exported functions arguments break with this commit and will be fixed in following commits.
This is now the same as the preexisting `generate_function_trait_sig`
This determination happens by looking up the origin definition of a resource, not the leaf possibly-aliased type.
Resources don't have their representation asserted since the bare type itself doesn't implement `ComponentType`, but otherwise generate type aliases the same way as other type aliases.
No need to update `wit-bindgen` to 0.10.0 just yet, that'll happen in a future update if necessary.
|
One more round of changes at Nor2-io#2 (again happy to go through anything if you'd like), and then I think this is good to go 👍 |
More resources-related changes
|
Also @silesmo would you be up for filing an issue for follow-up work? |
Yes I can do that! 🙂 |
|
Pat's busy with streams and such so he'll take a look at this after-the-fact, so I'm going to go ahead and flag this for merge. Thanks again @silesmo! |
|
oh I think this may need a rebase for merging now? |
Head branch was pushed to by a user without write access
I merged the latest changes now |
* added trap to resource dtor * basic resources * fixed signature and trait bounds * fixed static function * fixed trait signature * basic resources * added result to resource drop * reverted formatting * added doc comment to resource `dtor` * fixed merge issue * fixed another merge issue * added resource import codegen test * Revert changes to `Cargo.lock` * Update `Cargo.lock` with the wit-bindgen update * Add `cargo vet` entries for new crates * Restore old-style of printing types in bindgen Avoid collecting resources-as-we-go in favor of doing that more declaratively elsewhere. Additionally handle imported-vs-exported resource via the order that interfaces are visited. * Update the shape of resource traits: * Use `&mut self` instead of `StoreContextMut` * Make resource traits as supertraits of the `Host` trait generated for each resource. * More uniformly handle types/returns with resource methods. * Fix derivations of `Clone` and `Copy` for where handles are contained. * Fix generation of handle typedefs * Support resources-in-worlds * Remove now-duplicate function This is now the same as the preexisting `generate_function_trait_sig` * Fix classifying handles as imported or exported This determination happens by looking up the origin definition of a resource, not the leaf possibly-aliased type. * Fix chains-of-use of resources Resources don't have their representation asserted since the bare type itself doesn't implement `ComponentType`, but otherwise generate type aliases the same way as other type aliases. * Revert `Cargo.lock` changes No need to update `wit-bindgen` to 0.10.0 just yet, that'll happen in a future update if necessary. * Add basic runtime tests for resources * fixed merge issue --------- Co-authored-by: Alex Crichton <alex@alexcrichton.com>
Implements host export resource generation #6722
After a discussion with @alexcrichton, I have broken out basic resource generation for host resource exports into it's own PR,
Host import resource generation will follow, as well as more ergonomic resource code generation that abstracts away the raw resource handles.
Example implementation: https://github.com/Nor2-io/wasmtime-resources-example/blob/main/example-host-export-index/src/main.rs
Example code generation output: https://github.com/Nor2-io/wasmtime-resources-example/blob/main/example-host-export-index/main%20copy%202.rs
Sneak peak of future additions that abstract away resources can be seen in the link below. (It doesn't work with nested resources yet.)
https://github.com/Nor2-io/wasmtime-resources-example/blob/main/example-host-import/src/main.rs
https://github.com/Nor2-io/wasmtime-resources-example/blob/main/example-host-export/src/main.rs
Added temporary exception for cargo deny on multiple versions until memfd has been updated: #62