Work around async_trait behavior

[dario23]

The #[async_trait] attribute/crate does a transformation to all async
methods, which as far as i can tell removes the &self lifetime, so our
logic to offset by 1 if has_self == true for both compared types has
an off-by-one-error here. This didn't error out earlier, since
get_region_from_params uses Vec::get, so "this index is out of
bounds" is just as None as "this generic param is not of kind
lifetime".

Also this is more a workaround than a fix. I'm not sure if we can do
something cleverer than "check if the last lifetim's name is
'async_trait".

[JohnTitor]

Somehow could we have a regression test?

[dario23]

hm, which direction of regression? i looked into adding an example testcase using #[async_trait], but async fn in general needs edition 2018, while the other tests are built without any edition flag in the rustc invocation, so they assume edition 2015, and would need changes for building with edition 2018. i think we might want that anyway, WDYT?

in the other direction i'm not so sure, i.e. how would we test that we're not creating regressions with this for other code, without knowing properly what the other code now regressing looks like; we don't know what we haven't thought of, i guess.

[dario23]

ah, and also we'd need the async-trait crate when building old.rs and new.rs in that case.. which makes the rustc invocation quite a bit more involved (need to build async-trait crate first, including dependencies syn, quote and proc-macro2, and pass it to the other rustc invocations). i think it might be a lot easier to do a full case using casbin:2.0.0 -> casbin:2.0.1 for example.

[JohnTitor]

So, you expect this PR to fix #271, right?
I mean "a test to check if the fix doesn't regress", we could add a crate-based test here, and don't have to check old/new outputs, just have to check it doesn't trigger an ICE.

[dario23]

this should work after #291 is merged.

[dario23]

update: i keep getting different results for casbin running the tests locally, some are just order of errors, but there's also two errors that don't come up every run :-/

• casbin-2.0.0-2.0.1.linux.1.txt
• casbin-2.0.0-2.0.1.linux.2.txt
• casbin-2.0.0-2.0.1.linux.3.txt

[dario23]

alright, found one that (at least locally) doesn't show the nondeterministic behavior.

[JohnTitor]

Thanks! @bors r+

[bors]

📌 Commit 75875d4 has been approved by JohnTitor

[bors]

⌛ Testing commit 75875d4 with merge 71d2bb1...

[bors]

☀️ Test successful - checks-actions
Approved by: JohnTitor
Pushing 71d2bb1 to master...