proc macro server dies too easily and loses panic messages

[lnicola]

I was playing with the example in #6868 (comment) and it seems really easy to make it crash:

[jonas-schievink]

I've noticed this too, the loss of backtraces is due to the macro-side proc_macro crate turning them off intentionally. Not sure if we can do much about that.

I don't know why the double panic happens, but that would be great to fix as it tears down the whole process.

[lnicola]

Another example: the server is crashing on the rust-analyzer repository.

[edwin0cheng]

I just checked that the code mentioned do not crash now (052e722) ?!

[lnicola]

I'm not sure how exactly I've made it crash -- are you getting no errors in the RA output channel?

It should also crash on its own in three RA repo.

[edwin0cheng]

I'm not sure how exactly I've made it crash -- are you getting no errors in the RA output channel?

Yes, no errors.

[lnicola]

I'm using a nightly Rust, maybe that's why?

2020-12-20.08-36-22.mp4

(Firefox doesn't play that video, try to download it or use Chrome)