debuginfo: Add temporary workaround for #14871. #14909
Conversation
|
@michaelwoerister thanks, I was literally just about to do this (but with a far less informative comment...) |
|
@pnkfelix Thanks for taking a look at this. Mind that I had to force-update this just now, as there was an unused variable warning in the first version |
|
Well, unfortunately it seems that the inconsistency is not so harmless after all. I'm running into LLVM assertions without the check. I guess, we're stuck with these crashes until I have time for a proper fix. Sorry for the inconvenience. However, I should be able to switch around my schedule for next week and take care of this tomorrow... |
|
Ah I had noticed that privately last night when giving my own version of this a shot, and had assumed (because I wasn't thinking) that it was an orthogonal bug. In hindsight that was silly of me. Thanks for being thorough. |
Run CI on nightly rust when proc-macro-* crates change
Whenever something is added to the debuginfo::TypeMap there's a consistency check, making sure that the code does what it is supposed to do. For some types (e.g. slices in a recursion cycle) this check rightfully fails. However, there's no need to actually abort compilation as the inconsistency won't lead to corrupt state, due to LLVM metadata uniquing.
A later, more thorough fix (coming next week) should remove the inconsistency altogether and also make this check non-fatal. For the time being, omit the check so people can compile their code with -g.