[MIR] A pass to type-check MIR #31474
Conversation
|
r? @jroesch (rust_highfive has picked a reviewer for you, use r? to override) |
| @@ -974,6 +978,7 @@ pub fn build_session_options(matches: &getopts::Matches) -> Options { | |||
| let parse_only = debugging_opts.parse_only; | |||
| let no_trans = debugging_opts.no_trans; | |||
| let treat_err_as_bug = debugging_opts.treat_err_as_bug; | |||
| let mir_opt_level = debugging_opts.mir_opt_level.unwrap_or(1); | |||
There was a problem hiding this comment.
I was considering a default of 1 + -C opt-level. This way the level of MIR “optimisations” would automatically get attached to the -O and -C opt-level flags as well.
There was a problem hiding this comment.
First make opt-levels greater than 1 do something, then we can talk.
|
☔ The latest upstream changes (presumably #31425) made this pull request unmergeable. Please resolve the merge conflicts. |
|
How fast do you expect the check to be? It would be nice if we could run the check between each pass, probably based on a -Z flag. |
|
cc me |
|
This pass should not take too much time. |
|
Let me know when you think this is ready for review, reading over it right now. |
arielb1
changed the title
|
the rvalue-type-checking code can be improved, but I will leave that to a separate PR - this one is growing too big. |
arielb1
force-pushed
the
mir-typeck
branch
3 times, most recently
from
59b1dd5
to
d210962
Compare
| span_mirbug_and_err!(self, parent, "bad type {:?}", ty) | ||
| } | ||
|
|
||
| fn sanitize_lvalue(&mut self, lvalue: &Lvalue<'tcx>) -> LvalueTy<'tcx> { |
There was a problem hiding this comment.
This prefix sanitize strikes me as odd. To me, it suggests something which will mutate the lvalue in place, removing artifacts or normalizing it in some way. But in fact sanitize_lvalue seems to compute the type of an lvalue.
There is also an unfortunate amount of overlap with the lvalue_ty routine -- these basically do the same thing, but just handle the error cases differently. Perhaps we could make lvalue_ty return a Result instead to eliminate the overlap?
There was a problem hiding this comment.
OK, reading a bit further I see that sanitize_projection actually does more consistency checking, so perhaps it is good to keep these two routines as separate. I think I would change sanitize to typeck_ or type_ or just check_.
There was a problem hiding this comment.
Perhaps verify_ would be the best name, given that this is part of the type verifier.
There was a problem hiding this comment.
it returns the sanitized lvalue ty, which will be TyErr if there are are late-bound regions etc. in the type.
|
☔ The latest upstream changes (presumably #31600) made this pull request unmergeable. Please resolve the merge conflicts. |
setting -Z mir-opt-level=0 will disable all MIR optimizations for easier debugging
this makes the the MIR assignment pass complete successfully
TODO: find a correct borrow region Fixes rust-lang#31482
also fix translation of "rust-call" functions, although that could use more optimizations
rather than being implicit quasi-subtyping. Nothing good can come out of quasi-subtyping.
|
@bors r=nikomatsakis |
|
📌 Commit 626dc3b has been approved by |
|
⌛ Testing commit 626dc3b with merge e768394... |
|
💔 Test failed - auto-mac-64-opt |
|
@bors r=nikomatsakis |
|
📌 Commit d84658e has been approved by |
|
⌛ Testing commit d84658e with merge 6c751e0... |
This should stop broken MIR from annoying us when we try to implement things
This should stop broken MIR from annoying us when we try to implement things