I'm working on updating the verification engine to use this Sema changes that save names from this PR. One hiccup is handling names for let
blocks. Since the bindings are truncated after evaluating the let
body (e.g., this line), can variable IDs be reused across multiple (non-nested) let
blocks?
Seems like we will need to do our own scoping on the verification side to handle this, which is fine, just wanted to confirm that this is what happens with the VarId
mapping. Thanks!
CC @Jamey Sharp
Uh. I didn't think about the implications of that truncate when I wrote that PR. I think we should just revert part of the PR and put the next_var
field back so that VarId
s are unique within a rule. It should be a very small patch since now we can just manage next_var
inside the add_var
function.
Okay, it's not quite so easy to fix as I hoped, since truncating also throws away the names I was trying to save for you (and, eventually, for me too!). I think BoundVar
should keep some notion of which scope the name was bound in, and Bindings::lookup
should respect that, rather than truncating bindings.seen
anywhere. I'm still thinking through the details...
Got it, I'll also add this to the agenda for our next meeting. Thanks!
I'm out of time for today and haven't figured out precisely how to fix this yet. I can say for sure though that I don't want you to have to walk the AST again to get the names out. We should be able to expose the right information from sema, I'm just having trouble thinking through the details.
Actually, after taking a break, suddenly it seemed much easier: https://github.com/bytecodealliance/wasmtime/pull/5538
Last updated: Dec 23 2024 at 12:05 UTC