internal: Make HirFileId, EditionedFileId and macro files Salsa struct
#19617
Conversation
rustbot
added
the
S-waiting-on-review
ChayimFriedman2
force-pushed
the
more-actual
branch
4 times, most recently
from
c75f6d8 to
018402c
Compare
| #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, salsa::Supertype)] | ||
| pub enum HirFileId { | ||
| FileId(EditionedFileId), | ||
| MacroFile(MacroCallId), | ||
| } |
There was a problem hiding this comment.
Note that it would be nicei f we could shrink this to 4 bytes again (as was done with the bit packing before). Iirc back when I implemented that it saved ~10mb. Not relevant to this PR for now though.
There was a problem hiding this comment.
We can do this if we store it in places as salsa::Id (or some newtype wrapper). Methods can still take HirFileId.
There was a problem hiding this comment.
Yes, I believe Niko's initial supertype proposal wouldvebeen something like that, where instead of it being a derive it was an attribute that collapsed the variants into an erased wrapper. I wonder if we could implement that as a separate thing to the derive as well. I do like the derive in most cases, given it doesn't require you to extract the active variant manually.
ChayimFriedman2
force-pushed
the
more-actual
branch
from
018402c to
593aa72
Compare
And make more queries non-interned. Also flip the default for queries, now the default is to not intern and to intern a query you need to say `invoke_interned`.
ChayimFriedman2
force-pushed
the
more-actual
branch
from
593aa72 to
c58ddaf
Compare
And make more queries non-interned.
Also flip the default for queries, now the default is to not intern and to intern a query you need to say
invoke_interned.This saves 161mb on
analysis-stats ..This is a big PR, but most of it is tedious replacements: the interesting changes are in
span,base-dbandhir-expand/src/lib.rs. Oh, andquery-group-macro. The other files could use a skim review.