Remove the requirement that ast->hir lowering be reproducible #33296
Conversation
jseyfried
force-pushed
the
non_idempotent_lowering
branch
2 times, most recently
from
b1b28c5
to
ce9e894
Compare
|
What’s the purpose of the change? Seems like a potentially plugin-breaking-change to me. |
|
Side note: if |
The purpose is to simplify lowering (by removing the id caching system) and to remove the lowering context from the Also, it will be significantly simpler to do name resolution during lowering if lowering doesn't need to be reproducible. I don't think it's worth the extra complexity needed to be reproducible when we can just use the HIR map to re-lower (almost) any AST node. |
|
@petrochenkov good point, I added a commit that clears the MTWT tables after lowering for save-analysis. |
jseyfried
force-pushed
the
non_idempotent_lowering
branch
from
0cc3573
to
864b4f8
Compare
jseyfried
changed the title
| @@ -2121,21 +2121,8 @@ fn signal_block_expr(lctx: &LoweringContext, | |||
|
|
|||
| #[cfg(test)] | |||
| mod test { | |||
There was a problem hiding this comment.
I think you can remove this whole module
jseyfried
force-pushed
the
non_idempotent_lowering
branch
from
7a000ee
to
522b6ed
Compare
| // This whole system relies on node ids being incremented one at a time and | ||
| // all increments being for lowering. This means that you should not call any | ||
| // non-lowering function which will use new node ids. | ||
| // | ||
| // We must also cache gensym'ed Idents to ensure that we get the same Ident |
There was a problem hiding this comment.
Good point, forgot about that -- done.
|
r+ with that comment removed |
|
@bors: r+ |
|
📌 Commit ef69ef8 has been approved by |
Remove the requirement that ast->hir lowering be reproducible This PR changes the ast->hir lowerer to be non-reproducible, and it removes the lowering context's id cache. If the `hir` of an `ast` node needs to be reproduced, we can use the hir map instead of the lowerer -- for example, `tcx.map.expect_expr(expr.id)` instead of `lower_expr(lcx, expr)`. r? @nrc
|
What exactly is getting non-reproducible with this change? Does this mean that if I compile the same Rust program twice, I would get two different binaries (as in, binaries that are not bit-for-bit identical)? That would be a huge bummer -- reproducible builds are a cornerstone of building trustworthy binaries. If Rust can't build reproducibly, This is actively used, for example, by the Tor project and some BitCoin clients. it would be unsuited for any application that needs to establish trust in the generated binaries. See https://reproducible-builds.org/ for more details. |
I don't know if Rust has this property, but this PR isn't related to reproducible builds. Lowering is still deterministic after this PR (i.e. it behaves the same when building the same code). This PR changed lowering so that if the same AST node is lowered multiple times during the same build, we aren't guaranteed to get same result. Perhaps a better word would be "idempotent" instead of "reproducible". |
…nique, r=nrc Give `ast::ExprKind::Paren` no-op expressions the same ids as their children. Having `ast::ExprKind::Paren` expressions share ids with their children - reduces the number of unused `NodeId`s in the hir map and - guarantees that `tcx.map.expect_expr(ast_expr.id)` is the hir corresponding to `ast_expr`. This fixes the bug from rust-lang#34327, which was introduced in rust-lang#33296 when I assumed the above guarantee. r? @nrc
This PR changes the ast->hir lowerer to be non-reproducible, and it removes the lowering context's id cache.
If the
hirof anastnode needs to be reproduced, we can use the hir map instead of the lowerer -- for example,tcx.map.expect_expr(expr.id)instead oflower_expr(lcx, expr).r? @nrc