Make CrateMetadata and CStore thread-safe #48936
Conversation
rust-highfive
added
the
S-waiting-on-review
Zoxc
force-pushed
the
cstore
branch
3 times, most recently
from
0d9b6ee to
2de21be
Compare
There was a problem hiding this comment.
Thanks, @Zoxc! Looks good to me.
What do you think of adding an InitOnce<T> primitive to rustc_data_structures::sync? Especially things like codemap_import_info would profit from this, I think. I haven't done any performance testing but even if it's not significantly faster, it makes the semantics much clearer which is always a win.
| @@ -335,7 +334,9 @@ impl<'a> CrateLoader<'a> { | |||
| if data.root.macro_derive_registrar.is_some() { | |||
| dep_kind = DepKind::UnexportedMacrosOnly; | |||
| } | |||
| data.dep_kind.set(cmp::max(data.dep_kind.get(), dep_kind)); | |||
| data.dep_kind.with_lock(|data_dep_kind| { | |||
There was a problem hiding this comment.
We should probably add a separate primitive for state that is initialized once and then never modified again. There must be more efficient ways to do this than a mutex.
There was a problem hiding this comment.
parking_lot::Once looks like a good building block for such a thing.
|
I did think about adding something like |
| metadata_loader, | ||
| } | ||
| } | ||
|
|
||
| /// You cannot use this function to allocate a CrateNum in a thread-safe manner. | ||
| /// It is currently only used in CrateLoader which is single-threaded code. |
There was a problem hiding this comment.
I added a comment about this function. next_crate_num is used once in CrateLoader to get the next free CrateNum. CrateLoader may then create multiple CrateNums based on that.
At least for |
|
📌 Commit 5b8f9c5 has been approved by |
bors
added
S-waiting-on-bors
Make CrateMetadata and CStore thread-safe r? @michaelwoerister
|
☀️ Test successful - status-appveyor, status-travis |
r? @michaelwoerister