Per the discussion on #34765, we make one `DepNode::Mir` variant and use
it to represent both the MIR tracking map as well as passes that operate
on MIR. We also track loads of cached MIR (which naturally comes from
metadata).
Note that the "HAIR" pass adds a read of TypeckItemBody because it uses
a myriad of tables that are not individually tracked.
I cannot figure out how to write a test for this, but I observed
incorrect edges as a result of not using memoized pattern here
(e.g., LateLintCheck -> SizedConstraint).
When we hash the inputs to a MetaData node, we have to hash them in a
consistent order. We achieve this by sorting the stringfied `DefPath`
entries. Also, micro-optimie by cache more results across the saving
process.
The reads will occur naturally as the HIR/MIR is fetched from the
tracked tables, and this winds up adding reads to the hir of foreign
def-ids somehow.
The biggest problem, actually, is krate numbers being removed entirely,
which can lead to array-index-out-of-bounds errors.
cc #35123 -- not a complete fix, since really we ought to "map" the old
crate numbers to the new ones, not just detect changes.
The way we do HIR inlining introduces reads of the "Hir" into the graph,
but this Hir in fact belongs to other crates, so when we try to load
later, we ICE because the Hir nodes in question don't belond to the
crate (and we haven't done inlining yet). This pass rewrites those HIR
nodes to the metadata from which the inlined HIR was loaded.
rustc_trans: apply the debug location for the MIR Assert panic call.
Helps `libcore` build with MIR trans and debuginfo; libcore has the body of `panic`, which resulted in:
```
inlinable function call in a function with debug info must have a !dbg location
call void @_ZN4core9panicking5panic17h585bd70cda921012E({ %str_slice, %str_slice, i32 }* @panic_loc12745)
LLVM ERROR: Broken function found, compilation aborted!
```
tcp-stress-test: Pull out thread count as a constant
This PR factors out the number of concurrent threads used in `tcp-stress-test.rs` to a constant at the top of the file.
We at @NixOS had to lower our thread count as the chrooted-builds don't allow that many threads.
This change will make it easier to lower/increase the count in the future (I actually forgot to change the second `1000` when I was working on this). Another benefit is the removal of magic numbers in the test suite.
This is related to https://github.com/rust-lang/rust/issues/35107
Move caching of inlined HIR into CrateStore
So far we've had separate HIR-inlining caches for each codegen unit and the caching for things inlined during constant evaluation had some holes. Consequently, things would be inlined multiple times if they were used from different codegen units, etc, leading to
- wasted memory,
- multiple `NodeId`s per `DefId` and,
- for things inlined during constant evaluation, no way to map a `NodeId` back to it's original `DefId`.
This PR moves all caching into the CrateStore, solving all of the above problems. It also fixes some bugs in the inlining code, like cyclic in the parent-chains in the HIR map and some `NodeId`'s being translated to more or less random values. There are assertions in place now that should prevent this kind of thing in the future.
This PR based on top of #35090, which contains some necessary fixes.
Use "x86-64" as the target CPU for NetBSD and Bitrig on amd64.
Using "generic" disables a number of features that are present on all
x86_64 cpus, the "x86-64" target cpu is the common denominator for that
arch.
Refs #20777
Methods `Fn(Mut,Once)::call(mut,once)` are gated with two feature gates, remove one of them
Methods `Fn::call`, `FnMut::call_mut` and `FnOnce::call_once` are gated with usual library feature `fn_traits` and also hardcoded in the compiler and gated once more with feature `unboxed_closures`
This patch removes the `unboxed_closures`feature gate from these methods (`unboxed_closures` is still used for other things though), now they are gated only with `fn_traits`.
All unnecessary `#![feature(unboxed_closures)]`s are removed, many of them are old and were already unnecessary before the change this PR does.
LLVM upgrade
As discussed in https://internals.rust-lang.org/t/need-help-with-emscripten-port/3154/46 I'm trying to update the used LLVM checkout in Rust.
I basically took @shepmaster's code and applied it on top (though I did the commits manually, the [original commits have better descriptions](https://github.com/rust-lang/rust/compare/master...avr-rust:avr-support).
With these changes I was able to build rustc. `make check` throws one last error on `run-pass/issue-28950.rs`. Output: https://gist.github.com/badboy/bcdd3bbde260860b6159aa49070a9052
I took the metadata changes as is and they seem to work, though it now uses the module in another step. I'm not sure if this is the best and correct way.
Things to do:
* [x] ~~Make `run-pass/issue-28950.rs` pass~~ unrelated
* [x] Find out how the `PositionIndependentExecutable` setting is now used
* [x] Is the `llvm::legacy` still the right way to do these things?
cc @brson @alexcrichton
trans: Avoid weak linkage for closures when linking with MinGW.
This PR proposes one possible solution to #34793, the problem that prevents https://github.com/servo/servo/pull/12393 from landing. It applies the same strategy, that we already use for monomorphizations, to closures, that is, instead of emitting symbols with `weak_odr` linkage in order to avoid symbol conflicts, we emit them with `internal` linkage, with the side effect that we have to copy code instead of just linking to it, if more than one codegen unit is involved.
With this PR, the compiler will only apply this strategy for targets where we would actually run into a problem when using `weak_odr` linkage, in other words nothing will change for platforms except for MinGW.
The solution implemented here has one restriction that could be lifted with some more effort, but it does not seem to be worth the trouble since it will go away once we use only MIR-trans: If someone compiles code
1. on MinGW,
2. with more than one codegen unit,
3. *not* using MIR-trans,
4. and runs into a closure inlined from another crate
then the compiler will abort and suggest to compile either with just one codegen unit or `-Zorbit`.
What's nice about this is that I lays a foundation for also doing the same for generics: using weak linkage where possible and thus enabling some more space optimizations that the linker can do.
~~This PR also contains a test case for compiling a program that contains more than 2^15 closures. It's a huge, generated file with almost 100K LOCs. I did not commit the script for generating the file but could do so. Alternatively, maybe someone wants to come up with a way of doing this with macros.~~
The test file is implemented via macros now (thanks @alexcrichton!)
Opinions?
Fixes#34793.
cc @rust-lang/compiler
typeck: use a TypeVisitor in ctp
Use a TypeVisitor in ctp instead of `ty::walk`
This fixes a few cases where a region could be projected out of a trait while not being constrained by the type parameters, violating rust-lang/rfcs#447 and breaking soundness. As such, this is a [breaking-change].
Fixes#35139
r? @eddyb
Avoid writing a temporary closure kind
We used to write a temporary closure kind into the inference table, but
this could lead to obligations being incorrectled resolved before
inference had completed. This result could then be cached, leading to
further trouble. This patch avoids writing any closure kind until the
computation is complete.
Fixes#34349.
r? @arielb1 -- what do you think?
diagnostically note source of overruling outer forbid
When we emit E0453 (lint level attribute overruled by outer `forbid`
lint level), it could be helpful to note where the `forbid` level was
set, for the convenience of users who, e.g., believe that the correct
fix is to weaken the `forbid` to `deny`.
Suggest use of `--print target-list` when target is not found.
If given target could not be found suggest using `--print target-list`.
Previously, error has been reported as:
$ rustc --target 86-unknown-linux-gnu
error: Error loading target specification: Could not find specification for target "86-unknown-linux-gnu"
After changes it looks as follows:
$ rustc --target 86-unknown-linux-gnu
error: Error loading target specification: Could not find specification for target "x86-unknown-linux-gnu". Use `--print target-list` for a list of supported targets
If given target could not be found suggest using `--print target-list`.
Previously, error has been reported as:
$ rustc --target x86-unknown-linux-gnu
error: Error loading target specification: Could not find specification for target "x86-unknown-linux-gnu"
After changes it looks as follows:
rustc --target x86-unknown-linux-gnu
error: Error loading target specification: Could not find specification for target "x86-unknown-linux-gnu"
help: Use `--print target-list` for a list of built-in targets
Add ARM MUSL targets
Rebase of #33189.
I tested this by producing a std for `arm-unknown-linux-musleabi` then I cross compiled Hello world to said target. Checked that the produced binary was statically linked and verified that the binary worked under QEMU.
This depends on rust-lang/libc#341. I'll have to update this PR after that libc PR is merged.
I'm also working on generating ARM musl cross toolchain via crosstool-ng. Once I verified those work, I'll send a PR to rust-buildbot.
r? @alexcrichton
cc @timonvo
Fix build of compiler-rt on FreeBSD
Broken since ee6011fc71 removed cmake from the
process. There are likely other platforms still broken, but I didn't test on them.
Properly feature gate all unstable ABIs
Fixes https://github.com/rust-lang/rust/issues/34900
[breaking-change]
r? @pnkfelix
---
Function-visiting machinery for AST/HIR is surprisingly error-prone, it's *very* easy to miss some cases or visit something twice while writing a visitor. This is the true problem behind https://github.com/rust-lang/rust/issues/34900. I'll try to restructure these visitors a bit and send one more PR later.
