Add -Z no-parallel-llvm flag
Codegen issues commonly only manifest under specific circumstances,
e.g. if multiple codegen units are used and ThinLTO is enabled.
However, these configuration are threaded, making the use of LLVM
debugging facilities hard, as output is interleaved.
This patch adds a -Z no-parallel-llvm flag, which allows disabling
parallelization of codegen and linking, while otherwise preserving
behavior with regard to codegen units and LTO.
Make sure that queries have predictable symbol names.
Some recent refactorings led to query names not showing up in the corresponding symbol names. [perf-focus](https://github.com/nikomatsakis/perf-focus) and manual profiling have been broken by this. This PR makes sure that query providers always get their own symbol and that that symbol has a predictable name.
Since this adds `#[inline(never)]` to a function that wraps the provider call, let's check if this does not regress performance before merging.
r? @nikomatsakis
rust-gdb: work around the re-used -d argument in cgdb
Use --directory= to pass $GDB_PYTHON_MODULE_DIRECTORY instead of -d, because [cgdb](https://github.com/cgdb/cgdb) reuses '-d' to select a debugger, whereas the long option works with both gdb and cgdb.
rustc: Correctly pretty-print macro delimiters
This commit updates the `Mac_` AST structure to keep track of the delimiters
that it originally had for its invocation. This allows us to faithfully
pretty-print macro invocations not using parentheses (e.g. `vec![...]`). This in
turn helps procedural macros due to #43081.
Closes#50840
"crate-ify" paths that begin with a renamed crate
This does two things:
- crate-ify paths that begin with a renamed crate (i.e., add `crate::`) to the front
Fixes https://github.com/rust-lang/rust/issues/50996
I also added tests for a few other scenarios.
r? @alexcrichton
Use different datastructure for MIRI relocations
This PR makes relocations in MIRI used a sorted vector instead of a `BTreeMap` which should make a few common operations more efficient. Let's see if that's true.
r? @oli-obk
Remove attribute_cache from CrateMetadata
This PR will fix#50508 by removing the `attribute_cache` from the `CrateMetadata` struct. Seeing as performance was referenced in the original issue, I also cleaned up a `self.entry(node_id);` call which might have occasionally happened redundantly.
r? @michaelwoerister
Generate "invalidates" facts when -Znll-facts is passed
Most of the new code is copied directly from the heart of the MIR borrowchecker. I was expecting more fundamental structural changes, hence the copying. This appears to work as it stands, but I'd like to submit a follow-up PR to reduce code duplication. I figured that could wait though, since this is blocking a large amount of work in the borrow check repository and I'm out of time for tonight =).
r? @nikomatsakis
rustc: Fix another double-lint issue with `crate::`
This commit fixes another issue in the `absolute_path_not_starting_with_crate`
lint where it warns twice about an import which may contain `self`. It turns out
there were a few more locations that needed updating to use `root_id` and
`root_span` introduced in #50970 and after that it looks to work like a charm!
Closes#50978
This commit fixes another issue in the `absolute_path_not_starting_with_crate`
lint where it warns twice about an import which may contain `self`. It turns out
there were a few more locations that needed updating to use `root_id` and
`root_span` introduced in #50970 and after that it looks to work like a charm!
Closes#50978
`Liveness::users` is a vector that is occasionally enormous. For
example, doing a "clean incremental" check build of `inflate`, there is
one instance that represents 5,499 live nodes and 1087 vars, which
requires 5,977,413 entries. At 24 bytes per entry, that is 143MB.
This patch changes LiveNode from a usize to a u32. On 64-bit machines
that halves the size of these entries, significantly reducing peak
memory usage and memory traffic, and speeding up "clean incremental"
builds of `inflate` by about 10%.
This commit updates the `Mac_` AST structure to keep track of the delimiters
that it originally had for its invocation. This allows us to faithfully
pretty-print macro invocations not using parentheses (e.g. `vec![...]`). This in
turn helps procedural macros due to #43081.
Closes#50840
Codegen issues commonly only manifest under specific circumstances,
e.g. if multiple codegen units are used and ThinLTO is enabled.
However, these configuration are threaded, making the use of LLVM
debugging facilities hard, as output is interleaved.
This patch adds a -Z no-parallel-llvm flag, which allows disabling
parallelization of codegen and linking, while otherwise preserving
behavior with regard to codegen units and LTO.
Right-size the `VecDeque` in `coerce_unsized`.
The default capacity of a VecDeque is 8, which is excessive here. In a
"base incremental" check build of rustc-perf's tuple-stress benchmark,
this decreases total heap allocation by 26%. I couldn't see a clear
speedup, but it can't hurt.
Micro-optimization on PR#50697
We should stop iterating through the indices in the `init_path_map` once we've already found a match for the local.
r? @nikomatsakis or @pnkfelix
Add the 2018 edition of the book to doc.rust-lang.org
The second edition of the book is on its way to the printers, and as such, is frozen. We've forked off the 2018 edition to add new stuff to; this PR now builds it so that people can read it on doc.rust-lang.org.
