Remove 'speculative evaluation' of predicates
Performing 'speculative evaluation' introduces caching bugs that
cannot be fixed without invasive changes to projection.
Hopefully, we can win back most of the performance lost by
re-adding 'cache completion'
Fixes#90662
rustdoc: make `--passes` and `--no-defaults` have no effect
Fixes#91714
One potential issue is that currently there is no stable way to achieve `--document-hidden-items`. This affects test `issue-15347`.
I also had to modify the tests `issue-42875` and `no-compiler-export`. Regardless of combinations of `--document-hidden-items` and `--document-private-items`, I was unable to get these to pass without the modifications. I left behind a comment noting the change.
Eliminate `ObligationCauseData`
This makes `Obligation` two words bigger, but avoids allocating a lot of the time.
I previously tried this in #73983 and it didn't help much, but local timings look more promising now.
r? `@ghost`
I think that s == "" is the only edge case (as it makes iter.next() return None the first time). The early return is necessary so that the last character of 'out' isn't popped if s == "" && !frag.need_backline
This makes `Obligation` two words bigger, but avoids allocating a lot of
the time.
I previously tried this in #73983 and it didn't help much, but local
timings look more promising now.
Rollup of 4 pull requests
Successful merges:
- #91791 (Fix an ICE when lowering a float with missing exponent magnitude)
- #91878 (Remove `in_band_lifetimes` from `rustc_infer`)
- #91895 (Remove `in_band_lifetimes` for `rustc_monomorphize`)
- #92029 (Explicitly set no ELF flags for .rustc section)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Explicitly set no ELF flags for .rustc section
For a data section, the object crate will set the SHF_ALLOC by default, which is exactly what we don't want. Explicitly set sh_flags to zero to avoid this.
I checked with `objdump -h` that this produces the right flags for ELF.
Fixes#92013.
Remove `in_band_lifetimes` from `rustc_infer`
See #91867 for more information.
This crate actually had a typo `'ctx` in one of its functions:
```diff
-pub fn same_type_modulo_infer(a: Ty<'tcx>, b: Ty<'ctx>) -> bool {
+pub fn same_type_modulo_infer<'tcx>(a: Ty<'tcx>, b: Ty<'tcx>) -> bool {
```
Also, I wasn't entirely sure about the lifetimes in `suggest_new_region_bound`:
```diff
pub fn suggest_new_region_bound(
- tcx: TyCtxt<'tcx>,
+ tcx: TyCtxt<'_>,
err: &mut DiagnosticBuilder<'_>,
fn_returns: Vec<&rustc_hir::Ty<'_>>,
```
Should all of those lifetimes really be distinct?
rustdoc: Remove unused `collapsed` field
`render/context` always runs after `run_global_context`, so it was always set to `true`.
This is a holdover from when rustdoc allowed configuring passes, but the `collapse-docs` pass was
removed ages ago, and the ability to configure passes is about to be removed.
Found while reviewing https://github.com/rust-lang/rust/pull/91305.
Update example code for Vec::splice to change the length
The current example for `Vec::splice` illustrates the replacement of a section of length 2 with a new section of length 2. This isn't a particularly interesting case for splice, and makes it look a bit like a shorthand for the kind of manipulations that could be done with a mutable slice.
In order to provide a stronger example, this updates the example to use different lengths for the source and destination regions, and uses a slice from the middle of the vector to illustrate that this does not necessarily have to be at the beginning or the end.
Resolves#92067
Enable `#[thread_local]` for all windows-msvc targets
As it stands, `#[thread_local]` is enabled haphazardly for msvc. It seems all 64-bit targets have it enabled, but not 32-bit targets unless they're also UWP targets (perhaps because UWP was added more recently?). So this PR simply enables it for 32-bit targets as well. I can't think of a reason not to and I've confirmed by running tests locally which pass.
See also #91659
Remove `in_band_lifetimes` from `rustc_middle`
See #91867
This was mostly straightforward. In several places, I take advantage
of the fact that lifetimes are non-hygenic: a macro declares the
'tcx' lifetime, which is then used in types passed in as macro
arguments.
Revert "Temporarily rename int_roundings functions to avoid conflicts"
This reverts commit 3ece63b64e.
This should be okay because #90329 has been merged.
r? `@joshtriplett`
Remove `SymbolStr`
This was originally proposed in https://github.com/rust-lang/rust/pull/74554#discussion_r466203544. As well as removing the icky `SymbolStr` type, it allows the removal of a lot of `&` and `*` occurrences.
Best reviewed one commit at a time.
r? `@oli-obk`
Remove `SymbolStr`
This was originally proposed in https://github.com/rust-lang/rust/pull/74554#discussion_r466203544. As well as removing the icky `SymbolStr` type, it allows the removal of a lot of `&` and `*` occurrences.
Best reviewed one commit at a time.
r? `@oli-obk`
rustdoc: Add `UrlPartsBuilder`
This is a type for efficiently and easily constructing the part of a URL
after the domain: `nightly/core/str/struct.Bytes.html`.
It allows simplifying some code and avoiding some allocations in the
`href_*` functions.
It will also allow making `Cache.paths` et al. use `Symbol` without
having to allocate `String`s in the `href_*` functions. `String`s would
be necessary otherwise because `Symbol::as_str()` returns `SymbolStr`,
whose `Deref<Target = str>` impl requires the `str` to not outlive it.
This is the primary motivation for the addition of `UrlPartsBuilder`.
Re-introduce concept of projection cache 'completion'
Instead of clearing out the cache entirely, we store
the intermediate evaluation result into the cache entry.
This accomplishes several things:
* We avoid the performance hit associated with re-evaluating
the sub-obligations
* We avoid causing issues with incremental compilation, since
the final evaluation result is always the same
* We avoid affecting other uses of the same `InferCtxt` which
might care about 'side effects' from processing the sub-obligations
(e,g. region constraints). Only code that is specifically aware
of the new 'complete' code is affected
Instead of clearing out the cache entirely, we store
the intermediate evaluation result into the cache entry.
This accomplishes several things:
* We avoid the performance hit associated with re-evaluating
the sub-obligations
* We avoid causing issues with incremental compilation, since
the final evaluation result is always the same
* We avoid affecting other uses of the same `InferCtxt` which
might care about 'side effects' from processing the sub-obligations
(e,g. region constraints). Only code that is specifically aware
of the new 'complete' code is affected
rustdoc: Write doc-comments directly instead of using FromIterator
The FromIterator impl made the code much harder to understand. The types
don't make sense until you realize there's a custom FromIterator impl.
This is the first commit from https://github.com/rust-lang/rust/pull/91305; since ``@camelid`` wrote it originally I don't feel bad unilaterally approving it.
r? ``@ghost``
``@bors`` r+
Note that this will conflict with https://github.com/rust-lang/rust/pull/92078.
Add user seed to `-Z randomize-layout`
Allows users of -`Z randomize-layout` to provide `-Z layout-seed=<seed>` in order to further randomizing type layout randomization. Extension of [compiler-team/#457](https://github.com/rust-lang/compiler-team/issues/457), allows users to change struct layouts without changing code and hoping that item path hashes change, aiding in detecting layout errors
Update browser-ui-test version and improve rustdoc-gui tests readability
Since the `0.5.1`, we can use trailing commas. I also used the opportunity to clean up the existing tests.
r? `@notriddle`
Avoid sorting in hash map stable hashing
Suggested by `@the8472` [here](https://github.com/rust-lang/rust/pull/89404#issuecomment-991813333). I hope that I understood it right, I replaced the sort with modular multiplication, which should be commutative.
Can I ask for a perf. run? However, locally it didn't help at all. Creating the `StableHasher` all over again is probably slowing it down quite a lot. And using `FxHasher` is not straightforward, because the keys and values only implement `HashStable` (and probably they shouldn't be just hashed via `Hash` anyway for it to actually be stable).
Maybe the `StableHash` interface could be changed somehow to better suppor these scenarios where the hasher is short-lived. Or the `StableHasher` implementation could have variants with e.g. a shorter buffer for these scenarios.
Rollup of 7 pull requests
Successful merges:
- #91566 (Apply path remapping to DW_AT_GNU_dwo_name when producing split DWARF)
- #91926 (Remove `in_band_lifetimes` from `rustc_metadata`)
- #91931 (Remove `in_band_lifetimes` from `rustc_codegen_llvm`)
- #92024 (rustc_codegen_llvm: Give each codegen unit a unique DWARF name on all platforms, not just Apple ones.)
- #92037 (Use a const ParamEnv when in default_method_body_is_const)
- #92047 (Set `RUST_BACKTRACE=0` when running location-detail tests)
- #92050 (Add a space and 2 grave accents )
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
For a data section, the object crate will set the SHF_ALLOC by
default, which is exactly what we don't want. Explicitly set
sh_flags to zero to avoid this.
`render/context` always runs after `run_global_context`, so it was always set to `true`.
This is a holdover from when rustdoc allowed configuring passes, but the `collapse-docs` pass was
removed ages ago, and the ability to configure passes is about to be removed.
