This isn't sound without validation. We don't want to report errors in case of failure to intern and validate, we just don't want to const prop. Interning and const prop is not built for this, let's not do it until we have a clearer picture on aggregate propagation.
Update LLVM submodule to include lld NOLOAD fix
> Rust nightly 2020-05-22 and later ships lld with a regression related to linker scripts: NOLOAD sections incorrectly generate sections filled with 0s. This causes gdb and other elf loaders to write to reserved or otherwise invalid addresses (gdb also seems confused by the resulting ELF files and spits out a warning about the sections). This is particularly a problem for embedded rust projects that use lld by default and have affected linker scripts (cortex-m-rt based projects for instance).
https://github.com/rust-lang/llvm-project/pull/64
Note that this also pulls in llvm changes from #72937
Count the beta prerelease number just from master
We were computing a merge-base between the remote beta and master
branches, but this was giving incorrect answers for the first beta if
the remote hadn't been pushed yet. For instance, `1.45.0-beta.3359`
corresponds to the number of merges since the 1.44 beta, but we really
want just `.1` for the sole 1.45 beta promotion merge.
We don't really need to query the remote beta at all -- `master..HEAD`
suffices if we assume that we're on the intended beta branch already.
validate basic sanity for TerminatorKind
r? @jonas-schievink
This mainly checks that all `BasicBlock` actually exist. On top of that, it checks that `Call` actually calls something of `FnPtr`/`FnDef` type, and `Assert` has to work on a `bool`. Also `SwitchInt` cannot have an empty target list.
linker: Add a linker rerun hack for gcc versions not supporting -static-pie
Which mirrors the existing `-no-pie` linker rerun hack, but the logic is a bit more elaborated in this case.
If the linker (gcc or clang) errors on `-static-pie` we rerun in with `-static` instead.
We must also replace CRT objects corresponding to `-static-pie` with ones corresponding to `-static` in this case.
(One sanity check for CRT objects in target specs is also added as a drive-by fix.)
To do in the future: refactor all linker rerun hacks into separate functions and share more code with `add_(pre,post)_link_objects`.
This PR accompanies https://github.com/rust-lang/rust/pull/71804 and unblocks https://github.com/rust-lang/rust/pull/70740.
Make `PolyTraitRef::self_ty` return `Binder<Ty>`
This came up during review of #71618. The current implementation is the same as a call to `skip_binder` but harder to audit. Make it preserve binding levels and add a call to `skip_binder` at all use sites so they can be audited as part of #72507.
de-promote Duration::from_secs
In https://github.com/rust-lang/rust/pull/67531, we removed the `rustc_promotable` attribute from a bunch of `Duration` methods, but not from `Duration::from_secs`. This makes the current list of promotable functions the following (courtesy of @ecstatic-morse):
* `INT::min_value`, `INT::max_value`
* `std::mem::size_of`, `std::mem::align_of`
* `RangeInclusive::new` (backing `x..=y`)
* `std::ptr::null`, `std::ptr::null_mut`
* `RawWaker::new`, `RawWakerVTable::new` ???
* `Duration::from_secs`
I feel like the last one stands out a bit here -- the rest are all very core language primitives, and `RawWaker` has a strong motivation for getting a `'static` vtable. But a `&'static Duration`? That seems unlikely. So I propose we no longer promote calls to `Duration::from_secs`, which is what this PR does.
https://github.com/rust-lang/rust/pull/67531 saw zero regressions and I am not aware of anyone complaining that this broke their (non-cratered) code, so I consider it likely the same will be true here, but of course we'd do a crater run.
See [this document](https://github.com/rust-lang/const-eval/blob/master/promotion.md) for some more background on promotion and https://github.com/rust-lang/const-eval/issues/19 for some of the concerns around promoting function calls.
resolve: Sort E0408 errors by Symbol str
This is a request for comments implementing my suggested solution to https://github.com/rust-lang/rust/issues/72913
Previously errors were sorted by Symbol index instead of the string. The indexes are not the same between architectures because Symbols for architecture extensions (e.g. x86 AVX or RISC-V d) are interned before the source file is parsed. RISC-V's naming of extensions after single letters led to it having errors sorted differently for test cases using single letter variable names. Instead sort the errors by the Symbol string so that it is stable across architectures.
While I was at it, there's also 8edb05c2 skipping some ui tests which I think are irrelevant for risc-v.
Rollup of 5 pull requests
Successful merges:
- #72260 (Spell out `Self` in async function return)
- #72996 (Remove unsused `NodeId` related APIs in hir map)
- #73010 (Update RELEASES.md)
- #73017 (Use assert_eq for liballoc test)
- #73019 (add test for #72960)
Failed merges:
r? @ghost
Add myself to mailmap
Now that I have (accidentally) made contributions using two separate email addresses, I've added the rest of the addresses I typical use.
Add more assert to Vec with_capacity docs
Show assertion on len too to show them how adding new items will affect both the
length and capacity, before and after.
We were computing a merge-base between the remote beta and master
branches, but this was giving incorrect answers for the first beta if
the remote hadn't been pushed yet. For instance, `1.45.0-beta.3359`
corresponds to the number of merges since the 1.44 beta, but we really
want just `.1` for the sole 1.45 beta promotion merge.
We don't really need to query the remote beta at all -- `master..HEAD`
suffices if we assume that we're on the intended beta branch already.
Previously errors were sorted by Symbol index instead of the string. The
indexes are not the same between architectures because Symbols for
architecture extensions (e.g. x86 AVX or RISC-V d) are interned before
the source file is parsed. RISC-V's naming of extensions after single
letters led to it having errors sorted differently for test cases using
single letter variable names. Instead sort the errors by the Symbol
string so that it is stable across architectures.
save_analysis: work on HIR tree instead of AST
In order to reduce the uses of `NodeId`s in the compiler, `save_analysis` crate has been reworked to operate on the HIR tree instead of the AST.
cc #50928
