LLVM's memory dependence analysis doesn't properly account for calls
that could unwind and thus effectively act as a branching point. This
can lead to stores that are only visible when the call unwinds being
removed, possibly leading to calls to drop() functions with b0rked
memory contents.
As there is no fix for this in LLVM yet and we want to keep
compatibility to current LLVM versions anyways, we have to workaround
this bug by omitting the noalias attribute on &mut function arguments.
Benchmarks suggest that the performance loss by this change is very
small.
Thanks to @RalfJung for pushing me towards not removing too many
noalias annotations and @alexcrichton for helping out with the test for
this bug.
Fixes#29485
The pass removes the unwind branch of each terminator, thus moving the responsibility of handling
the -Z no-landing-pads flag to a small self-contained pass… instead of polluting the translator.
r? @Manishearth
I just noticed they can't be rolled up (often modifying the same line(s) in imports). So once I reach the critical amount for them to be merged I'll create a PR that merges all of them.
We no longer have a separate powerpc64 and powerpc64le target_arch, and instead use target_endian to select between the two. These patches fix a couple of remaining issues.
LLVM's memory dependence analysis doesn't properly account for calls
that could unwind and thus effectively act as a branching point. This
can lead to stores that are only visible when the call unwinds being
removed, possibly leading to calls to drop() functions with b0rked
memory contents.
As there is no fix for this in LLVM yet and we want to keep
compatibility to current LLVM versions anyways, we have to workaround
this bug by omitting the noalias attribute on &mut function arguments.
Benchmarks suggest that the performance loss by this change is very
small.
Thanks to @RalfJung for pushing me towards not removing too many
noalias annotations and @alexcrichton for helping out with the test for
this bug.
Fixes#29485
* We don't have SEH-based unwinding yet.
For this reason we don't need operand bundles in MIR trans.
* Refactored some uses of fcx.
* Refactored some calls to `with_block`.
Here's another go at adding emscripten support. This needs to wait again on new [libc definitions](https://github.com/rust-lang-nursery/libc/pull/122) landing. To get the libc definitions right I had to add support for i686-unknown-linux-musl, which are very similar to emscripten's, which are derived from arm/musl.
This branch additionally removes the makefile dependency on the `EMSCRIPTEN` environment variable by not building the unused compiler-rt.
Again, this is not sufficient for actually compiling to asmjs since it needs additional LLVM patches.
r? @alexcrichton
Backtraces, and the compilation of libbacktrace for asmjs, are disabled.
This port doesn't use jemalloc so, like pnacl, it disables jemalloc *for all targets*
in the configure file.
It disables stack protection.
The scope of these refactorings is a little bit bigger than the title implies. See each commit for details.
I’m submitting this for nitpicking now (the first 4 commits), because I feel the basic idea/implementation is sound and should work. I will eventually expand this PR to cover the translator changes necessary for all this to work (+ tests), ~~and perhaps implement a dynamic dropping scheme while I’m at it as well.~~
r? @nikomatsakis
If a new cleanup is added to a cleanup scope, the cached exits for that
scope are cleared, so all previous cleanups have to be translated
again. In the worst case this means that we get N distinct landing pads
where the last one has N cleanups, then N-1 and so on.
As new cleanups are to be executed before older ones, we can instead
cache the number of already translated cleanups in addition to the
block that contains them, and then only translate new ones, if any and
then jump to the cached ones, getting away with linear growth instead.
For the crate in #31381 this reduces the compile time for an optimized
build from >20 minutes (I cancelled the build at that point) to about 11
seconds. Testing a few crates that come with rustc show compile time
improvements somewhere between 1 and 8%. The "big" winner being
rustc_platform_intrinsics which features code similar to that in #31381.
Fixes#31381
The first commit improves detection of unused imports -- it should have been part of #30325. Right now, the unused import in the changed test would not be reported.
The rest of the commits are miscellaneous, independent clean-ups in resolve that I didn't think warranted individual PRs.
r? @nrc
The structure of the old translator as well as MIR assumed that drop glue cannot possibly panic and
translated the drops accordingly. However, in presence of `Drop::drop` this assumption can be
trivially shown to be untrue. As such, the Rust code like the following would never print number 2:
```rust
struct Droppable(u32);
impl Drop for Droppable {
fn drop(&mut self) {
if self.0 == 1 { panic!("Droppable(1)") } else { println!("{}", self.0) }
}
}
fn main() {
let x = Droppable(2);
let y = Droppable(1);
}
```
While the behaviour is allowed according to the language rules (we allow drops to not run), that’s
a very counter-intuitive behaviour. We fix this in MIR by allowing `Drop` to have a target to take
on divergence and connect the drops in such a way so the leftover drops are executed when some drop
unwinds.
Note, that this commit still does not implement the translator part of changes necessary for the
grand scheme of things to fully work, so the actual observed behaviour does not change yet. Coming
soon™.
See #14875.
We used to have CallKind only because there was a requirement to have all successors in a
contiguous memory block. Now that the requirement is gone, remove the CallKind and instead just
have the necessary information inline.
Awesome!
After the truly incredible and embarrassing mess I managed to make in my last pull request, this should be a bit less messy.
Fixes#31267 - with this change, the code mentioned in the issue compiles.
Found and fixed another issue as well - constants of zero-size types, when used in ExprRepeats inside associated constants, were causing the compiler to crash at the same place as #31267. An example of this:
```
struct Bar;
const BAZ: Bar = Bar;
struct Foo([Bar; 1]);
struct Biz;
impl Biz {
const BAZ: Foo = Foo([BAZ; 1]);
}
fn main() {
let foo = Biz::BAZ;
println!("{:?}", foo);
}
```
However, I'm fairly certain that my fix for this is not as elegant as it could be. The problem seems to occur only with an associated constant of a tuple struct containing a fixed size array which is initialized using a repeat expression, and when the element to be repeated provided to the repeat expression is another constant which is of a zero-sized type. The fix works by looking for constants and associated constants which are zero-width and consequently contain no data, but for which rustc is still attempting to emit an LLVM value; it simply stops rustc from attempting to emit anything. By my logic, this should work fine since the only values that are emitted in this case (according to the comments) are for closures with side effects, and constants will never have side effects, so it's fine to simply get rid of them. It fixes the error and things compile fine with it, but I have a sneaking suspicion that it could be done in a far better manner.
r? @nikomatsakis
If a new cleanup is added to a cleanup scope, the cached exits for that
scope are cleared, so all previous cleanups have to be translated
again. In the worst case this means that we get N distinct landing pads
where the last one has N cleanups, then N-1 and so on.
As new cleanups are to be executed before older ones, we can instead
cache the number of already translated cleanups in addition to the
block that contains them, and then only translate new ones, if any and
then jump to the cached ones, getting away with linear growth instead.
For the crate in #31381 this reduces the compile time for an optimized
build from >20 minutes (I cancelled the build at that point) to about 11
seconds. Testing a few crates that come with rustc show compile time
improvements somewhere between 1 and 8%. The "big" winner being
rustc_platform_intrinsics which features code similar to that in #31381.
Fixes#31381
This mirrors the behavior of `clang-cl.exe` by adding a `CodeView` global
variable when emitting debug information. This should in turn help stack traces
that are generated when code is compiled with debuginfo enabled.
Closes#28133
Currently the `mipsel-unknown-linux-gnu` target doesn't actually set the
`target_arch` value to `mipsel` but it rather uses `mips`. Alternatively the
`powerpc64le` target does indeed set the `target_arch` as `powerpc64le`,
causing a bit of inconsistency between theset two.
As these are just the same instance of one instruction set, let's use
`target_endian` to switch between them and only set the `target_arch` as one
value. This should cut down on the number of `#[cfg]` annotations necessary and
all around be a little more ergonomic.
This mirrors the behavior of `clang-cl.exe` by adding a `CodeView` global
variable when emitting debug information. This should in turn help stack traces
that are generated when code is compiled with debuginfo enabled.
Closes#28133
These commits perform a few high-level changes with the goal of enabling i686 MSVC unwinding:
* LLVM is upgraded to pick up the new exception handling instructions and intrinsics for MSVC. This puts us somewhere along the 3.8 branch, but we should still be compatible with LLVM 3.7 for non-MSVC targets.
* All unwinding for MSVC targets (both 32 and 64-bit) are implemented in terms of this new LLVM support. I would like to also extend this to Windows GNU targets to drop the runtime dependencies we have on MinGW, but I'd like to land this first.
* Some tests were fixed up for i686 MSVC here and there where necessary. The full test suite should be passing now for that target.
In terms of landing this I plan to have this go through first, then verify that i686 MSVC works, then I'll enable `make check` on the bots for that target instead of just `make` as-is today.
Closes#25869
This commit transitions the compiler to using the new exception handling
instructions in LLVM for implementing unwinding for MSVC. This affects both 32
and 64-bit MSVC as they're both now using SEH-based strategies. In terms of
standard library support, lots more details about how SEH unwinding is
implemented can be found in the commits.
In terms of trans, this change necessitated a few modifications:
* Branches were added to detect when the old landingpad instruction is used or
the new cleanuppad instruction is used to `trans::cleanup`.
* The return value from `cleanuppad` is not stored in an `alloca` (because it
cannot be).
* Each block in trans now has an `Option<LandingPad>` instead of `is_lpad: bool`
for indicating whether it's in a landing pad or not. The new exception
handling intrinsics require that on MSVC each `call` inside of a landing pad
is annotated with which landing pad that it's in. This change to the basic
block means that whenever a `call` or `invoke` instruction is generated we
know whether to annotate it as part of a cleanuppad or not.
* Lots of modifications were made to the instruction builders to construct the
new instructions as well as pass the tagging information for the call/invoke
instructions.
* The translation of the `try` intrinsics for MSVC has been overhauled to use
the new `catchpad` instruction. The filter function is now also a
rustc-generated function instead of a purely libstd-defined function. The
libstd definition still exists, it just has a stable ABI across architectures
and leaves some of the really weird implementation details to the compiler
(e.g. the `localescape` and `localrecover` intrinsics).
This brings some routine upgrades to the bundled LLVM that we're using, the most
notable of which is a bug fix to the way we handle range asserts when loading
the discriminant of an enum. This fix ended up being very similar to f9d4149c
where we basically can't have a range assert when loading a discriminant due to
filling drop, and appropriate flags were added to communicate this to
`trans::adt`.
When cross compiling for a target that has a larger usize type than the
host system, we use a truncated value to mark data as dropped,
eventually leading to drop calls on already dropped data. To properly
handle this, the drop pattern needs to be of type u64.
Since C_integral truncates its given value to the requested size anyway,
we can also drop the function that chose between the u32 and u64 values,
and always use the u64 constant.
Fixes#31139
The purpose of the translation item collector is to find all monomorphic instances of functions, methods and statics that need to be translated into LLVM IR in order to compile the current crate.
So far these instances have been discovered lazily during the trans path. For incremental compilation we want to know the set of these instances in advance, and that is what the trans::collect module provides.
In the future, incremental and regular translation will be driven by the collector implemented here.
This is a fix for #30741. It simplifies dep-graph tracking for trait matching. I was experimenting with having a greater resolution here, but decided to pare back to just have one dep node for "trait resolutions on trait `Foo`", which means that adding an impl to the trait `Foo` will invalidate all fns that had to do any trait matching at all on `Foo`. This seems like a reasonable starting place.
Independently, I realized I had neglected to record a dependency from trans on typeck -- this is obviously needed, since trans consumes a bunch of data structures that typeck produces (but which are not currently individually tracked) -- and because trans assumes that typeck has been done. Eventually those are going to go away and be replaced with MIR, which will be tracked, so this edge would presumably be derived automatically then, but it's an obvious enough thing to want for now.
r? @arielb1
cc @michaelwoerister -- this might indirectly fix the problem you observed with the trans cache, though it'd be nice to try and craft an independent test case for that.
This commit stabilizes and deprecates the FCP (final comment period) APIs for
the upcoming 1.7 beta release. The specific APIs which changed were:
Stabilized
* `Path::strip_prefix` (renamed from `relative_from`)
* `path::StripPrefixError` (new error type returned from `strip_prefix`)
* `Ipv4Addr::is_loopback`
* `Ipv4Addr::is_private`
* `Ipv4Addr::is_link_local`
* `Ipv4Addr::is_multicast`
* `Ipv4Addr::is_broadcast`
* `Ipv4Addr::is_documentation`
* `Ipv6Addr::is_unspecified`
* `Ipv6Addr::is_loopback`
* `Ipv6Addr::is_unique_local`
* `Ipv6Addr::is_multicast`
* `Vec::as_slice`
* `Vec::as_mut_slice`
* `String::as_str`
* `String::as_mut_str`
* `<[T]>::clone_from_slice` - the `usize` return value is removed
* `<[T]>::sort_by_key`
* `i32::checked_rem` (and other signed types)
* `i32::checked_neg` (and other signed types)
* `i32::checked_shl` (and other signed types)
* `i32::checked_shr` (and other signed types)
* `i32::saturating_mul` (and other signed types)
* `i32::overflowing_add` (and other signed types)
* `i32::overflowing_sub` (and other signed types)
* `i32::overflowing_mul` (and other signed types)
* `i32::overflowing_div` (and other signed types)
* `i32::overflowing_rem` (and other signed types)
* `i32::overflowing_neg` (and other signed types)
* `i32::overflowing_shl` (and other signed types)
* `i32::overflowing_shr` (and other signed types)
* `u32::checked_rem` (and other unsigned types)
* `u32::checked_shl` (and other unsigned types)
* `u32::saturating_mul` (and other unsigned types)
* `u32::overflowing_add` (and other unsigned types)
* `u32::overflowing_sub` (and other unsigned types)
* `u32::overflowing_mul` (and other unsigned types)
* `u32::overflowing_div` (and other unsigned types)
* `u32::overflowing_rem` (and other unsigned types)
* `u32::overflowing_neg` (and other unsigned types)
* `u32::overflowing_shl` (and other unsigned types)
* `u32::overflowing_shr` (and other unsigned types)
* `ffi::IntoStringError`
* `CString::into_string`
* `CString::into_bytes`
* `CString::into_bytes_with_nul`
* `From<CString> for Vec<u8>`
* `From<CString> for Vec<u8>`
* `IntoStringError::into_cstring`
* `IntoStringError::utf8_error`
* `Error for IntoStringError`
Deprecated
* `Path::relative_from` - renamed to `strip_prefix`
* `Path::prefix` - use `components().next()` instead
* `os::unix::fs` constants - moved to the `libc` crate
* `fmt::{radix, Radix, RadixFmt}` - not used enough to stabilize
* `IntoCow` - conflicts with `Into` and may come back later
* `i32::{BITS, BYTES}` (and other integers) - not pulling their weight
* `DebugTuple::formatter` - will be removed
* `sync::Semaphore` - not used enough and confused with system semaphores
Closes#23284
cc #27709 (still lots more methods though)
Closes#27712Closes#27722Closes#27728Closes#27735Closes#27729Closes#27755Closes#27782Closes#27798
The only way to get a value for a zero-sized type is `undef`, so
there's really no point in actually having a return type other than
void for such types. Also, while the comment in return_type_is_void
mentioned something about aiding C ABI support, @eddyb correctly
pointed out on IRC that there is no such thing as a zero-sized type in
C. And even with clang, which allows empty structs, those get
translated as void return types as well.
Fixes#28766
This is groundwork for #30587 (typestrong constant integrals), but imo it's a change that in itself is good, too, since we don't just juggle `u64`s around anymore.
`ty::Disr` will be changed to a `ConstInt` in #30587
This commit stabilizes and deprecates the FCP (final comment period) APIs for
the upcoming 1.7 beta release. The specific APIs which changed were:
Stabilized
* `Path::strip_prefix` (renamed from `relative_from`)
* `path::StripPrefixError` (new error type returned from `strip_prefix`)
* `Ipv4Addr::is_loopback`
* `Ipv4Addr::is_private`
* `Ipv4Addr::is_link_local`
* `Ipv4Addr::is_multicast`
* `Ipv4Addr::is_broadcast`
* `Ipv4Addr::is_documentation`
* `Ipv6Addr::is_unspecified`
* `Ipv6Addr::is_loopback`
* `Ipv6Addr::is_unique_local`
* `Ipv6Addr::is_multicast`
* `Vec::as_slice`
* `Vec::as_mut_slice`
* `String::as_str`
* `String::as_mut_str`
* `<[T]>::clone_from_slice` - the `usize` return value is removed
* `<[T]>::sort_by_key`
* `i32::checked_rem` (and other signed types)
* `i32::checked_neg` (and other signed types)
* `i32::checked_shl` (and other signed types)
* `i32::checked_shr` (and other signed types)
* `i32::saturating_mul` (and other signed types)
* `i32::overflowing_add` (and other signed types)
* `i32::overflowing_sub` (and other signed types)
* `i32::overflowing_mul` (and other signed types)
* `i32::overflowing_div` (and other signed types)
* `i32::overflowing_rem` (and other signed types)
* `i32::overflowing_neg` (and other signed types)
* `i32::overflowing_shl` (and other signed types)
* `i32::overflowing_shr` (and other signed types)
* `u32::checked_rem` (and other unsigned types)
* `u32::checked_neg` (and other unsigned types)
* `u32::checked_shl` (and other unsigned types)
* `u32::saturating_mul` (and other unsigned types)
* `u32::overflowing_add` (and other unsigned types)
* `u32::overflowing_sub` (and other unsigned types)
* `u32::overflowing_mul` (and other unsigned types)
* `u32::overflowing_div` (and other unsigned types)
* `u32::overflowing_rem` (and other unsigned types)
* `u32::overflowing_neg` (and other unsigned types)
* `u32::overflowing_shl` (and other unsigned types)
* `u32::overflowing_shr` (and other unsigned types)
* `ffi::IntoStringError`
* `CString::into_string`
* `CString::into_bytes`
* `CString::into_bytes_with_nul`
* `From<CString> for Vec<u8>`
* `From<CString> for Vec<u8>`
* `IntoStringError::into_cstring`
* `IntoStringError::utf8_error`
* `Error for IntoStringError`
Deprecated
* `Path::relative_from` - renamed to `strip_prefix`
* `Path::prefix` - use `components().next()` instead
* `os::unix::fs` constants - moved to the `libc` crate
* `fmt::{radix, Radix, RadixFmt}` - not used enough to stabilize
* `IntoCow` - conflicts with `Into` and may come back later
* `i32::{BITS, BYTES}` (and other integers) - not pulling their weight
* `DebugTuple::formatter` - will be removed
* `sync::Semaphore` - not used enough and confused with system semaphores
Closes#23284
cc #27709 (still lots more methods though)
Closes#27712Closes#27722Closes#27728Closes#27735Closes#27729Closes#27755Closes#27782Closes#27798
This provides limited support for using associated consts on type parameters. It generally works on things that can be figured out at trans time. This doesn't work for array lengths or match arms. I have another patch to make it work in const expressions.
CC @eddyb @nikomatsakis
The only way to get a value for a zero-sized type is `undef`, so
there's really no point in actually having a return type other than
void for such types. Also, while the comment in return_type_is_void
mentioned something about aiding C ABI support, @eddyb correctly
pointed out on IRC that there is no such thing as a zero-sized type in
C. And even with clang, which allows empty structs, those get
translated as void return types as well.
Fixes#28766
(Note that it might be a good idea to replace *all* calls of
`alloc_ty` with calls to `alloc_ty_init`, to encourage programmers to
consider the appropriate value for the `init` flag when creating
temporary values.)
includes bugfixes pointed out during review:
* Only `call_lifetime_start` for an alloca if the function entry does
not itself initialize it to "dropped."
* Remove `schedule_lifetime_end` after writing an *element* into a
borrowed slice. (As explained by [dotdash][irc], "the lifetime end
that is being removed was for an element in the slice, which is not
an alloca of its own and has no lifetime start of its own")
[irc]: https://botbot.me/mozilla/rust-internals/2016-01-13/?msg=57844504&page=3
This will correctly add the thread_local attribute to the external static
variable "errno":
extern {
#[thread_local]
static errno: c_int;
}
Before this commit, the thread_local attribute is ignored.
In particular, bring back the `zero` flag for `lvalue_scratch_datum`,
which controls whether the alloca's created immediately at function
start are uninitialized at that point or have their embedded
drop-flags initialized to "dropped".
Then made `to_lvalue_datum_in_scope` pass "dropped" as `zero` flag.
Previously it was returning a value, mostly for the two reasons:
* Cloning Lvalue is very cheap most of the time (i.e. when Lvalue is not a Projection);
* There’s users who want &mut lvalue and there’s users who want &lvalue. Returning a value allows
to make either one easier when pattern matching (i.e. Some(ref dest) or Some(ref mut dest)).
However, I’m now convinced this is an invalid approach. Namely the users which want a mutable
reference may modify the Lvalue in-place, but the changes won’t be reflected in the final MIR,
since the Lvalue modified is merely a clone.
Instead, we have two accessors `destination` and `destination_mut` which return a reference to the
destination in desired mode.
`TypeFoldable`s can currently be visited inefficiently with an identity folder that is run only for its side effects. This creates a more efficient visitor for `TypeFoldable`s and uses it to implement `RegionEscape` and `HasProjectionTypes`, fixing cleanup issue #20298.
This is a pure refactoring.
It was recently realized that we accept defaulted type parameters everywhere, without feature gate, even though the only place that we really *intended* to accept them were on types. This PR adds a lint warning unless the "type-parameter-defaults" feature is enabled. This should eventually become a hard error.
This is a [breaking-change] in that new feature gates are required (or simply removing the defaults, which is probably a better choice as they have little effect at this time). Results of a [crater run][crater] suggest that approximately 5-15 crates are affected. I didn't do the measurement quite right so that run cannot distinguish "true" regressions from "non-root" regressions, but even the upper bound of 15 affected crates seems relatively minimal.
[crater]: https://gist.github.com/nikomatsakis/760c6a67698bd24253bf
cc @rust-lang/lang
r? @pnkfelix
This is roughly the same as my previous PR that created a dependency graph, but that:
1. The dependency graph is only optionally constructed, though this doesn't seem to make much of a difference in terms of overhead (see measurements below).
2. The dependency graph is simpler (I combined a lot of nodes).
3. The dependency graph debugging facilities are much better: you can now use `RUST_DEP_GRAPH_FILTER` to filter the dep graph to just the nodes you are interested in, which is super help.
4. The tests are somewhat more elaborate, including a few known bugs I need to fix in a second pass.
This is potentially a `[breaking-change]` for plugin authors. If you are poking about in tcx state or something like that, you probably want to add `let _ignore = tcx.dep_graph.in_ignore();`, which will cause your reads/writes to be ignored and not affect the dep-graph.
After this, or perhaps as an add-on to this PR in some cases, what I would like to do is the following:
- [x] Write-up a little guide to how to use this system, the debugging options available, and what the possible failure modes are.
- [ ] Introduce read-only and perhaps the `Meta` node
- [x] Replace "memoization tasks" with node from the map itself
- [ ] Fix the shortcomings, obviously! Notably, the HIR map needs to register reads, and there is some state that is not yet tracked. (Maybe as a separate PR.)
- [x] Refactor the dep-graph code so that the actual maintenance of the dep-graph occurs in a parallel thread, and the main thread simply throws things into a shared channel (probably a fixed-size channel). There is no reason for dep-graph construction to be on the main thread. (Maybe as a separate PR.)
Regarding performance: adding this tracking does add some overhead, approximately 2% in my measurements (I was comparing the build times for rustdoc). Interestingly, enabling or disabling tracking doesn't seem to do very much. I want to poke at this some more and gather a bit more data -- in some tests I've seen that 2% go away, but on others it comes back. It's not entirely clear to me if that 2% is truly due to constructing the dep-graph at all.
The next big step after this is write some code to dump the dep-graph to disk and reload it.
r? @michaelwoerister