Avoid sorting predicates by `DefId`
Fixes issue #82920
Even if an item does not change between compilation sessions, it may end
up with a different `DefId`, since inserting/deleting an item affects
the `DefId`s of all subsequent items. Therefore, we use a `DefPathHash`
in the incremental compilation system, which is stable in the face of
changes to unrelated items.
In particular, the query system will consider the inputs to a query to
be unchanged if any `DefId`s in the inputs have their `DefPathHash`es
unchanged. Queries are pure functions, so the query result should be
unchanged if the query inputs are unchanged.
Unfortunately, it's possible to inadvertantly make a query result
incorrectly change across compilations, by relying on the specific value
of a `DefId`. Specifically, if the query result is a slice that gets
sorted by `DefId`, the precise order will depend on how the `DefId`s got
assigned in a particular compilation session. If some definitions end up
with different `DefId`s (but the same `DefPathHash`es) in a subsequent
compilation session, we will end up re-computing a *different* value for
the query, even though the query system expects the result to unchanged
due to the unchanged inputs.
It turns out that we have been sorting the predicates computed during
`astconv` by their `DefId`. These predicates make their way into the
`super_predicates_that_define_assoc_type`, which ends up getting used to
compute the vtables of trait objects. This, re-ordering these predicates
between compilation sessions can lead to undefined behavior at runtime -
the query system will re-use code built with a *differently ordered*
vtable, resulting in the wrong method being invoked at runtime.
This PR avoids sorting by `DefId` in `astconv`, fixing the
miscompilation. However, it's possible that other instances of this
issue exist - they could also be easily introduced in the future.
To fully fix this issue, we should
1. Turn on `-Z incremental-verify-ich` by default. This will cause the
compiler to ICE whenver an 'unchanged' query result changes between
compilation sessions, instead of causing a miscompilation.
2. Remove the `Ord` impls for `CrateNum` and `DefId`. This will make it
difficult to introduce ICEs in the first place.
Now that future-incompat-report support has landed in nightly Cargo, we
can start to make progress towards removing the various proc-macro
back-compat hacks that have accumulated in the compiler.
This PR introduces a new lint `proc_macro_back_compat`, which results in
a future-incompat-report entry being generated. All proc-macro
back-compat warnings will be grouped under this lint. Note that this
lint will never actually become a hard error - instead, we will remove
the special cases for various macros, which will cause older versions of
those crates to emit some other error.
I've added code to fire this lint for the `time-macros-impl` case. This
is the easiest case out of all of our current back-compat hacks - the
crate was renamed to `time-macros`, so seeing a filename with
`time-macros-impl` guarantees that an older version of the parent `time`
crate is in use.
When Cargo's future-incompat-report feature gets stabilized, affected
users will start to see future-incompat warnings when they build their
crates.
Fixes#83112
The location information for a dummy span isn't real, so don't encode
it. This brings the incr comp cache code into line with the Span
`StableHash` impl, which doesn't hash the location information for dummy
spans.
Previously, we would attempt to load the 'original' file from a dummy
span - if the file id changed (e.g. due to being moved on disk), we would get an
ICE, since the Span was still valid due to its hash being unchanged.
Prevent JS error when there is no dependency or other crate documented (or --disable-per-crate-search has been used)
When there is only one crate, the dropdown is removed, creating an error (that you can see pretty easily on docs.rs for example).
r? `@jyn514`
Rollup of 10 pull requests
Successful merges:
- #81465 (Add documentation about formatting `Duration` values)
- #82121 (Implement Extend and FromIterator for OsString)
- #82617 (Document `everybody_loops`)
- #82789 (Get with field index from pattern slice instead of directly indexing)
- #82798 (Rename `rustdoc` to `rustdoc::all`)
- #82804 (std: Fix a bug on the wasm32-wasi target opening files)
- #82943 (Demonstrate best practice for feeding stdin of a child processes)
- #83066 (Add `reverse` search alias for Iterator::rev())
- #83070 (Update cargo)
- #83081 (Fix panic message of `assert_failed_inner`)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Update cargo
7 commits in 970bc67c3775781b9708c8a36893576b9459c64a..32da9eaa5de5be241cf8096ca6b749a157194f77
2021-03-07 18:09:40 +0000 to 2021-03-13 01:18:40 +0000
- Fix logic for determining prefer-dynamic for a dylib. (rust-lang/cargo#9252)
- Fix issue with filtering exclusive target dependencies. (rust-lang/cargo#9255)
- Update pkgid-spec docs. (rust-lang/cargo#9249)
- Wordsmith the edition documentation a bit more (rust-lang/cargo#9233)
- Package ID specification urls must contain a host (rust-lang/cargo#9188)
- Add documentation for JSON message_path. (rust-lang/cargo#9247)
- Fix filter_platform to run on targets other than x86. (rust-lang/cargo#9246)
Add `reverse` search alias for Iterator::rev()
When searching for "reverse" in rustdoc you can't find the rev method on Iterator so here is a search alias for that.
Demonstrate best practice for feeding stdin of a child processes
Documentation change.
It's possible to create a deadlock with stdin/stdout I/O on a single thread:
* the child process may fill its stdout buffer, and have to wait for the parent process to read it,
* but the parent process may be waiting until its stdin write finishes before reading the stdout.
Therefore, the parent process should use separate threads for writing and reading.
These examples are not deadlocking in practice, because they use short strings, but I think it's better to demonstrate code that works even for long writes. The problem is non-obvious and tricky to debug (it seems that even libstd has a similar issue: #45572).
This also demonstrates how to use stdio with threads: it's not obvious that `.take()` can be used to avoid fighting with the borrow checker.
I've checked that the modified examples run fine.
std: Fix a bug on the wasm32-wasi target opening files
This commit fixes an issue pointed out in #82758 where LTO changed the
behavior of a program. It turns out that LTO was not at fault here, it
simply uncovered an existing bug. The bindings to
`__wasilibc_find_relpath` assumed that the relative portion of the path
returned was always contained within thee input `buf` we passed in. This
isn't actually the case, however, and sometimes the relative portion of
the path may reference a sub-portion of the input string itself.
The fix here is to use the relative path pointer coming out of
`__wasilibc_find_relpath` as the source of truth. The `buf` used for
local storage is discarded in this function and the relative path is
copied out unconditionally. We might be able to get away with some
`Cow`-like business or such to avoid the extra allocation, but for now
this is probably the easiest patch to fix the original issue.
Rename `rustdoc` to `rustdoc::all`
When rustdoc lints were changed to be tool lints, the `rustdoc` group was removed, leading to spurious warnings like
```
warning: unknown lint: `rustdoc`
```
The lint group still worked when rustdoc ran, since rustdoc added the group itself.
This renames the group to `rustdoc::all` for consistency with `clippy::all` and the rest of the rustdoc lints.
Follow-up to #80527.
r? ``@Manishearth``
Implement Extend and FromIterator for OsString
Add the following trait impls:
- `impl Extend<OsString> for OsString`
- `impl<'a> Extend<&'a OsStr> for OsString`
- `impl FromIterator<OsString> for OsString`
- `impl<'a> FromIterator<&'a OsStr> for OsString`
Because `OsString` is a platform string with no particular semantics, concatenating them together seems acceptable.
I came across a use case for these trait impls in https://github.com/artichoke/artichoke/pull/1089:
Artichoke is a Ruby interpreter. Its CLI accepts multiple `-e` switches for executing inline Ruby code, like:
```console
$ cargo -q run --bin artichoke -- -e '2.times {' -e 'puts "foo: #{__LINE__}"' -e '}'
foo: 2
foo: 2
```
I use `clap` for command line argument parsing, which collects these `-e` commands into a `Vec<OsString>`. To pass these commands to the interpreter for `Eval`, I need to join them together. Combining these impls with `Iterator::intersperse` https://github.com/rust-lang/rust/issues/79524 would enable me to build a single bit of Ruby code.
Currently, I'm doing something like:
```rust
let mut commands = commands.into_iter();
let mut buf = if let Some(command) = commands.next() {
command
} else {
return Ok(Ok(()));
};
for command in commands {
buf.push("\n");
buf.push(command);
}
```
If there's interest, I'd also like to add impls for `Cow<'a, OsStr>`, which would avoid allocating the `"\n"` `OsString` in the concatenate + intersperse use case.
It seems there are two copies of it: one in `src/test/ui/attributes/`
and one in `src/test/rustdoc-ui/`. I'm guessing this is to test that the
lint is emitted both when you run the compiler and when you run rustdoc.
