Tiny fixes to make compiletest work for CloudABI cross builds
I'm currently working toward getting a `src/ci/docker` container working to do isolated/automated builds and testing of `x86_64-unknown-cloudabi`. This is working pretty well, but still requires some fixes to `libtest` and `compiletest`. Here is the first set of fixes that I had to apply.
Fixes#47094
Besides fixing that bug, this change has a user-visible effect on the spans in the "incompatible repr hints" warning and another error: they now point at `foo` and/or `bar` in `repr(foo, bar)` instead of the whole attribute. This is sometimes more precise (e.g., `#[repr(C, packed)]` on an enum points at the `packed`) but sometimes not. I moved a compile-fail test to a ui test to illustrate how it now looks in the common case of only one attribute.
CloudABI doesn't do dynamic linking. For this reason, there is no need
to handle any other TLS model than local-exec. CloudABI's C library
doesn't provide a __tls_get_addr() function to do Dynamic TLS.
By forcing local-exec to be used here, we ensure that we don't generate
function calls to __tls_get_addr().
Use memchr for str::find(char)
This is a 10x improvement for searching for characters.
This also contains the patches from https://github.com/rust-lang/rust/pull/46713 . Feel free to land both separately or together.
cc @mystor @alexcrichton
r? @bluss
fixes#46693
This structure doesn't seem to be used by libtest itself. It is used by
compiletest, but never passed on to anything externally. This makes it
easier to get the testing framework to work for CloudABI crossbuilds, as
CloudABI currently lacks PathBuf, which is used by TestPaths.
Add compiler docs testing to CI.
Fixes#47025.
I don't know if `x86_64-gnu` is the right builder for this, but there seems to be time left on [Travis](https://travis-ci.org/rust-lang/rust/jobs/307488864).
Remaining problems blocking this PR:
- [x] broken links caused by rustdoc issues:
- [x] `pub use self::Enum::...`: #46766 and #46767 (fixed by #47050, thanks @ollie27!)
- [x] `impl Deref for DerefToStdType`: #32129 (ignored in linkchecker)
- [x] `#[feature(decl_macro)]` and `use std::vec`: #47038 (ignored in linkchecker)
- [x] `rustc_data_structures::sync::{Lrc, RwLock}` aliases `std` types: #32130 (ignored in linkchecker)
- [x] markdown differences, in rust repository and in external crates, now failing the build with #46880 merged (all fixed)
- [x] multiple crate updates needed: `rand`, `log`, `parking_lot_core`, `flate2`
- [x] submodule updates needed to deduplicate dependencies: `rust-installer`, ~`cargo`~ (done by #47052)
- [x] #44953 test broken by `log` update (removed, this can be controversial)
- [x] Waiting `x86_64-gnu` build results ([done](https://travis-ci.org/rust-lang/rust/builds/323451069))
See individual commits for more details.
Update `rand` crate to `0.3.19`.
Update `log` crate to `0.3.9` and `0.4.1`.
Update `parking_lot_core` crate to `0.2.9`.
Upgrade all flate2 dependencies to `1.0.1`.
- Update `rust-installer` submodule.
It tested #44953.
`log` macros in newer versions are no longer recursive, so these duplicated
error messages (about unstable feature uses) previously occurring at each
level of recursion are no longer possible, even with the fix by #45540.
Furthermore this test breaks when multiple versions of `log` are in the
sysroot (`log 0.3.9` depends on`log 0.4.1`)
Reword trying to operate in immutable fields
The previous message ("cannot assign/mutably borrow immutable field")
when trying to modify a field of an immutable binding gave the
(incorrect) impression that fields can be mutable independently of their
ADT's binding. Slightly reword the message to read "cannot
assign/mutably borrow field of immutable binding".
Re #35937.
Allow lifetimes in macros
This is a resurrection of PR #41927 which was a resurrection of #33135, which is intended to fix#34303.
In short, this allows macros_rules! to use :lifetime as a matcher to match 'lifetimes.
Still to do:
- [x] Feature gate
Add a tidy check for missing or too many trailing newlines.
I've noticed recently there are lots of review comments requesting to fix trailing newlines. If this is going to be an official style here, it's better to let the CI do this repetitive check.
Improved error messages for linking failure
Partial fix for #46998
It's unnecessary to print the linker options if there is no linker installed in the first place. Currently, for libraries, the output is still printed, but that should be cleaned up in the future. If you don't have gcc or g++ installed, so that no linker is installed on the system, the output is now this:
```
$ ./rustc hello.rs
error: linker `cc` not found
|
= note: No such file or directory (os error 2)
error: aborting due to previous error
```
For libraries, the linker arguments are still printed, but they should be cleaned up in further commits.
The previous message ("cannot assign/mutably borrow immutable field")
when trying to modify a field of an immutable binding gave the
(incorrect) impression that fields can be mutable independently of their
ADT's binding. Slightly reword the message to read "cannot
assign/mutably borrow field of immutable binding".
Don't announce CloudABI as being UNIX.
This was originally brought in, because the definitions are based on
those of FreeBSD, Linux, etc. Even though CloudABI is based on POSIX, it
uses a subset that is so small that it's not reasonable to call it POSIX.
Now that I'm porting libstd, I'm running into some spots where I have to
explicitly disable code paths that were enabled by cfg(unix).
It's unnecessary to print the linker options if there is no linker installed.
Currently, for libraries, the output is still printed, see #46998 for
discussion
Add CloudABI to the list of systems on which we stub out alloc_jemalloc.
The official jemalloc sources don't build cleanly on CloudABI yet, for
the reason that some of its tracing frameworks try to access the global
filesystem namespace, which CloudABI doesn't provide.
Always make use of the malloc implementation used by the C library,
which already happens to be jemalloc with some tiny build fixes.
in which leading zeroes on tuple-struct accesses are abjured
Resolves#47073. If accepted, a point in the compatibility section of the release notes is warranted.
This was originally brought in, because the definitions are based on
those of FreeBSD, Linux, etc. Even though CloudABI is based on POSIX, it
uses a subset that is so small that it's not reasonable to call it POSIX.
Now that I'm porting libstd, I'm running into some spots where I have to
explicitly disable code paths that were enabled by cfg(unix).
Requires tools to test-pass if the corresponding submodule is updated.
Follow up of #46554. Breaking a tool would not stop bors from accepting a merge, but this also means when we intend to *fix* a tool but failed, bors will *still* accept the PR. This behavior is not helpful to the tool maintainers.
This PR attempts to fix this by rejecting the tool-update merge when a tool failed. It recognizes a PR as "tool-updating" when the corresponding submodule is changed, compared with the previous commit.
