llvm: change data layout bug to an error and make it trigger more
Fixes#33446.
Don't skip the inconsistent data layout check for custom LLVMs or non-built-in targets.
With #118708, all targets will have a simple test that would trigger this error if LLVM's data layouts do change - so data layouts would be corrected during the LLVM upgrade. Therefore, with builtin targets, this error won't happen with our LLVM because each target will have been confirmed to work. With non-builtin targets, this error is probably useful to have because you can change the data layout in your target and if it is wrong then that could lead to bugs.
When using a custom LLVM, the same justification makes sense for non-builtin targets as with our LLVM, the user can update their target to match their LLVM and that's probably a good thing to do. However, with a custom LLVM, the user cannot change the builtin target data layouts if they don't match - though given that the compiler's data layout is used for layout computation and a bunch of other things - you could get some bugs because of the mismatch and probably want to know about that. I'm not sure if this is something that people do and is okay, but I doubt it?
`CFG_LLVM_ROOT` was also always set during local development with `download-ci-llvm` so this bug would never trigger locally.
In #33446, two points are raised:
- In the issue itself, changing this from a `bug!` to a proper error is what is suggested, by using `isCompatibleDataLayout` from LLVM, but that function still just does the same thing that we do and check for equality, so I've avoided the additional code necessary to do that FFI call.
- `@Mark-Simulacrum` suggests a different check is necessary to maintain backwards compatibility with old LLVM versions. I don't know how often this comes up, but we can do that with some simple string manipulation + LLVM version checks as happens already for LLVM 17 just above this diff.
Don't skip the inconsistent data layout check for custom LLVMs.
With #118708, all targets will have a simple test that would trigger this
check if LLVM's data layouts do change - so data layouts would be
corrected during the LLVM upgrade. Therefore, with builtin targets, this
check won't trigger with our LLVM because each target will have been
confirmed to work. With non-builtin targets, this check is probably
useful to have because you can change the data layout in your target and
if its wrong then that could lead to bugs.
When using a custom LLVM, the same justification makes sense for
non-builtin targets as with our LLVM, the user can update their target to
match their LLVM and that's probably a good thing to do. However, with
a custom LLVM, the user cannot change the builtin target data layouts if
they don't match - though given that the compiler's data layout is used
for layout computation and a bunch of other things - you could get some
bugs because of the mismatch and probably want to know about that.
`CFG_LLVM_ROOT` was also always set during local development with
`download-ci-llvm` so this bug would never trigger locally.
Signed-off-by: David Wood <david@davidtw.co>
libtest: Fix padding of benchmarks run as tests
### Summary
The first commit adds regression tests for libtest padding.
The second commit fixes padding for benches run as tests and updates the blessed output of the regression tests to make it clear what effect the fix has on padding.
Closes#104092 which is **E-help-wanted** and **regression-from-stable-to-stable**
### More details
Before this fix we applied padding _before_ manually doing what `convert_benchmarks_to_tests()` does which affects padding calculations. Instead use `convert_benchmarks_to_tests()` first if applicable and then apply padding afterwards so it becomes correct.
Benches should only be padded when run as benches to make it easy to compare the benchmark numbers. Not when run as tests.
r? `@ghost` until CI passes.
The output is produced by printf from C code in these cases, and printf prints in text mode, which means `\n` will be printed as `\r\n` on Windows.
In --bless mode the new output with `\r\n` will replace expected output in `tests/run-make/raw-dylib-*\output.txt` files, which use \n, always resulting in dirty files in the repo.
`tokenstream::Spacing` appears on all `TokenTree::Token` instances,
both punct and non-punct. Its current usage:
- `Joint` means "can join with the next token *and* that token is a
punct".
- `Alone` means "cannot join with the next token *or* can join with the
next token but that token is not a punct".
The fact that `Alone` is used for two different cases is awkward.
This commit augments `tokenstream::Spacing` with a new variant
`JointHidden`, resulting in:
- `Joint` means "can join with the next token *and* that token is a
punct".
- `JointHidden` means "can join with the next token *and* that token is a
not a punct".
- `Alone` means "cannot join with the next token".
This *drastically* improves the output of `print_tts`. For example,
this:
```
stringify!(let a: Vec<u32> = vec![];)
```
currently produces this string:
```
let a : Vec < u32 > = vec! [] ;
```
With this PR, it now produces this string:
```
let a: Vec<u32> = vec![] ;
```
(The space after the `]` is because `TokenTree::Delimited` currently
doesn't have spacing information. The subsequent commit fixes this.)
The new `print_tts` doesn't replicate original code perfectly. E.g.
multiple space characters will be condensed into a single space
character. But it's much improved.
`print_tts` still produces the old, uglier output for code produced by
proc macros. Because we have to translate the generated code from
`proc_macro::Spacing` to the more expressive `token::Spacing`, which
results in too much `proc_macro::Along` usage and no
`proc_macro::JointHidden` usage. So `space_between` still exists and
is used by `print_tts` in conjunction with the `Spacing` field.
This change will also help with the removal of `Token::Interpolated`.
Currently interpolated tokens are pretty-printed nicely via AST pretty
printing. `Token::Interpolated` removal will mean they get printed with
`print_tts`. Without this change, that would result in much uglier
output for code produced by decl macro expansions. With this change, AST
pretty printing and `print_tts` produce similar results.
The commit also tweaks the comments on `proc_macro::Spacing`. In
particular, it refers to "compound tokens" rather than "multi-char
operators" because lifetimes aren't operators.
Avoid adding builtin functions to `symbols.o`
We found performance regressions in #113923. The problem seems to be that `--gc-sections` does not remove these symbols. I tested that lld removes these symbols, but ld and gold do not.
I found that `used` adds symbols to `symbols.o` at 3e202ead60/compiler/rustc_codegen_ssa/src/back/linker.rs (L1786-L1791).
The PR removes builtin functions.
Note that under LTO, ld still preserves these symbols. (lld will still remove them.)
The first commit also fixes#118559. But I think the second commit also makes sense.
Before this fix we applied padding before manually doing what
`convert_benchmarks_to_tests()` does. Instead use
`convert_benchmarks_to_tests()` if applicable and then apply padding
afterwards so it becomes correct. (Benches should only be padded when
run as benches to make it easy to compare the benchmark numbers.)
`build_session` is passed an `EarlyErrorHandler` and then constructs a
`Handler`. But the `EarlyErrorHandler` is still used for some time after
that.
This commit changes `build_session` so it consumes the passed
`EarlyErrorHandler`, and also drops it as soon as the `Handler` is
built. As a result, `parse_cfg` and `parse_check_cfg` now take a
`Handler` instead of an `EarlyErrorHandler`.
Report errors in jobserver inherited through environment variables
This pr attempts to catch situations, when jobserver exists, but is not being inherited.
r? `@petrochenkov`
As you can see the padding is wrong when running benches as tests. This
will be fixed in the next commit. (Benches should only be padded when
run as benches to make it easy to compare the benchmark numbers.)
Restore `#![no_builtins]` crates participation in LTO.
After #113716, we can make `#![no_builtins]` crates participate in LTO again.
`#![no_builtins]` with LTO does not result in undefined references to the error. I believe this type of issue won't happen again.
Fixes#72140. Fixes#112245. Fixes#110606. Fixes#105734. Fixes#96486. Fixes#108853. Fixes#108893. Fixes#78744. Fixes#91158. Fixes https://github.com/rust-lang/cargo/issues/10118. Fixes https://github.com/rust-lang/compiler-builtins/issues/347.
The `nightly-2023-07-20` version does not always reproduce problems due to changes in compiler-builtins, core, and user code. That's why this issue recurs and disappears.
Some issues were not tested due to the difficulty of reproducing them.
r? pnkfelix
cc `@bjorn3` `@japaric` `@alexcrichton` `@Amanieu`
Added linker_arg(s) Linker trait methods for link-arg to be prefixed "-Wl," for cc-like linker args and not verbatim
https://github.com/rust-lang/rust/issues/99427#issuecomment-1234443468
> here's one possible improvement to -l link-arg making it more portable between linkers and useful - befriending it with the verbatim modifier (https://github.com/rust-lang/rust/issues/99425).
>
> -l link-arg:-verbatim=-foo would add -Wl,-foo (or equivalent) when C compiler is used as a linker, and just -foo when bare linker is used.
> -l link-arg:+verbatim=-bar on the other hand would always pass just -bar.
They've been deprecated for four years.
This commit includes the following changes.
- It eliminates the `rustc_plugin_impl` crate.
- It changes the language used for lints in
`compiler/rustc_driver_impl/src/lib.rs` and
`compiler/rustc_lint/src/context.rs`. External lints are now called
"loaded" lints, rather than "plugins" to avoid confusion with the old
plugins. This only has a tiny effect on the output of `-W help`.
- E0457 and E0498 are no longer used.
- E0463 is narrowed, now only relating to unfound crates, not plugins.
- The `plugin` feature was moved from "active" to "removed".
- It removes the entire plugins chapter from the unstable book.
- It removes quite a few tests, mostly all of those in
`tests/ui-fulldeps/plugin/`.
Closes#29597.
Allow target specs to use an LLD flavor, and self-contained linking components
This PR allows:
- target specs to use an LLD linker-flavor: this is needed to switch `x86_64-unknown-linux-gnu` to using LLD, and is currently not possible because the current flavor json serialization fails to roundtrip on the modern linker-flavors. This can e.g. be seen in https://github.com/rust-lang/rust/pull/115622#discussion_r1321312880 which explains where an `Lld::Yes` is ultimately deserialized into an `Lld::No`.
- target specs to declare self-contained linking components: this is needed to switch `x86_64-unknown-linux-gnu` to using `rust-lld`
- adds an end-to-end test of a custom target json simulating `x86_64-unknown-linux-gnu` being switched to using `rust-lld`
- disables codegen backends from participating because they don't support `-Zgcc-ld=lld` which is the basis of mcp510.
r? `@petrochenkov:` if the approach discussed https://github.com/rust-lang/rust/pull/115622#discussion_r1329403467 and on zulip would work for you: basically, see if we can emit only modern linker flavors in the json specs, but accept both old and new flavors while reading them, to fix the roundtrip issue.
The backwards compatible `LinkSelfContainedDefault` variants are still serialized and deserialized in `crt-objects-fallback`, while the spec equivalent of e.g. `-Clink-self-contained=+linker` is serialized into a different json object (with future-proofing to incorporate `crt-objects-fallback` in the future).
---
I've been test-driving this in https://github.com/rust-lang/rust/pull/113382 to test actually switching `x86_64-unknown-linux-gnu` to `rust-lld` (and fix what needs to be fixed in CI, bootstrap, etc), and it seems to work fine.
Chrome links .rlibs with /WHOLEARCHIVE or -Wl,--whole-archive to prevent
the linker from discarding static initializers. This works well, except
on Windows x86, where lld complains:
error: /safeseh: lib.rmeta is not compatible with SEH
The fix is simply to mark the .rmeta as SAFESEH aware. This is trivially
true, since the metadata file does not contain any executable code.
Most coverage metadata is encoded into two sections in the final executable.
The `__llvm_covmap` section mostly just contains a list of filenames, while the
`__llvm_covfun` section contains encoded coverage maps for each instrumented
function.
The catch is that each per-function record also needs to contain a hash of the
filenames list that it refers to. Historically this was handled by assembling
most of the per-function data into a temporary list, then assembling the
filenames buffer, then using the filenames hash to emit the per-function data,
and then finally emitting the filenames table itself.
However, now that we build the filenames table up-front (via a separate
traversal of the per-function data), we can hash and emit that part first, and
then emit each of the per-function records immediately after building. This
removes the awkwardness of having to temporarily store nearly-complete
per-function records.
Mention the syntax for `use` on `mod foo;` if `foo` doesn't exist
Newcomers might get confused that `mod` is the only way of defining scopes, and that it can be used as if it were `use`.
Fix#69492.
Implement rustc part of RFC 3127 trim-paths
This PR implements (or at least tries to) [RFC 3127 trim-paths](https://github.com/rust-lang/rust/issues/111540), the rustc part. That is `-Zremap-path-scope` with all of it's components/scopes.
`@rustbot` label: +F-trim-paths
Use `YYYY-MM-DDTHH_MM_SS` as datetime format for ICE dump files
Windows paths do not support `:`, so use a datetime format in ICE dump paths that Windows will accept.
CC #116809, fix#115180.
MacOS (and all apple targets) have -Csplit-debuginfo=packed as default
instead of "off" like all other targets (well there is Windows, but we
don't test it in those tests), also -Csplit-debuginfo is not stable on
all targets so we only set in on Darwin where is matters.
When building with LTO, builtin functions that are defined but whose calls have not been inserted yet, get internalized.
We need to prevent these symbols from being internalized at LTO time.
Refer to https://reviews.llvm.org/D49434.
After #113716, we can make `#![no_builtins]` crates participate in LTO again.
`#![no_builtins]` with LTO does not result in undefined references to the error.
