| Age | Commit message (Collapse) | Author | Lines |
|---|
|
|
|
Remove reachable coverage without counters to maintain invariant that
either there is no coverage at all or there is a live coverage counter
left that provides the function source hash.
The motivating example would be a following closure:
```rust
let f = |x: bool| {
debug_assert!(x);
};
```
Which, with span changes from #93967, with disabled debug assertions,
after the final CFG simplifications but before removal of dead blocks,
gives rise to MIR:
```rust
fn main::{closure#0}(_1: &[closure@a.rs:2:13: 2:22], _2: bool) -> () {
debug x => _2;
let mut _0: ();
bb0: {
Coverage::Expression(4294967295) = 1 - 2;
return;
}
...
}
```
|
|
To generate a function coverage we need at least one coverage counter,
so a coverage from unreachable blocks is retained only when some live
counters remain.
The previous implementation incorrectly retained unreachable coverage,
because it didn't account for the fact that those live counters can
belong to another function due to inlining.
|
|
|
|
r=wesleywiser
Stabilize `-Z instrument-coverage` as `-C instrument-coverage`
(Tracking issue for `instrument-coverage`: https://github.com/rust-lang/rust/issues/79121)
This PR stabilizes support for instrumentation-based code coverage, previously provided via the `-Z instrument-coverage` option. (Continue supporting `-Z instrument-coverage` for compatibility for now, but show a deprecation warning for it.)
Many, many people have tested this support, and there are numerous reports of it working as expected.
Move the documentation from the unstable book to stable rustc documentation. Update uses and documentation to use the `-C` option.
Addressing questions raised in the tracking issue:
> If/when stabilized, will the compiler flag be updated to -C instrument-coverage? (If so, the -Z variant could also be supported for some time, to ease migrations for existing users and scripts.)
This stabilization PR updates the option to `-C` and keeps the `-Z` variant to ease migration.
> The Rust coverage implementation depends on (and automatically turns on) -Z symbol-mangling-version=v0. Will stabilizing this feature depend on stabilizing v0 symbol-mangling first? If so, what is the current status and timeline?
This stabilization PR depends on https://github.com/rust-lang/rust/pull/90128 , which stabilizes `-C symbol-mangling-version=v0` (but does not change the default symbol-mangling-version).
> The Rust coverage implementation implements the latest version of LLVM's Coverage Mapping Format (version 4), which forces a dependency on LLVM 11 or later. A compiler error is generated if attempting to compile with coverage, and using an older version of LLVM.
Given that LLVM 13 has now been released, requiring LLVM 11 for coverage support seems like a reasonable requirement. If people don't have at least LLVM 11, nothing else breaks; they just can't use coverage support. Given that coverage support currently requires a nightly compiler and LLVM 11 or newer, allowing it on a stable compiler built with LLVM 11 or newer seems like an improvement.
The [tracking issue](https://github.com/rust-lang/rust/issues/79121) and the [issue label A-code-coverage](https://github.com/rust-lang/rust/labels/A-code-coverage) link to a few open issues related to `instrument-coverage`, but none of them seem like showstoppers. All of them seem like improvements and refinements we can make after stabilization.
The original `-Z instrument-coverage` support went through a compiler-team MCP at https://github.com/rust-lang/compiler-team/issues/278 . Based on that, `@pnkfelix` suggested that this needed a stabilization PR and a compiler-team FCP.
|
|
|
|
If we do not add code coverage instrumentation to the `Body` of a
function, then when we go to generate the function record for it, we
won't write any data and this later causes llvm-cov to fail when
processing data for the entire coverage report.
I've identified two main cases where we do not currently add code
coverage instrumentation to the `Body` of a function:
1. If the function has a single `BasicBlock` and it ends with a
`TerminatorKind::Unreachable`.
2. If the function is created using a proc macro of some kind.
For case 1, this typically not important as this most often occurs as
the result of function definitions that take or return uninhabited
types. These kinds of functions, by definition, cannot even be called so
they logically should not be counted in code coverage statistics.
For case 2, I haven't looked into this very much but I've noticed while
testing this patch that (other than functions which are covered by case
1) the skipped function coverage debug message is occasionally triggered
in large crate graphs by functions generated from a proc macro. This may
have something to do with weird spans being generated by the proc macro
but this is just a guess.
I think it's reasonable to land this change since currently, we fail to
generate *any* results from llvm-cov when a function has no coverage
instrumentation applied to it. With this change, we get coverage data
for all functions other than the two cases discussed above.
|
|
These options primarily exist to work around bugs, and those bugs have
largely been fixed. Avoid stabilizing them, so that we don't have to
support them indefinitely.
|
|
Continue supporting -Z instrument-coverage for compatibility for now,
but show a deprecation warning for it.
Update uses and documentation to use the -C option.
Move the documentation from the unstable book to stable rustc
documentation.
|
|
|
|
As discovered in #85461, the MSVC linker treats weak symbols slightly
differently than unix-y linkers do. This causes link.exe to fail with
LNK1227 "conflicting weak extern definition" where as other targets are
able to link successfully.
This changes the dead functions from being generated as weak/hidden to
private/default which, as the LLVM reference says:
> Global values with “private” linkage are only directly accessible by
objects in the current module. In particular, linking code into a module
with a private global value may cause the private to be renamed as
necessary to avoid collisions. Because the symbol is private to the
module, all references can be updated. This doesn’t show up in any
symbol table in the object file.
This fixes the conflicting weak symbols but doesn't address the reason
*why* we have conflicting symbols for these dead functions. The test
cases added in this commit contain a minimal repro of the fundamental
issue which is that the logic used to decide what dead code functions
should be codegen'd in the current CGU doesn't take into account that
functions can be duplicated across multiple CGUs (for instance, in the
case of `#[inline(always)]` functions).
Fixing that is likely to be a more complex change (see
https://github.com/rust-lang/rust/issues/85461#issuecomment-985005805).
Fixes #85461
|
|
Update some `C-unwind` bits and then
|
|
This commit intends to fill out some of the remaining pieces of the
C-unwind ABI. This has a number of other changes with it though to move
this design space forward a bit. Notably contained within here is:
* On `panic=unwind`, the `extern "C"` ABI is now considered as "may
unwind". This fixes a longstanding soundness issue where if you
`panic!()` in an `extern "C"` function defined in Rust that's actually
UB because the LLVM representation for the function has the `nounwind`
attribute, but then you unwind.
* Whether or not a function unwinds now mainly considers the ABI of the
function instead of first checking the panic strategy. This fixes a
miscompile of `extern "C-unwind"` with `panic=abort` because that ABI
can still unwind.
* The aborting stub for non-unwinding ABIs with `panic=unwind` has been
reimplemented. Previously this was done as a small tweak during MIR
generation, but this has been moved to a separate and dedicated MIR
pass. This new pass will, for appropriate functions and function
calls, insert a `cleanup` landing pad for any function call that may
unwind within a function that is itself not allowed to unwind. Note
that this subtly changes some behavior from before where previously on
an unwind which was caught-to-abort it would run active destructors in
the function, and now it simply immediately aborts the process.
* The `#[unwind]` attribute has been removed and all users in tests and
such are now using `C-unwind` and `#![feature(c_unwind)]`.
I think this is largely the last piece of the RFC to implement.
Unfortunately I believe this is still not stabilizable as-is because
activating the feature gate changes the behavior of the existing `extern
"C"` ABI in a way that has no replacement. My thinking for how to enable
this is that we add support for the `C-unwind` ABI on stable Rust first,
and then after it hits stable we change the behavior of the `C` ABI.
That way anyone straddling stable/beta/nightly can switch to `C-unwind`
safely.
|
|
Fixes: #84018
With `-Z instrument-coverage`, coverage reporting of dead blocks
(for example, blocks dropped because a conditional branch is dropped,
based on const evaluation) is now supported.
Note, this PR relands an earlier, reverted PR that failed when compiling
generators. The prior issues with generators has been resolved and a new
test was added to prevent future regressions.
Check out the resulting changes to test coverage of dead blocks in the
test coverage reports in this PR.
|
|
r=tmandry"
This reverts commit e5f83d24aee866a14753a7cedbb4e301dfe5bef5, reversing
changes made to ac888e8675182c703c2cd097957878faf88dad94.
|
|
Report coverage `0` of dead blocks
Fixes: #84018
With `-Z instrument-coverage`, coverage reporting of dead blocks
(for example, blocks dropped because a conditional branch is dropped,
based on const evaluation) is now supported.
If `instrument-coverage` is enabled, `simplify::remove_dead_blocks()`
finds all dropped coverage `Statement`s and adds their `code_region`s as
`Unreachable` coverage `Statement`s to the `START_BLOCK`, so they are
still included in the coverage map.
Check out the resulting changes in the test coverage reports in this PR (in [commit 1](https://github.com/rust-lang/rust/pull/84797/commits/0b0d293c7c46bdadf80e5304a667e34c53c0cf7e)).
r? `@tmandry`
cc: `@wesleywiser`
|
|
Coverage instruments closure bodies in macros (not the macro body)
Fixes: #84884
This solution might be considered a compromise, but I think it is the
better choice.
The results in the `closure.rs` test correctly resolve all test cases
broken as described in #84884.
One test pattern (in both `closure_macro.rs` and
`closure_macro_async.rs`) was also affected, and removes coverage
statistics for the lines inside the closure, because the closure
includes a macro. (The coverage remains at the callsite of the macro, so
we lose some detail, but there isn't a perfect choice with macros.
Often macro implementations are split across the macro and the callsite,
and there doesn't appear to be a single "right choice" for which body
should be covered. For the current implementation, we can't do both.
The callsite is most likely to be the preferred site for coverage.
r? `@tmandry`
cc: `@wesleywiser`
|
|
And adds tests to validate it still works.
There is an anticipated feature request to support a compiler flag that
only adds coverage for specific files (or perhaps mods). As I thought
about where that change would need to be supported, I realized that
checking the attribute in mapgen (for unused functions) was unnecessary.
The unused functions are only synthesized if they have MIR coverage, and
functions with the `no_coverage` attribute will not have been
instrumented with MIR coverage statements in the first place.
New tests confirm this.
Also, while adding tests, I updated resolved comments and FIXMEs in
other tests.
|
|
Fixes: #84884
This solution might be considered a compromise, but I think it is the
better choice.
The results in the `closure.rs` test correctly resolve all test cases
broken as described in #84884.
One test pattern (in both `closure_macro.rs` and
`closure_macro_async.rs`) was also affected, and removes coverage
statistics for the lines inside the closure, because the closure
includes a macro. (The coverage remains at the callsite of the macro, so
we lose some detail, but there isn't a perfect choice with macros.
Often macro implementations are split across the macro and the callsite,
and there doesn't appear to be a single "right choice" for which body
should be covered. For the current implementation, we can't do both.
The callsite is most likely to be the preferred site for coverage.
I applied this fix to all `MacroKinds`, not just `Bang`.
I'm trying to resolve an issue of lost coverage in a
`MacroKind::Attr`-based, function-scoped macro. Instead of only
searching for a body_span that is "not a function-like macro" (that is,
MacroKind::Bang), I'm expanding this to all `MacroKind`s. Maybe I should
expand this to `ExpnKind::Desugaring` and `ExpnKind::AstPass` (or
subsets, depending on their sub-kinds) as well, but I'm not sure that's
a good idea.
I'd like to add a test of the `Attr` macro on functions, but I need time
to figure out how to constract a good, simple example without external
crate dependencies. For the moment, all tests still work as expected (no
change), this new commit shouldn't have a negative affect, and more
importantly, I believe it will have a positive effect. I will try to
confirm this.
|
|
Fixes: #84018
With `-Z instrument-coverage`, coverage reporting of dead blocks
(for example, blocks dropped because a conditional branch is dropped,
based on const evaluation) is now supported.
If `instrument-coverage` is enabled, `simplify::remove_dead_blocks()`
finds all dropped coverage `Statement`s and adds their `code_region`s as
`Unreachable` coverage `Statement`s to the `START_BLOCK`, so they are
still included in the coverage map.
Check out the resulting changes in the test coverage reports in this PR.
|
|
|
|
|
|
|
|
Fixes: #84561
|
|
Adds feature-gated `#[no_coverage]` function attribute, to fix derived Eq `0` coverage issue #83601
Derived Eq no longer shows uncovered
The Eq trait has a special hidden function. MIR `InstrumentCoverage`
would add this function to the coverage map, but it is never called, so
the `Eq` trait would always appear uncovered.
Fixes: #83601
The fix required creating a new function attribute `no_coverage` to mark
functions that should be ignored by `InstrumentCoverage` and the
coverage `mapgen` (during codegen).
Adding a `no_coverage` feature gate with tracking issue #84605.
r? `@tmandry`
cc: `@wesleywiser`
|
|
|
|
The Eq trait has a special hidden function. MIR `InstrumentCoverage`
would add this function to the coverage map, but it is never called, so
the `Eq` trait would always appear uncovered.
Fixes: #83601
The fix required creating a new function attribute `no_coverage` to mark
functions that should be ignored by `InstrumentCoverage` and the
coverage `mapgen` (during codegen).
While testing, I also noticed two other issues:
* spanview debug file output ICEd on a function with no body. The
workaround for this is included in this PR.
* `assert_*!()` macro coverage can appear covered if followed by another
`assert_*!()` macro. Normally they appear uncovered. I submitted a new
Issue #84561, and added a coverage test to demonstrate this issue.
|
|
Fixes: #84180
For chained function calls separated by the `?` try operator, the
function call following the try operator produced a MIR `Call` span that
matched the span of the first call. The `?` try operator started a new
span, so the second call got no span.
It turns out the MIR `Call` terminator has a `func` `Operand`
for the `Constant` representing the function name, and the function
name's Span can be used to reset the starting position of the span.
|
|
coverage of async function bodies should match non-async
This fixes some missing coverage within async function bodies.
Commit 1 demonstrates the problem in the fixed issue, and commit 2 corrects it.
Fixes: #83985
|
|
The body_span was assumed to be in the Span root context, but this was
not the case for async function bodies.
|
|
The initial commit demonstrates the issue, but the fix is not yet
implemented.
Once corrected...
Fixes: #83985
|
|
`continue` statements were missing coverage. This was particularly
noticeable in a match pattern that contained only a `continue`
statement, leaving the branch appear uncounted. This PR addresses the
problem and adds tests to prove it.
|
|
A colleague contacted me and asked why Rust's counters start at 1, when
Clangs appear to start at 0. There is a reason why Rust's internal
counters start at 1 (see the docs), and I tried to keep them consistent
when codegenned to LLVM's coverage mapping format. LLVM should be
tolerant of missing counters, but as my colleague pointed out,
`llvm-cov` will silently fail to generate a coverage report for a
function based on LLVM's assumption that the counters are 0-based.
See:
https://github.com/llvm/llvm-project/blob/main/llvm/lib/ProfileData/Coverage/CoverageMapping.cpp#L170
Apparently, if, for example, a function has no branches, it would have
exactly 1 counter. `CounterValues.size()` would be 1, and (with the
1-based index), the counter ID would be 1. This would fail the check
and abort reporting coverage for the function.
It turns out that by correcting for this during coverage map generation,
by subtracting 1 from the Rust Counter ID (both when generating the
counter increment intrinsic call, and when adding counters to the map),
some uncovered functions (including in tests) now appear covered! This
corrects the coverage for a few tests!
|
|
This change reduces the risk of impacting coverage tests on unrelated
changes (such as MIR and Span changes), and reduces the burden when
blessing coverage changes in case it is necessary.
* Remove all spanview tests. The spanview tests were useful during
development, but they can be generated as needed, via compiler command
line flags. They aren't critical to confirming coverage results. (The
coverage report tests are sufficient.)
When spanview regeneration was necessary, the diffs were way too hard
to read to be useful anyway. So I'm removing them to reduce friction
from a feature that is no longer useful.
* Remove the requirement for `llvm-cov show --debug` when blessing
tests. The `--debug` flag is, unfortunately, only available if LLVM is
built with `optimize = false` (in Rust's config.toml). This adds
significant time and resource burdens to the contributor's build. As
it turns out, for other reasons in the past, I wasn't actually using
the debug output (counter info) to validate coverage anymore either,
so it was required for no reason, I now realize.
|
|
|
|
This matches the behavior of Clang and allows us to remove several
hacks which were needed to ensure functions weren't optimized away
before reaching the instrumentation pass.
|
|
Adjusted LLVM codegen for code compiled with `-Zinstrument-coverage` to
address multiple, somewhat related issues.
Fixed a significant flaw in prior coverage solution: Every counter
generated a new counter variable, but there should have only been one
counter variable per function. This appears to have bloated .profraw
files significantly. (For a small program, it increased the size by
about 40%. I have not tested large programs, but there is anecdotal
evidence that profraw files were way too large. This is a good fix,
regardless, but hopefully it also addresses related issues.
Fixes: #82144
Invalid LLVM coverage data produced when compiled with -C opt-level=1
Existing tests now work up to at least `opt-level=3`. This required a
detailed analysis of the LLVM IR, comparisons with Clang C++ LLVM IR
when compiled with coverage, and a lot of trial and error with codegen
adjustments.
The biggest hurdle was figuring out how to continue to support coverage
results for unused functions and generics. Rust's coverage results have
three advantages over Clang's coverage results:
1. Rust's coverage map does not include any overlapping code regions,
making coverage counting unambiguous.
2. Rust generates coverage results (showing zero counts) for all unused
functions, including generics. (Clang does not generate coverage for
uninstantiated template functions.)
3. Rust's unused functions produce minimal stubbed functions in LLVM IR,
sufficient for including in the coverage results; while Clang must
generate the complete LLVM IR for each unused function, even though
it will never be called.
This PR removes the previous hack of attempting to inject coverage into
some other existing function instance, and generates dedicated instances
for each unused function. This change, and a few other adjustments
(similar to what is required for `-C link-dead-code`, but with lower
impact), makes it possible to support LLVM optimizations.
Fixes: #79651
Coverage report: "Unexecuted instantiation:..." for a generic function
from multiple crates
Fixed by removing the aforementioned hack. Some "Unexecuted
instantiation" notices are unavoidable, as explained in the
`used_crate.rs` test, but `-Zinstrument-coverage` has new options to
back off support for either unused generics, or all unused functions,
which avoids the notice, at the cost of less coverage of unused
functions.
Fixes: #82875
Invalid LLVM coverage data produced with crate brotli_decompressor
Fixed by disabling the LLVM function attribute that forces inlining, if
`-Z instrument-coverage` is enabled. This attribute is applied to
Rust functions with `#[inline(always)], and in some cases, the forced
inlining breaks coverage instrumentation and reports.
|
|
|
|
|
|
|
|
Updates `tools.mk` to explicitly require `SHELL := /bin/dash`, since CI
uses `dash` but other environments (including developer local machines)
may default to `bash`.
Replaces bash-specific shell command in one Makefile with a
dash-compatible alternative, and re-enables the affected Makefile test.
Removes apparently redundant definition of `UNAME`.
|
|
ad-hoc internal solution.
In particular: the specific code to define LLVM_VERSION_11_PLUS here was, for
some reason, using `$(shell ...)` with bash-specific variable replacement code.
On non-bash platforms like dash, that `shell` invocation would fail, and the
LLVM_VERSION_11_PLUS check would always fail, the test would always be ignored,
and thus be treated as a "success" (in the sense that `--bless` would never do
anything).
This was causing me a lot of pain.
|
|
Fixes: #80060
Also adds additional test cases for coverage of doctests.
|
|
This uses the `SourceMap::doctest_offset_line` method to re-map line
numbers from doctests. Remapping columns is not yet done.
Part of issue #79417.
|
|
Fixes: #79569
Fixes: #79566
Fixes: #79565
For the first issue (#79569), I got hit a `debug_assert!()` before
encountering the reported error message (because I have `debug = true`
enabled in my config.toml).
The assertion showed me that some `SwitchInt`s can have more than one
target pointing to the same `BasicBlock`.
I had thought that was invalid, but since it seems to be possible, I'm
allowing this now.
I added a new test for this.
----
In the last two cases above, both tests (intentionally) fail to compile,
but the `InstrumentCoverage` pass is invoked anyway.
The MIR starts with an `Unreachable` `BasicBlock`, which I hadn't
encountered before. (I had assumed the `InstrumentCoverage` pass
would only be invoked with MIRs from successful compilations.)
I don't have test infrastructure set up to test coverage on files that
fail to compile, so I didn't add a new test.
|
|
The original test produced a single crate with two mods, which was not
the goal of the test.
|
|
Added one more test (two files) showing coverage of generics and unused
functions across crates.
Created and referenced new Issues, as requested.
Added comments.
Added a note about the possible effects of compiler options on LLVM
coverage maps.
|
|
Fixes multiple issue with counters, with simplification
Includes a change to the implicit else span in ast_lowering, so coverage
of the implicit else no longer spans the `then` block.
Adds coverage for unused closures and async function bodies.
Fixes: #78542
Adding unreachable regions for known MIR missing from coverage map
Cleaned up PR commits, and removed link-dead-code requirement and tests
Coverage no longer depends on Issue #76038 (`-C link-dead-code` is
no longer needed or enforced, so MSVC can use the same tests as
Linux and MacOS now)
Restrict adding unreachable regions to covered files
Improved the code that adds coverage for uncalled functions (with MIR
but not-codegenned) to avoid generating coverage in files not already
included in the files with covered functions.
Resolved last known issue requiring --emit llvm-ir workaround
Fixed bugs in how unreachable code spans were added.
|
|
Tested and validate results for panic unwind, panic abort, assert!()
macro, TerminatorKind::Assert (for example, numeric overflow), and
async/await.
Implemented a previous documented idea to change Assert handling to be
the same as FalseUnwind and Goto, so it doesn't get its own
BasicCoverageBlock anymore. This changed a couple of coverage regions,
but I validated those changes are not any worse than the prior results,
and probably help assure some consistency (even if some people might
disagree with how the code region is consistently computed).
Fixed issue with async/await. AggregateKind::Generator needs to be
handled like AggregateKind::Closure; coverage span for the outer async
function should not "cover" the async body, which is actually executed
in a separate "closure" MIR.
|
|
* `rustc` should now compile under LLVM 9 or 10
* Compiler generates an error if `-Z instrument-coverage` is specified
but LLVM version is less than 11
* Coverage tests that require `-Z instrument-coverage` and run codegen
should be skipped if LLVM version is less than 11
|
