ankurrj7 wrote:
Pushed one more fix for the RDBMS-style cases we found.
The issue was not just "macro coverage" by itself. The continuation region we
create after a call starts out without a source start, and later it can inherit
an end location from the surrounding region. In a few real shapes that
inherited end was stale:
```c
p = alloc();
if (!p)
return 2;
if (ret) {
f();
return 1;
}
```
and also in the cleanup macro cases where a call continuation is followed by
repeated function-like cleanup macros.
When the region was popped, that stale inherited end could make the mapping
range go out of source order after macro/include unnesting. That is what was
tripping the coverage mapping assertion.
The fix now does two things:
* If a call-continuation region does not have its own end, it looks for the
closest enclosing end location that still stays in source order after the same
unnesting that `popRegions()` will do. So we avoid blindly using a parent end
that belongs to an older source range.
* For function-like macro continuations, the new region starts at the macro
invocation location. I limited that only to call-continuation regions. I had a
broader version first, but that changed existing macro mapping
(`switchmacro.c`), so this keeps the change scoped to the new continuation
regions.
I added coverage-mapping regressions for both the plain guard case and the
repeated cleanup macro case.
Validation I ran after the final version:
```text
ninja -C /scratch/ankurraj/clang_experiment_23/build/cov-broad-build clang
llvm-cov llvm-profdata
git diff --check
git-clang-format --diff HEAD ...
llvm-lit -sv clang/test/CoverageMapping/call-continuations.c \
clang/test/CoverageMapping/call-continuations-tight.c \
clang/test/Profile/call-continuations.c \
clang/test/Driver/coverage-call-continuations.c \
clang/test/CoverageMapping/terminate-statements.cpp \
clang/test/CoverageMapping/switchmacro.c
llvm-lit -sv clang/test/CoverageMapping
```
The focused lit set passed, full `clang/test/CoverageMapping` passed `77/77`,
and I also reran the local RDBMS scenario matrix plus the scratch reducer cases
we collected from the failures.
https://github.com/llvm/llvm-project/pull/201079
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
