This is the follow-up patch for http://reviews.llvm.org/D19436
- Update the discriminator reading algorithm to match the assignment algorithm.
- Add test to cover the new algorithm.
Details
Details
Diff Detail
Diff Detail
Event Timeline
Comment Actions
Usually we try to reduce the source code before generating the IR test case
(looks like this IR may've been hand reduced further? But I may be
mistaken) for debug info tests & include the source in a comment at the top
of the test so it's easy to understand the basic intent of the test case -
since reconstructing the intent/semantics from the metadata is somewhat
non-trivial.
(also, sometimes we checkin binaries to test the dumper - we don't always,
and I'm not suggesting it here, just mentioning it as a thought - sometimes
we just treat the dumper as "golden" and assume we're just testing the LLVM
IR/CodeGen behavior)
- Dave
| test/DebugInfo/X86/discriminator2.ll | ||
|---|---|---|
| 10 | Would it be simpler/sufficient to do this: void f1();
void f2() {
f1(); f1();
f1(); f1();
}(with column info off? Or, to keep column info on, #define CALLS f1(); f1() and have the body of the function be: CALLS; CALLS; ) | |
Comment Actions
Unfortunately, we can't because I need to have two consecutive
instructions that are both from discriminator 1, but different lines
to trigger this bug. With your proposed change, the code sequence will
be:
line 4 discriminator 0
line 4 discriminator 1
line 5 discriminator 0
line 5 discriminator 1
But with my test case, the code sequence will be:
line 4 discriminator 0
line 5 discriminator 0
line 4 discriminator 1
line 5 discriminator 1
Comment Actions
Fixed another bug in the implemantation to make it consistent with gnu assembler.
PTAL, Thanks
Would it be simpler/sufficient to do this:
void f1(); void f2() { f1(); f1(); f1(); f1(); }(with column info off? Or, to keep column info on, #define CALLS f1(); f1() and have the body of the function be:
)