Refactor processing of BranchRegions associated with an MCDCDecisionRegion #78819
Conversation
|
|
||
| const auto &RegionsBegin = Record.MappingRegions.begin(); | ||
| const auto &RegionsEnd = Record.MappingRegions.end(); | ||
| for (auto It = RegionsBegin; It != RegionsEnd; ++It) { |
There was a problem hiding this comment.
I would've preferred to keep the range-based loop as preferred by the LLVM coding standards, but the iterator is useful in being able to scan ahead in the list.
| const auto &RegionsEnd = Record.MappingRegions.end(); | ||
| for (auto It = RegionsBegin; It != RegionsEnd; ++It) { | ||
| const auto &Region = *It; | ||
|
|
||
| // If an MCDCDecisionRegion is seen, track the BranchRegions that follow | ||
| // it according to Region.NumConditions. | ||
| if (Region.Kind == CounterMappingRegion::MCDCDecisionRegion) { |
There was a problem hiding this comment.
I think the handling of DecisionRegions is much cleaner -- being assembled and processed in one iteration rather than rely on successive iterations of the loop.
| // Test visualization of MC/DC constructs for branches in macro expansions. | ||
|
|
||
| // RUN: llvm-profdata merge %S/Inputs/mcdc-macro.proftext -o %t.profdata | ||
| // RUN: llvm-cov show --show-expansions --show-branches=count --show-mcdc %S/Inputs/mcdc-macro.o -instr-profile %t.profdata -path-equivalence=/tmp,%S/Inputs %S/Inputs/mcdc-macro.c | FileCheck %s |
There was a problem hiding this comment.
I rely on a pre-built object file for now (as I did in a handful of other cases) because it allows for a targeted check. I know these are tedious to update. Based on suggestions from others in the last code reviews, I'm considering changing these tests to be runtime profile tests in the future, removing the need for the object files. However, for now, I think it's useful for this test.
…egion. This fixes MC/DC issue llvm#77871 in which branches under ExpansionRegions were not being included in the creation of the MC/DC record. The fix is a slight refactor in how branches associated with an MCDCDecisionRegion are gathered such that ExpansionRegions can be scanned recursively.
evodius96
force-pushed
the
llvm-cov-mcdc-macros
branch
from
bf66961
to
57d5647
Compare
|
@evodius96 This also works for me, with #78966. I cannot reproduce any more. AFAIK each group of branch is clustering and your impl can handle them as well. This is shorter however my impl is more descriptive (I think). Both are working to me. Which could we take? |
I'm inclined to go with your approach. I don't see any practical downside to it. Thanks! |
|
Closing in deference to #78969 |
This fixes MC/DC issue #77871 in which branches under ExpansionRegions were not being included in the creation of the MC/DC record. The fix is a slight refactor in how branches associated with an MCDCDecisionRegion are gathered such that ExpansionRegions can be scanned recursively.