This is particularly useful to avoid diverging the modules between a PCH and a translation-unit compilation.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
clang/tools/clang-scan-deps/ClangScanDeps.cpp | ||
---|---|---|
958 | This assumes FD is not empty, i.e. the -format experimental-full argument. We should probably error out early if that's not the case. WDYT about changing the new -brief flag into something like -format experimental-brief? We could avoid the dependency between arguments and we'd also make it clear the make and p1689 don't have a brief variant. |
clang/tools/clang-scan-deps/ClangScanDeps.cpp | ||
---|---|---|
958 |
Ah, good catch! I'll fix.
I consider -brief orthogonal to the format kind, there's no reason we can't have brief versions of the other formats. |
Remove -optimize-args from the test invocations since it's not relevant for the test.
clang/tools/clang-scan-deps/ClangScanDeps.cpp | ||
---|---|---|
958 |
How would a brief make or P1689 output look like? |
clang/tools/clang-scan-deps/ClangScanDeps.cpp | ||
---|---|---|
958 | make could print the number of file dependencies and P1689 maybe also prints the number of modules (I'm not familiar with that). In the apple fork there are additional formats added that would have the same -brief output as in this patch, without any changes. |
LGTM. Note that I think we should consider renaming the -format flag in the future. Ideally, it should reflect the fact that the scanner is doing different work, not just formatting the output differently.
This assumes FD is not empty, i.e. the -format experimental-full argument. We should probably error out early if that's not the case.
WDYT about changing the new -brief flag into something like -format experimental-brief? We could avoid the dependency between arguments and we'd also make it clear the make and p1689 don't have a brief variant.