The plan is to change how multilib works under the hood and this test
gives confidence that the changes won't break which multilib variant is
selected.
Details
- Reviewers
phosek
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Thanks, some of these are no longer needed which could simplify this change so I sent out D143050 to clean up the Fuchsia driver first.
clang/lib/Driver/ToolChains/Fuchsia.cpp | ||
---|---|---|
316–317 | Could we pass const ArgList &Args instead? I'm concerned that this approach won't scale in the future as we keep on adding more flags. | |
clang/unittests/Driver/FuchsiaTest.cpp | ||
18–23 | Nit: we prefer // for comments. | |
30–43 | While exhaustive, I'm concerned about the scalability of this approach. We plan on adding more multilibs in the future so the number of combinations is going to grow exponentially. A more scalable approach would be to check only the minimal set of combination necessary to achieve full coverage. |
A more scalable approach would be to check only the minimal set of combination necessary to achieve full coverage.
In that case the testing you've already got in place in clang/test/Driver/fuchsia.cpp is adequate. There was a potential bug here so I was being extra cautious, but in practice the test added in this change didn't pick up anything that wasn't also caught by the existing tests. I'll abandon this change and remove it from the stack. The change stack now begins at D142893.
Could we pass const ArgList &Args instead? I'm concerned that this approach won't scale in the future as we keep on adding more flags.