https://github.com/atrosinenko requested changes to this pull request.
Oh, I just realized that we have to check that the backend understands the
new-style module flags. Previously, it was technically possible to use them
like `if (M.getModuleFlag("...")) { ... }` without actually checking the value
to be non-zero (or to be exactly `1`)! This may possibly apply to LLD as well.
Thankfully, the backend (at least the mainline one) seems to check the actual
values everywhere. Though, such behavior is only seems to be explicitly tested
for the ``ptrauth-elf-got"` flag. Furthermore, the
`"aarch64-elf-pauthabi-(platform|version)"`flags are frequently tested for
"present vs. absent" (validation that if one of these two flags is set, both
should be) - for such tests, zero value seems to be equivalent to the "absent"
case (is it?).
Considering auto-upgrading the old modules, we probably have nothing better
than keep treating an absent flag as having the value zero: IIRC the module is
not required to contain a target triple, so we cannot decide whether the
missing flag should be set to zero or kept absent. If so, I would explicitly
mention such behavior in a comment somewhere. But note that this makes use of
the fortunate behavior of the mainline backend and might not hold for some
downstream users.
https://github.com/llvm/llvm-project/pull/204226
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
