Commit Graph

2 Commits

Author SHA1 Message Date
Nico Weber 49532137d0 Make AST reading work better with LLVM_APPEND_VC_REV=NO
With LLVM_APPEND_VC_REV=NO, Modules/merge-lifetime-extended-temporary.cpp
would fail if it ran before a0f50d7316 (which changed
the serialization format) and then after, for these reasons:

1. With LLVM_APPEND_VC_REV=NO, the module hash before and after the
   change was the same.

2. Modules/merge-lifetime-extended-temporary.cpp is the only test
   we have that uses -fmodule-cache-path=%t that
   a) actually writes to the cache path
   b) doesn't do `rm -rf %t` at the top of the test

So the old run would write a module file, and then the new run would
try to load it, but the serialized format changed.

Do several things to fix this:

1. Include clang::serialization::VERSION_MAJOR/VERSION_MINOR in
   the module hash, so that when the AST format changes (...and
   we remember to bump these), we use a different module cache dir.
2. Bump VERSION_MAJOR, since a0f50d7316 changed the
   on-disk format in a way that a gch file written before that change
   can't be read after that change.
3. Add `rm -rf %t` to all tests that pass -fmodule-cache-path=%t.
   This is unnecessary from a correctness PoV after 1 and 2,
   but makes it so that we don't amass many cache dirs over time.
   (Arguably, it also makes it so that the test suite doesn't catch
   when we change the serialization format but don't bump
   clang::serialization::VERSION_MAJOR/VERSION_MINOR; oh well.)

Differential Revision: https://reviews.llvm.org/D73202
2020-01-27 22:08:37 -05:00
Richard Smith 5159bbad8b PR38627: Fix handling of exception specification adjustment for
destructors.

We previously tried to patch up the exception specification after
completing the class, which went wrong when the exception specification
was needed within the class body (in particular, by a friend
redeclaration of the destructor in a nested class). We now mark the
destructor as having a not-yet-computed exception specification
immediately after creating it.

This requires delaying various checks against the exception
specification (where we'd previously have just got the wrong exception
specification, and now find we have an exception specification that we
can't compute yet) when those checks fire while the class is being
defined.

This also exposed an issue that we were missing a CodeSynthesisContext
for computation of exception specifications (otherwise we'd fail to make
the module containing the definition of the class visible when computing
its members' exception specs). Adding that incidentally also gives us a
diagnostic quality improvement.

This has also exposed an pre-existing problem: making the exception
specification evaluation context a non-SFINAE context (as it should be)
results in a bootstrap failure; PR38850 filed for this.

llvm-svn: 341499
2018-09-05 22:30:37 +00:00