]> granicus.if.org Git - clang/commit
Revert r143864, as the problem I was actually looking at can be solved
authorChandler Carruth <chandlerc@gmail.com>
Sun, 6 Nov 2011 08:21:04 +0000 (08:21 +0000)
committerChandler Carruth <chandlerc@gmail.com>
Sun, 6 Nov 2011 08:21:04 +0000 (08:21 +0000)
commit0170fff9f4c667fa85267420b87b2c2ba89aaf31
tree98c0e38e1cc38d56a9932375edffdd4232c43895
parent5ffa3eed8b1ca69e1638cffbee2e464276a1ce06
Revert r143864, as the problem I was actually looking at can be solved
a better way. The more I think about it the more worried I am that this
hammer is simply too large. We should only be reaching out to the
filesystem when doing interesting "detection" things, not gratuitously.

Original commit message:
Start pruning down the set of flags passed to CC1 for header search.
This cleans up the CC1 invocations, and reduces the overhead there.
We're still hammering the filesystem looking for the C++ standard
libraries though.

The only reservation I have about this policy is the case of virtualized
files inside of CC1, but it's not clear what the best way to solve that
is. The Driver consistently queries the actual filesystem to make its
decisions. Changing that would be a very large undertaking. It might be
worthwhile, but it's not an immediate goal.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@143865 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Driver/ToolChains.cpp