]> granicus.if.org Git - clang/commit
There were some frustrating problems with the implementation of
authorChandler Carruth <chandlerc@gmail.com>
Sat, 23 Apr 2011 06:30:43 +0000 (06:30 +0000)
committerChandler Carruth <chandlerc@gmail.com>
Sat, 23 Apr 2011 06:30:43 +0000 (06:30 +0000)
commit50465d1d2f5f6dd3d8e18e2cf7bad4691e5ca6a8
tree21a8f19bcb43994f52ca72f5b51afbbbf04db994
parent81ef3e664d8ae250fbb68b2b6ccdeebb6c13ede5
There were some frustrating problems with the implementation of
-Wwrite-strings. First and foremost, once the positive form of the flag
was passed, it could never be disabled by passing -Wno-write-strings.
Also, the diagnostic engine couldn't in turn use -Wwrite-strings to
control diagnostics (as GCC does) because it was essentially hijacked to
drive the language semantics.

Fix this by giving CC1 a clean '-fconst-strings' flag to enable
const-qualified strings in C and ObjC compilations. Corresponding
'-fno-const-strings' is also added. Then the driver is taught to
introduce '-fconst-strings' in the CC1 command when '-Wwrite-strings'
dominates.

This entire flag is basically GCC-bug-compatibility driven, so we also
match GCC's bug where '-w' doesn't actually disable -Wwrite-strings. I'm
open to changing this though as it seems insane.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@130051 91177308-0d34-0410-b5e6-96231b3b80d8
include/clang/Driver/CC1Options.td
include/clang/Driver/Options.td
lib/Driver/Tools.cpp
lib/Frontend/CompilerInvocation.cpp
test/Driver/clang_f_opts.c
test/Sema/warn-write-strings.c
test/SemaObjC/warn-write-strings.m