Various hacks for running a clang tool on Windows.

Various hacks for running a clang tool on Windows.

BUG=424436

[dcheng, 2014-11-11 23:36:06]

dcheng@chromium.org changed reviewers:
+ hans@chromium.org, thakis@chromium.org

[dcheng, 2014-11-11 23:36:10]

Not really for review. This is what I ended up doing to get the clang tools
working on Windows. If I have more time in the future, I'll try to figure out
proper fixes and get them landed in the appropriate place.

Not included in this patch:
- A small shim CMakeLists.txt in third_party/llvm/tools/chrometools that pulled
in the stuff in tools/clang.
- Editing WinNT.h to remove __asm { } blocks. Despite a bunch of info from many
people, I never was able to get this to work.

https://codereview.chromium.org/718873004/diff/1/include_rsp.py
File include_rsp.py (right):

https://codereview.chromium.org/718873004/diff/1/include_rsp.py#newcode40
include_rsp.py:40: # e['command'] += ' --driver-mode=cl'
Lines 38-40 were various attempts to get clang-cl working in cl compatibility
mode. They all failed, hence the awkward argument rewriting section at the top.

https://codereview.chromium.org/718873004/diff/1/include_rsp.py#newcode59
include_rsp.py:59: compile_db = [e for e in compile_db if
e['command'].find('.obj.rsp') == -1]
For some reason, the compile DB had two entries for a lot of files--a nacl one,
and a normal one. ninja never seemed to generate a rsp for the nacl versions,
and having duplicate entries in the compile DB confused the tool--so I just
filtered out everything where we couldn't find a corresponding .rsp file.

https://codereview.chromium.org/718873004/diff/1/tools/clang/rewrite_scoped_r...
File tools/clang/rewrite_scoped_refptr/CMakeLists.txt (right):

https://codereview.chromium.org/718873004/diff/1/tools/clang/rewrite_scoped_r...
tools/clang/rewrite_scoped_refptr/CMakeLists.txt:5: X86AsmParser
This turned out to be unnecessary.

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/run_tool.py
File tools/clang/scripts/run_tool.py (right):

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/run_tool...
tools/clang/scripts/run_tool.py:57: args = ['git.bat', 'ls-files']
Yay, Windows.

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/run_tool...
tools/clang/scripts/run_tool.py:143: self.__edit_count = 0
I'll probably upstream this elsewhere--it's always nice to know there's forward
progress.

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/run_tool...
tools/clang/scripts/run_tool.py:143: self.__edit_count = 0
I'll probably upstream this elsewhere--it's always nice to know there's forward
progress.

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/update.py
File tools/clang/scripts/update.py (right):

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/update.p...
tools/clang/scripts/update.py:26: LLVM_WIN_REVISION = '217738'
I temporarily pinned the revision when I was building locally.

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/update.p...
tools/clang/scripts/update.py:75: return
Rebuilds were getting really long =(

https://codereview.chromium.org/718873004/diff/1/tools/clang/scripts/update.p...
tools/clang/scripts/update.py:197: WriteStampFile(LLVM_WIN_REVISION)
The small CMakeLists.txt shim in third_party/llvm/tools made the ASan build
pretty unhappy, so I had to return before.

[hans, 2014-11-11 23:42:44]

https://codereview.chromium.org/718873004/diff/1/include_rsp.py
File include_rsp.py (right):

https://codereview.chromium.org/718873004/diff/1/include_rsp.py#newcode40
include_rsp.py:40: # e['command'] += ' --driver-mode=cl'
On 2014/11/11 23:36:07, dcheng wrote:
> Lines 38-40 were various attempts to get clang-cl working in cl compatibility
> mode. They all failed, hence the awkward argument rewriting section at the
top.

Is it invoked as clang-cl but not getting into cl mode? That sounds weird.

Otherwise, the way to get it into cl-mode is to use --driver-mode=cl, but it has
to be the first argument or it won't work.

[dcheng, 2014-11-11 23:46:07]

On 2014/11/11 at 23:42:44, hans wrote:
> https://codereview.chromium.org/718873004/diff/1/include_rsp.py
> File include_rsp.py (right):
> 
> https://codereview.chromium.org/718873004/diff/1/include_rsp.py#newcode40
> include_rsp.py:40: # e['command'] += ' --driver-mode=cl'
> On 2014/11/11 23:36:07, dcheng wrote:
> > Lines 38-40 were various attempts to get clang-cl working in cl
compatibility
> > mode. They all failed, hence the awkward argument rewriting section at the
top.
> 
> Is it invoked as clang-cl but not getting into cl mode? That sounds weird.
> 
> Otherwise, the way to get it into cl-mode is to use --driver-mode=cl, but it
has to be the first argument or it won't work.

Ah, I think it probably wasn't the first argument. The duplicate entries for
nacl vs non-nacl would have still been problematic, though I guess I could try
to use another rule to filter them out.