[cfe-commits] [PATCH] Adds a FixedCompilationDatabase

[Manuel Klimek]

On Tue, Apr 10, 2012 at 10:07 PM, Douglas Gregor <dgregor at apple.com> wrote:
>
> On Apr 10, 2012, at 1:05 PM, Manuel Klimek <klimek at google.com> wrote:
>
>> On Tue, Apr 10, 2012 at 7:54 PM, Douglas Gregor <dgregor at apple.com> wrote:
>>>
>>> On Apr 10, 2012, at 5:17 AM, Manuel Klimek <klimek at google.com> wrote:
>>>
>>>> Added parsing code and integrated it into clang-check (which I'm now
>>>> heavily testing in my vim session :)
>>>>
>>>> As a nice side effect this gives us a beautiful way to write FileCheck
>>>> integration tests for clang tools.
>>>> I'd still like to be able to pull something out that encapsulates most
>>>> of the command line parsing for tools, so it's less code, but I want
>>>> to leave that for later.
>>>
>>> Patch generally looks good, although this…
>>>
>>> +std::vector<CompileCommand>
>>> +FixedCompilationDatabase::getCompileCommands(StringRef FilePath) const {
>>> +  std::vector<CompileCommand> Result(CompileCommands);
>>> +  Result[0].CommandLine.push_back(FilePath);
>>> +  return Result;
>>> +}
>>>
>>> doesn't actually seem right. What if the CompileCommands contains multiple source files, e.g.,
>>>
>>>        clang-check -- a.cpp b.cpp
>>>
>>> shouldn't we filter out the other non-source files, or return an empty compile command if the command line didn't specify the given file name (say, if the CompilationDatabase is asked to return a compile command for c.cpp)?
>>
>> I tried to document that in the chandler-length comments of the
>> FixedCompilationDatabase, but apparently I failed :)
>
> I just missed it, sorry.
>
>> The idea is that you'll specify the TUs to work on, the same way you
>> do for other clang tools, before the "--".
>> Your example would be
>> clang-check . a.cpp b.cpp -- -c ...
>
>
> I can live with that, although I'll note that it's still a little unfortunate that I can't drop in "clang-check" as if it were a compiler and have it do the right thing.

Yep, I generally agree. The problem is that then we have to be able to
parse a compile command line, which according to Chandler means to let
the Driver create a Compiler instance, and then using that to drive
the tool. I'm not sure how this would turn out architecture wise, and
I don't expect it to be very important, as, if you want to run the
tool like a compiler, just use a clang plugin - that's what they're
really good at :)

Cheers,
/Manuel