[cfe-commits] r160265 - in /cfe/trunk: include/clang/Tooling/CommandLineClangTool.h lib/Tooling/CMakeLists.txt lib/Tooling/CommandLineClangTool.cpp test/CMakeLists.txt test/Tooling/clang-ast-dump.cpp tools/CMakeLists.txt tools/Makefile tools/clan

[John McCall]

On Jul 20, 2012, at 8:48 AM, Douglas Gregor wrote:
> On Jul 19, 2012, at 10:00 AM, Alexander Kornienko <alexfh at google.com> wrote:
>> On Tue, Jul 17, 2012 at 7:55 PM, Douglas Gregor <dgregor at apple.com> wrote:
>> Okay, that's fair; but this could trivially be added as a parameter to the existing XML dumper. We don't need a whole separate, single-purpose tool for this.
>> ...
>> > Overall, a way to dump the AST is critical to writing clang tools,
>> > especially while you're learning the AST. Being able to dump parts of
>> > the AST of the source you want to work on makes it far easier to find
>> > out why the AST looks the way it does at a certain place where your
>> > refactoring doesn't work than trying to reproduce that with a minimal
>> > source file that doesn't use any non-builtin includes. Many tool
>> > writers are not hardcore C++ experts (the ones that are obviously
>> > don't need the tool that much), and one of the goals for us is to make
>> > writing C++ tools possible for normal people ;)
>> 
>> The XML dump isn't really a great way to learn the AST. Our normal dumper is far better for that purpose, IMO, because it's more complete and produces less noise.
>> 
>> Basically, we don't need XML, we only need a format which is closer to an actual AST structure than just a C++ listing. Both XML and normal dumper use the same format for statements, which is nice. The problem with the normal dumper is how it outputs declarations. It basically outputs fragments in C++, which isn't much useful for those trying to figure out how AST looks like for declarations.If we could change its output to that LISP-style format used for statements, it would be much more helpful in our use-case.
>> 
>> Are there reasons for the normal dumper to exist in its current state?
> 
> I think it's simply laziness.
> 
>> If there are no serious reasons, I could work on switching it's format. Otherwise we could add another dump format for declarations (consistent with the one used for statements), or replace XML with it.
>> 
>> What do you think about this?
> 
> I think this is the right way to go. -ast-dump should dump declarations in the same tree format used for statements. -ast-print will still persist for those who want to see pretty-printed ASTs rather than the actual form of the ASTs.

If we do this, we can just drop the XML dumper.

John.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20120722/338087ed/attachment.html>