[PATCH] D18565: Implement an "I'm dtrace, please retain all debug types" option.

[Adrian Prantl via cfe-commits]

> On Mar 30, 2016, at 2:19 PM, Joerg Sonnenberger via cfe-commits <cfe-commits at lists.llvm.org> wrote:
> 
> On Wed, Mar 30, 2016 at 10:54:32AM -0700, Adrian Prantl via cfe-commits wrote:
>>> Let's take a kernel as example. I want to include certain types for
>>> dtrace-like use with /dev/kmem, but not all the 100KB+ of boring types.
>>> Enough for doing some basic post-mortem debugging even on embedded
>>> systems.
>> 
>> If I understand this correctly, you want to pass a list of “interesting”
>> types to clang that are guaranteed to make it into the debug info?
>> At the moment this is not directly possible. What you could do is
>> collect all the interesting types in a header file and then build a
>> Clang module (using -gmodules) from that header. The resulting .pcm
>> will be and object file that has DWARF for all types in the module.
> 
> Oh, the "makes it into the debug info" is likely the easy part. Your
> patch does that :) The real question is whether there is a tool to
> *reduce* the set of defined types based on a whitelist. E.g. I want to
> have a new file with the DWARF definitions for "struct stat" and all
> types used by that, but nothing more. That's a lot of nasty source
> processing to emulate by hand, so I was curious if someone had written a
> tool for that already.

This is how I would do this:
1) Produce a preprocessed source file that includes the struct.
2) Write a shell script that compiles the source clang -o - | llvm-dwarfdump - | FileCheck checking for the struct.
3) Reduce the preprocessed source using delta [1] and your shell script.

-- adrian

[1] http://delta.tigris.org