llvm dev - Jun 2012 - [LLVMdev] Volunteers needed: Unit test scaffolding for lib/CodeGen

We don't currently have a way of writing unit tests for code in lib/CodeGen,
and we do have some code that could benefit from having tests. (See r158310).

The problem is that you can't just create a MachineFunction test instance
and start poking at it. You need to give it a target first, and certain helper
classes like MachineRegisterInfo are also expected to be present.

I am looking for a volunteer to build the scaffolding required in
unittests/CodeGen to make it possible to write unit tests for MachineFunction
and associated classes. You would need to:

- Set up a unittests/CodeGen target that links in libCodeGen.

- Create a mock target with minimal functionality, preferably without depending
on TableGen.

- Provide an easy way for individual tests to get a MachineFunction instance
with the required decorations.

This would be a good way to learn about target and codegen configuration.

/jakob

On 06/11/2012 08:25 AM, Jakob Stoklund Olesen wrote:
> We don't currently have a way of writing unit tests for code in
lib/CodeGen, and we do have some code that could benefit from having tests. (See
r158310).
>
> The problem is that you can't just create a MachineFunction test
instance and start poking at it. You need to give it a target first, and certain
helper classes like MachineRegisterInfo are also expected to be present.
>
> I am looking for a volunteer to build the scaffolding required in
unittests/CodeGen to make it possible to write unit tests for MachineFunction
and associated classes. You would need to:
>
> - Set up a unittests/CodeGen target that links in libCodeGen.
>
> - Create a mock target with minimal functionality, preferably without
depending on TableGen.
>
> - Provide an easy way for individual tests to get a MachineFunction
instance with the required decorations.
>
> This would be a good way to learn about target and codegen configuration.
>
> /jakob
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
I'm really interested to work on this but cannot start at the moment.
I need to finish the mips16 port first. Maybe in a month and a half I 
will have the time.

It's really dangerous to do optimizations without this kind of tool; no 
way to write decent test cases.

People that have worked on compilers for ages will tell you how peephole 
and other target specific optimization constantly yields bugs of the 
nature "this could never have worked", years after the optimization
was
added and supposedly tested.