Small tweaks with big performance gains

[Rouslan]

Subject: Small tweaks with big performance gains

Feature or Bugfix

• Refactoring

Purpose

• Increase performance

Detail

I was profiling Sphinx with the Breathe extension (it converts Doxygen output and relies on the cpp domain) on a large amount of code and found that the CPPDomain.resolve_xref method took about 460 seconds. By having ASTDeclaration.get_newest_id cache its output value, this went down to 160 seconds. By implementing ASTIdentifier.__eq__ instead of relying the base class implementation (which loops through __dict__), this went down to 110 seconds. There is no difference in the generated output.

[AA-Turner]

Hi Rouslan,

Please may you add a CHANGES entry? Do the current tests adequately cover the changes here?

A

[Rouslan]

I added the entry to CHANGES.

The changes I made shouldn't require any changes to the tests as there should be no difference to the interface of either changed class, with one caveat: the caching of ASTDeclaration.get_newest_id assumes that by the time it is called, no further modifications will be made to that instance of ASTDeclaration. As far as I can tell, this is the case.

[jakobandersen]

Great PR! Your observation about the newest ID is correct, so it's a safe optimization. I started on similar optimizations some time ago, but I never found time to finalize and merge it (master...jakobandersen:sphinx:c_cpp_optimizations). Perhaps there are some things that can be resurrected.

[AA-Turner]

@jakobandersen I've reflected the changes made here to the C++ domain in the C domain, I expect this will be fine.

A

[Rouslan]

Great, thanks for processing this so quickly!

[AA-Turner]

If you're able to find other pain points with your profiling work, I'd be more than happy to look to resolve them!

A

[Rouslan]

I don't plan to spend any more time on this but if you want, you can have the input files I was using for profiling. I uploaded it to OneDrive at https://1drv.ms/u/s!AjEmeqBHiaJDhBSZbQh2CVVA2BJU?e=thAiZq

The data is the Doxygen output from the https://github.com/espressif/esp-idf project. I included a copy of their license in the xml folder where the Doxygen output is, so this should be in compliance with their license.

The Breathe package needs to be installed. I recommend you install the fork I'm working on, which is several times faster: pip install 'breathe@git+https://github.com/Rouslan/breathe.git'

Just uncompress the file and run python3 run_profiler.py to run cProfile and output the results to "profile_results", or run some other profiler on run.py