Published

The next version of markata will be around a full second faster at building it’s docs, that’s a 30% bump in performance at the current state. This performance will come when virtual environments are stored in the same directory as the source code.

What happened?? #

I was looking through my profiler for some unexpected performance hits, and noticed that the docs plugin was taking nearly a full second (sometimes more), just to run glob.

    |  |- 1.068 glob  markata/plugins/docs.py:40
    |  |  |- 0.838 <listcomp>  markata/plugins/docs.py:82
    |  |  |  `- 0.817 PathSpec.match_file  pathspec/pathspec.py:165
    |  |  |        [14 frames hidden]  pathspec, <built-in>, <string>

Python scandir ignores hidden directories #

I started looking for different solutions and what I found was that I was hitting pathspec with way more files than I needed to.

len(list(Path().glob("**/*.py")))
# 6444
len([Path(f) for f in glob.glob("**/*.py", recursive=True)])
# 110

After digging into the docs I found that glob.glob uses os.scandir which ignores ‘.’ and ‘..’ directories while Path.glob does not.

https://docs.python.org/3/library/os.html#os.scandir

results? #

Now glob.py from the docs plugin does not even show up in the profiler.

I opened up ipython and saw the following results. For some reason as I hit docs.glob it was only hitting 488 ms from ipython, but it was still a massive improvement over the original.

%timeit docs.glob(m)
# 488 ms ± 3.05 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit docs.glob(m)
# 9.37 ms ± 90.9 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)