gh-107369: optimize textwrap.indent() by methane · Pull Request #107374 · python/cpython

methane

Now that you mention it, I can see that using isspace() is the most obvious way to do this. Why I did not see it earlier?

Because "".isspace() is False. We need to guarantee that "" is not used here.
x and not x.isspace() would be bit obvious, but little slower.

Algorithmically, isspace() looks more preferable, because it does not create a string. But on practice it may not matter in common cases. Did you compare variants with different inputs? For example Misc/NEWS.d/3.8.0a1.rst may show a very different result.

lstrip() is slow when every line has long indent. But Misc/NEWS.d/3.8.0a1.rst has almost no indents.

With 4c6a46a and https://gist.github.com/methane/5c6153c564d9508199a81c48d33161eb

> ./python.exe bench_indent.py Misc/NEWS.d/3.8.0a1.rst
filename='Misc/NEWS.d/3.8.0a1.rst' 8978 lines.
                   lstrip: 0.736msec
          not x.isspace(): 0.877msec
    x and not x.isspace(): 0.929msec

> ./python.exe bench_indent.py Objects/unicodeobject.c
filename='Objects/unicodeobject.c' 15332 lines.
                   lstrip: 1.812msec
          not x.isspace(): 1.877msec
    x and not x.isspace(): 1.970msec

If I add text = textwrap.indent(text, " "*32) before bench:

> ./python.exe bench_indent.py Objects/unicodeobject.c
filename='Objects/unicodeobject.c' 15332 lines.
                   lstrip: 2.259msec
          not x.isspace(): 2.356msec
    x and not x.isspace(): 2.437msec