Ref-strings in logging messages (was: Performance issue with CPython 3.10 + Cython)

[Julian Smith]

On Fri, 7 Oct 2022 18:28:06 +0100
Barry <barry at barrys-emacs.org> wrote:

> > On 7 Oct 2022, at 18:16, MRAB <python at mrabarnett.plus.com> wrote:
> > 
> > On 2022-10-07 16:45, Skip Montanaro wrote:  
> >>> On Fri, Oct 7, 2022 at 9:42 AM Andreas Ames <andreas.0815.qwertz at gmail.com>
> >>> wrote:
> >>> 1. The culprit was me. As lazy as I am, I have used f-strings all over the
> >>> place in calls to `logging.logger.debug()` and friends, evaluating all
> >>> arguments regardless of whether the logger was enabled or not.
> >>>   
> >> I thought there was some discussion about whether and how to efficiently
> >> admit f-strings to the logging package. I'm guessing that's not gone
> >> anywhere (yet).  
> > Letting you pass in a callable to call might help because that you could use lambda.  
> 
> Yep, that’s the obvious way to avoid expensive log data generation.
> Would need logging module to support that use case.

I have some logging code that uses eval() to evaluate expressions using
locals and globals in a parent stack frame, together with a parser to
find `{...}` items in a string.

I guess this constitutes a (basic) runtime implementation of f-strings.
As such it can avoid expensive evaluation/parsing when disabled, though
it's probably slow when enabled compared to native f-strings. It seems
to work quite well in practise, and also allows one to add some extra
formatting features.

For details see:

    https://git.ghostscript.com/?p=mupdf.git;a=blob;f=scripts/jlib.py;h=e85e9f2c4;hb=HEAD#l41

- Jules

-- 
http://op59.net