Consider using zopfli to generate .gz files

[twm]

Description

The Zopfli compressor generates gzip-compatible compressed files smaller than zlib, and often pretty close to Brotli. For example I use the Python bindings to compress some web assets here:

ORIGINAL   ZOPFLI    (.gz)  BROTLI   (.br)  FILE
---------  ---------------  --------------  ------------------------------------------
    6,138      1,686   27%     1,398   23%  normalize-8.0.1.css
  318,964    161,987   51%   141,848   44%  inconsolata-33dd95333e9e.ttf
      515        199   39%       178   35%  fonts-8e4fe55c0023.css
   22,141      4,573   21%     4,037   18%  main-b75bac5769ba.css
    9,310      2,512   27%     2,355   25%  main-b75bac5769ba.css.map
    5,805      1,149   20%     1,043   18%  icon-bd3da638f8c2.svg
    2,544      1,056   42%       976   38%  lettertype-dc874be920ee.svg
    8,957      1,940   22%     1,879   21%  logotype-a492afd710d0.svg
   24,838      4,849   20%     4,555   18%  icon-4a76ea6aede6.ico

The main downside is that Zopfli is much slower than zlib, like Brotli.

[adamchainz]

It is slower... wikipedia:

Under default settings, the output of Zopfli is typically 3–8% smaller than zlib's maximum compression, but takes around 80 times longer.

Whitenoise already adds a considerable overhead to deploys by doing so much work compressing up front, in serial (often including third party files that aren't served). I wouldn't want to slow it down further for savings when Brotli is already supported, and widely supported by browsers.

It could perhaps be an option, but I'd like to see more evidence that it's really desirable.

Is zopfli really the official library? It only has 12 stars, doesn't seem to be so popular.

[twm]

Whitenoise already adds a considerable overhead to deploys by doing so much work compressing up front, in serial (often including third party files that aren't served). I wouldn't want to slow it down further for savings when Brotli is already supported, and widely supported by browsers.

It could perhaps be an option, but I'd like to see more evidence that it's really desirable.

Agreed, there's a tradeoff here. I doubt it's worthwhile for large files. I just wanted to make sure you were aware it exists. Unfortunately I don't have any data on .gz vs .br hit rates to share.

Is zopfli really the official library? It only has 12 stars, doesn't seem to be so popular.

The only other one I've found is pyzopfli which appears unmaintained. IIRC I selected zopfli because it:

1. Provides wheels
2. Appeared to be maintained by Googlers (like Zopfli itself), though this no longer seems to be the case

[adamchainz]

Okay. I think the best course of action right now is to leave this open, and if others show support for the feature then we can look into it.