severe performance degradation between 0.5.2 and 0.6.2

[sandrotosi]

Hello,
i was building numpy documentation with pydata-sphinx-theme/0.5.2 (which is the version pinned in their requirements.txt), but that's affected by #394 which is fixed in 0.6.2, but when i've upgraded to that version, the build time spiked substantially.

I've run a test with these pydata-sphinx-theme versions, with the time (in mm:ss.00 format) it took to rebuild numpy's doc with each version:

• 0.5.2: 5:17.99
• 0.6.2: 25:20.05
• 0.6.3: 24:53.24
• 0.7.0: 24:34.35
• 0.7.1: 24:52.37

as you can see, in using 0.6.2 onward the time is 5x what it was with 0.5.2 (granted, numpy's doc fails with that version, but at the very end of the build process, so that's still relevant), and the only thing changing in this build environment is the pydata-sphinx-theme version used.

I'm wondering if this is something you're aware of and plan to address, or is this a news and maybe some investigation is required?

Thanks,
Sandro

[jorisvandenbossche]

This is a known issue (see eg numpy/numpy#18789 on the numpy side. For which version of numpy are you trying to build the docs?), with several possible solutions. See #364 for an explanation (and #381).

#492 is improving the documentation about the different options to deal with this, and that also led to a new package (https://github.com/executablebooks/sphinx-remove-toctrees)

[sandrotosi]

thanks! i was building numpy 1.21.2

[12rambau]

As this issue is related to numpy and the underlying problem is dealt with in other issues should we close this one ?

[choldgraf]

Yep we have made and documented ways around this in this PR:

• Document how to remove toctrees from pages #492

so I think we can close