Only put platforms list is less than 10, otherwise load with AJAX

[GuillaumeGomez]

Based on the same principle as #2113.

This PR does actually two things: only load releases if the "big menu" is opened and not if any menu is open. It allows to only load if needed even further.

The second part is actually adding the async load for the targets if there are more than 5 (having libc in mind for that).

If this approach looks good to you, I'll finish the PR by adding tests.

[GuillaumeGomez]

cc @Nemo157

[Nemo157]

I assume this is a significant part of the html sent for libc currently, have you measured how much it is?

[GuillaumeGomez]

I assume this is a significant part of the html sent for libc currently, have you measured how much it is?

It's currently 18.972 bytes. https://docs.rs/libc/latest/libc/ is 888.749 bytes. So it roughly represents 2.13%. Not that much for big pages, but for pages like https://docs.rs/libc/latest/libc/constant.ABDAY_1.html (which is 55.753 bytes), it represents 34%.

[GuillaumeGomez]

I added test for this feature and updated the failing ones (needed to update URLs mostly). It's now ready for review!

[GuillaumeGomez]

Updated the routes.

[GuillaumeGomez]

Fixed the target-redirect and updated the test checking that to ensure it's always set to the expected value. I also reverted the blacklist change I wrongly made.

[GuillaumeGomez]

Fixed the bug, improved the test a bit and updated the csp test.

[Nemo157]

Found a couple more bugs clicking around pages:

• /crate/libc/0.2.147/source/ requests /-/menus/platforms/crate/libc/0.2.147/source/ which 404's
• /crate/libc/0.2.147/builds/1436 gets a list, but they have links like /libc/0.2.147/arm-unknown-linux-musleabi/libc/1436 which search for 1436 in the docs
• /libc/0.2.147/libc/struct.Elf32_Chdr.html works correctly, but /libc/0.2.147/aarch64-unknown-linux-gnu/libc/struct.Elf32_Chdr.html gets links to paths like /crate/libc/0.2.147/target-redirect/arm-unknown-linux-gnueabi/libc/libc/struct.Elf32_Chdr.html, the doubled /libc/libc/ internal path causes it to not find the target file and fallback to a search.

[GuillaumeGomez]

Fixed all the URLs I could think of in /crate, which allowed me to discover cases where the URLs have a target-redirect whereas they shouldn't. You can see the cases in the tests I added.

[Nemo157]

I think that LGTM now, can't find any more issues clicking around 🎉

It might be good to have @syphar or someone to confirm the CSP change is fine, I can't think of any issues it could cause but I'm not that familiar with it.

[GuillaumeGomez]

🎉 Thanks for the review! I'll send a follow-up once merged about the target-redirect issue I mentioned.

[syphar]

reading about connect-src I believe the CSP change is fine.

Apart from that:
Looking back at #2113 we used the /:crate/ prefix so the cache invalidation would cover the release-list. While we don't return a caching header in get_all_releases (yet, which I definitely overlooked), we should keep CDN caching in mind also for the platform list fetching.

And using the /-/ prefix would bring the need to either add a new path to invalidations, or perhaps introduce a different prefix? ( I could imagine /:crate/-/..., if that doesn't overlap with rustdoc)

[GuillaumeGomez]

Changed the menus URLs to start with /crate/:name/:version/menus.

[GuillaumeGomez]

Anything else that remains to be done here?

[syphar]

Anything else that remains to be done here?

sorry for the delay, from my side I see that the routes were changed.
We could now also start actually caching the response, but I could imagine adding this separately later, especially since most platform lists will be directly rendered into the page.

I think a last manual test after the route change would be good, if @Nemo157 doesn't have time I could dig into this.

[Nemo157]

Just did another quick manual test and couldn't find any issues

[GuillaumeGomez]

🎉