Add caching to lychee links checker

[dmathieu]

This adds caching to the links checker, to hopefully reduce the false positive failures due to 429s.

[dmathieu]

I'm currently basing the cache on the GIT commit, but we can make it global if needed.

[MrAlias]

Won't caching on the git sha mean a new cache is used for each commit? That will make the cache ineffective as it will be new for each commit, right?

[dmathieu]

Won't caching on the git sha mean a new cache is used for each commit? That will make the cache ineffective as it will be new for each commit, right?

Based on the lychee and actions/cache documentations, I don't think that's accurate.
https://github.com/lycheeverse/lychee-action?tab=readme-ov-file#utilising-the-cache-feature

We do store the sha in the cache name, but if there is no cache hit, the restore-keys argument means we'll still restore the most recent one, no matter the sha.
See https://github.com/actions/cache?tab=readme-ov-file#inputs

[dmathieu]

Discussing with @pellared, we have decided to only cache if the run is fully successful, because lychee caches everything, including failures.
See lycheeverse/lychee#1400

[MrAlias]

Won't caching on the git sha mean a new cache is used for each commit? That will make the cache ineffective as it will be new for each commit, right?

Based on the lychee and actions/cache documentations, I don't think that's accurate. https://github.com/lycheeverse/lychee-action?tab=readme-ov-file#utilising-the-cache-feature

We do store the sha in the cache name, but if there is no cache hit, the restore-keys argument means we'll still restore the most recent one, no matter the sha. See https://github.com/actions/cache?tab=readme-ov-file#inputs

Doesn't this mean that we will always use the last cache from "restore-keys" in the normal operation? We almost never re-run the link checker on the same commit.

If this is the case using a global cache would be equivalent and reduce the number of caches stored in actions.

[pellared]

Doesn't this mean that we will always use the last cache from "restore-keys" in the normal operation? We almost never re-run the link checker on the same commit.

Yes

If this is the case using a global cache would be equivalent and reduce the number of caches stored in actions.

Can you please describe what do you mean by "a global cache"?

Do you mean using a single cache key value? The problem with it is that the action does not save an updates the cache when the cache was restored using the key. See #5160 (comment)

I think this is the main reason why lychee recommends such way of caching.

[MrAlias]

Do you mean using a single cache key value? The problem with it is that the action does not save an updates the cache when the cache was restored using the key. See #5160 (comment)

👍

[pellared]

Waiting for an non-Splunker approval.

[dashpole]

I haven't looked too deeply into lychee, but approving since this follows https://github.com/lycheeverse/lychee-action?tab=readme-ov-file#utilising-the-cache-feature, and based on other approvals.