refactor: migrate web server to Actix Web

[robjtede]

full-fat Actix Web conversion including (somewhat necessary) improvements to error handling

pros:

• –91 LoC (smallest final result)
• clearer error handling
• we can use automatic HTTP/2 connections from the reverse proxy if we want to
• access to large library of first- and third-party middleware

cons:

• biggest diff

tokei:

 Language            Files        Lines         Code     Comments       Blanks
===============================================================================
 Rust                   31         3861         3324           21          516
 |- Markdown             5           47            0           42            5
 (Total)                           3908         3324           63          521

[Enet4]

The pull requests are in need of a rebase after the migration to tracing, but other than that I hold no objections to moving forward with Actix Web once that is resolved.

[paolobarbolini]

I haven't used actix-web in a while and just seeing the much better out of the box error handling is refreshing.

we can use automatic HTTP/2 connections from the reverse proxy if we want to

Is that worth it given hyperium/h2#531? Or does actix-web have a solution for it?

[robjtede]

Actix Web is already thread-per-core so each worker would have its own "instance" of h2, meaning the mutexes are uncontended. IIRC things are fast in such cases.

[Enet4]

This is in need of a rebase again. Can you handle this?

[Enet4]

The crate_redirect endpoint (GET /crate/:name) stopped working with the new code. I could not understand why from looking at the code, but maybe you have a better clue.

[robjtede]

ahh yes

- #[get("/crate/:name")]
+ #[get("/crate/{name}")]

[Enet4]

That worked! I found no other issues in this migration. 👍