Cache parsing templates

[Kijewski]

It compiles, tests work just fine, but I did not benchmark yet if the caching is of any use performance-wise.

[Kijewski]

It might well be that we have to convert our Rcs to Arcs to get any performance gains, if rust does not call the derive macro in the same thread. Arcs are ridiculously slow compared to Rcs, but we could use triomphe to mitigate the slow downs a bit.

[GuillaumeGomez]

Better bench in any case to ensure any change you make going forward is positive (but you already know that, just love to state the obvious 😆 ).

[Kijewski]

Well, there's a tiny performance regression:

$ cd rinja_derive_standalone; cargo bench
hello_world             time:   [706.57 µs 774.11 µs 836.25 µs]
                        change: [+891.79% +965.79% +1053.5%] (p = 0.00 < 0.05)
                        Performance has regressed.

item_info.html          time:   [805.59 µs 852.78 µs 903.00 µs]
                        change: [+982.26% +1020.3% +1066.4%] (p = 0.00 < 0.05)
                        Performance has regressed.

ctrl+C

Not using any includes, etc., the cache is useless in this benchmark, of course, but I have no idea why the slow down are this terrible. :/

[GuillaumeGomez]

It'd be better to replace clone calls (here and above) with Rc::clone. Not useful for performance, but better for reading. :)

[GuillaumeGomez]

Just thought that it's not really useful to keep these statics as thread local.

[GuillaumeGomez]

Not using any includes, etc., the cache is useless in this benchmark, of course, but I have no idea why the slow down are this terrible. :/

Well, it's still less than 1ms. So not really that important I think. Still be interesting to try to understand why these two statics are so bad at this overall.

Also would be interesting to have test cases using a lot of include.

[Kijewski]

Also would be interesting to have test cases using a lot of include.

Yeah, but I have no idea how to easily implement this. :/ I guess we could use vfs to have a fake file-system, but implementing that would be kinda difficult I guess, and I'd defer that task to somewhen in the future.

[Kijewski]

Rinja_derive_standalone's benchmarks are useless now: We only measure how much faster the cache is than parsing. The parsing itself is not benchmarked anymore because it's part of the warm-up phase.

Benchmark results

hello_world             time:   [46.368 µs 46.431 µs 46.492 µs]
                        change: [-9.6500% -9.0053% -8.3628%] (p = 0.00 < 0.05)
                        Performance has improved.

item_info.html          time:   [56.031 µs 56.168 µs 56.293 µs]
                        change: [-18.971% -18.687% -18.425%] (p = 0.00 < 0.05)
                        Performance has improved.

item_union.html         time:   [141.11 µs 141.43 µs 141.80 µs]
                        change: [-26.505% -26.226% -25.940%] (p = 0.00 < 0.05)
                        Performance has improved.

page.html               time:   [607.29 µs 608.38 µs 609.53 µs]
                        change: [-24.460% -24.242% -23.995%] (p = 0.00 < 0.05)
                        Performance has improved.

print_item.html         time:   [130.40 µs 130.81 µs 131.36 µs]
                        change: [-20.143% -19.854% -19.521%] (p = 0.00 < 0.05)
                        Performance has improved.

short_item_info.html    time:   [111.89 µs 112.10 µs 112.34 µs]
                        change: [-21.130% -20.563% -19.876%] (p = 0.00 < 0.05)
                        Performance has improved.

sidebar.html            time:   [169.74 µs 170.02 µs 170.32 µs]
                        change: [-24.447% -24.216% -23.991%] (p = 0.00 < 0.05)
                        Performance has improved.

What you would expect: "hello world" with only one expression is only mildly faster, and the more complex a template, the bigger the speed-up.

[GuillaumeGomez]

Changes are much smaller, nicely done. :)

However, the MSRV seems to be 1.71. So just need to update it and then it's all good!

[Kijewski]

Caching reading the configuration is still missing, then it's ready. :)

[GuillaumeGomez]

I have to admit I didn't even think about the config file. Good catch! Please ping me once it's ready for ready.

[Kijewski]

It's ready to review! :) The last commit is a overhaul of the previous three commits, but it would be terror to refactor them. I hope that's okay! I think I made the diff as small as possible.

[GuillaumeGomez]

Nah it's fine, I don't care. 😆

Thanks a lot!