Comparison with Tera and Askama about ramhorns

[maciejhirsz]

Those benches are pretty old, but I've just re-run it against Askama 10.5 and it mostly still holds. As to why, I don't know, I haven't looked at Askama code.

In principle the way Askama works it could/should be faster, as Ramhorns is doing some work that Askama does not have to do. That piece of work in Ramhorns is however the single most optimized part of the whole crate, the whole thing is built around having as little overhead in dynamic templates as possible. There are more factors at play, and since we are in nanosecond range things like error handling when doing writes to the output buffer can have relatively significant effect on total time.

Someone once forked a fork of Askama, and that fork run faster than Ramhorns, but I believe there were some disagreements between the authors, including license issues, and I don't think any of those optimizations landed upstream.