How to reduce require time

[forthrin]

HTTPX has a (relatively!) high footprint on startup compared to other libraries. How can this be reduced?

time   require
0.001  date
0.002  time
0.003  readline
0.004  fileutils
0.008  m3u8
0.012  yaml
0.025  nokogiri
0.028  httpx/connection
0.031  httpx/transcoder
0.033  httpx/pool
0.094  httpx

[HoneyryderChuck]

I think you're comparing apples to oranges. Most of the examples are from bundled/standard gems; if you know how require works, it should be obvious why they're so fast. Also, a lot of gems are known for lazy loading code (and nokogiri is mostly a C extension, not a lot of ruby code to read/parse/evaluate).

If you want to make a fair comparison, you should put it against other http clients. and repeat your example a few times to take caches into account. And you'll see that, while there's substantial room for improvement in httpx, it's not that critical.

[forthrin]

Lazy loading is a good idea. The library shouldn't really need to do anything until you actually use it.

You may close the issue unless you want it as a reminder for future improvements.

[HoneyryderChuck]

Lazy loading is no free lunch either. For instance, you'll be working against CoW deployments if you do so. httpx also used to use autoload for a few things. It made requiring it quite unstable in multithreaded environments, because autoload is not thread-safe. I've also done a bit of introspection, and the outliers here are loading the openssl-related module (which works with stdlib openssl), and the HTTP/2 parser. So in the first case, the cost is diluted, as lots of other code will require openssl. In the last case, it's a feature other alternatives just do not provide.

I'll mark this as answered. Feel free to propose changes which improve the status quo that take the drawbacks into account.