Hacker News new | past | comments | ask | show | jobs | submit login
The Libcurl Transfer State Machine (haxx.se)
42 points by TangerineDream on May 10, 2021 | hide | past | favorite | 7 comments



The "do / doing / doing-more / did / done / completed" cluster just looks like a terrifying knot of possible routes through a bunch of states that have similar names (like a full half of them to me are synonyms...) and yet somehow is barely even mentioned in the text, which I feel "should" spend a ton of time explaining what is going on in that region, but all we get is "there are several such states to properly support all protocols but also for historical reasons". I feel like I want Yoda to show up to to point out "do or do not; there is no doing-more".


Exactly. The do / doing / doing-more / ... distinction is most likely just present, because C's threading/event handling model is too primitive. In a language like Go you would most likely write this as synchronous code without risking running out of OS threads. In Rust you would use tokio and async / await. No need to come up with all of these imaginary states.

Languages like C also make it notoriously hard to layer/decompose stuff. Why does DNS resolution and proxying need to be part of this state machine? I'm not saying that Go's net/http library is perfect, but at least it got this part right. The http.Client type uses a http.RoundTripper to do all of its networking. Separation of concerns.


It seems a little unfair to just jump to the conclusion that the problem is C, with basically no evidence of that. The multitude of "do" states did look confusing to me too, but without reading the code I can't say for sure that it's actually bad, and I definitely can not say why it became bad.


You can write async/await-style code in C with a simple switch-case state machine. It's not quite as convenient as aync-await support directly in the language, but what's happening underneath is the exact same thing.

(meaning: blaming "state machine layout" on the capabilities of the implementation language seems a bit of stretch to me, since async-await language support is essentially just (very high level) syntax sugar for a function that consists of a switch-case block and takes a context pointer as additional argument).


Could anyone recommend a textbook which covers the topic of state machines?



Ooh brilliant. Thank you!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: