Fix #2034

[yhirose]

No description provided.

[falbrechtskirchinger]

Wouldn't it make sense at some point to move those/some parameters into a Session or SocketConfig struct? It's quite unwieldy already.

[yhirose]

Yes, the code can be refactored later.

[falbrechtskirchinger]

Do you want to see my solution to this at all? It would have the added benefit of making requests cancellable in the future if that were ever desired. Also, it keeps additional timeout logic out of the stream classes, which partially clashes with my stream handler requirements (though, that can be worked around). And I intend to address the DNS issue as well. That's not (yet) included in your global timeout.

I'll need another day or two to get it ready.

Edit: Never mind, canceling doesn't require event pipes and is already implemented.

[yhirose]

Do you want to see my solution to this at all? It would have the added benefit of making requests cancellable in the future if that were ever desired. Also, it keeps additional timeout logic out of the stream classes, which partially clashes with my stream handler requirements (though, that can be worked around). And I intend to address the DNS issue as well. That's not (yet) included in your global timeout.

I'll need another day or two to get it ready.

Edit: Never mind, canceling doesn't require event pipes and is already implemented.

Currently, I am not very much interested in it unless it is absolutely needed. But if there are things that improve the httplib code base, could you split your change into several smaller pull requests? I usually don't accept a big pull request, because I don't have time to understand such.

As for the DNS issue, could you take a look at #1601? If you have a multi platform solution, could you send it as a pull request? Thanks!

[falbrechtskirchinger]

Do you want to see my solution to this at all? It would have the added benefit of making requests cancellable in the future if that were ever desired. Also, it keeps additional timeout logic out of the stream classes, which partially clashes with my stream handler requirements (though, that can be worked around). And I intend to address the DNS issue as well. That's not (yet) included in your global timeout.
I'll need another day or two to get it ready.
Edit: Never mind, canceling doesn't require event pipes and is already implemented.

Currently, I am not very much interested in it unless it is absolutely needed. But if there are things that improve the httplib code base, could you split your change into several smaller pull requests?

I've thought more about it and my code no longer adds anything of value to the core of httplib.

I usually don't accept a big pull request, because I don't have time to understand such.

Well, I hope you might accept a larger PR (WebSockets) as part of example/ or possibly contrib/.

As for the DNS issue, could you take a look at #1601? If you have a multi platform solution, could you send it as a pull request? Thanks!

Unfortunately, I was only interested in the issue as a vehicle to get WebSockets-related code included/reviewed. Sorry! I use httplib mostly for IPC using UNIX domain sockets and right now WebSockets for interacting locally with Chrome DevTools.
I would have just used threads. It's not great, but it works everywhere.