Currently, it seems that Xlib is kind of deprecated in favor of XCB (which is automatically generated from the Xorg XML protocol file).
I’m doing some work involving low level X programming, and I was wondering if you intended to switch to the new format anytime like you did for core{graphic,audio} on Mac ?
The main advantages of XCB is lower latency (request & answer can be queued, and become asynchronous), good multithreading support (no more global XLockDisplay), smaller size & memory footprint, multi-touch support (no more work is done in XLib).
The main cons: Having to rewrite the calls from Xlib to Xcb.
I’d like to, though I’ve got very little experience of xcb so it’d be a bit of a learning curve. What’s the general availability of xcb these days? Is it there by default in most distros?
I can’t say for the pletora of linux distribution, but since Xorg 1.6 and later, it’s the library used to communicate with X server (in fact the entire Xlib is rewritten on top of xcb on Debian, Ubuntu, and Gentoo).
All new features will be in xcb only (multi-touch, advanced compositing, double buffering, protocol compression), and it seems most highlevel libraries are switching to xcb too (clutter, compiz for one).
The issue with X in general is the poor documentation for the API (Xlib is better than xcb because it’s older, you can find more sources).
I realize this is a very old topic, but this is what came up when searching for “JUCE” and “xcb”.
Given that - as mentioned before - Xlib is now basically a wrapper around xcb, a switchover could be done quite gradually. It’s easily possible to retrieve the underlying xcb_connection_t from a Display (using XGetXCBConnection. That means we can take a small piece of existing code, where we work with Xlib, retrrieve the xcb_connection_t, do our work with pure xcb and leaving the rest of the code untouched.
Would PRs to this effect have any chance of getting merged?
I think that the JUCE team’s time would be better spent adding Wayland support instead.
However, if you can provide a PR that has a relatively high impact on performance without too much of a testing burden then we can certainly consider it.
