[Xcb] [PATCH 2/3] fix deadlock with xcb_take_socket/return_socket v3

[Josh Triplett]

On Fri, Jun 07, 2013 at 04:48:46PM +0200, Uli Schlachter wrote:
> On 05.06.2013 22:51, Jamey Sharp wrote:
> > I think this patch correctly does what it claims to do, which is always
> > a plus. :-) I haven't thought carefully about it, but it looks
> > plausible.
> > 
> > However, I have yet to see evidence that this patch fixes a bug. I
> > believe instead that it works around broken callers, which I'd rather
> > discourage.
> 
> Well, the only broken caller that was considered so far was "something calls
> into xcb via XGetXCBConnection() while the Display* is locked", right? You are
> saying that the caller shouldn't do this and has to be fixed instead.
> 
> However, how could such a fix look like? This rule would boil down to "you must
> not use XGetXCBConnection() in a library", because anyone up the call-chain
> could have used XLockDisplay(). Thus, XGetXCBConnection() and libxcb would
> become unusable for almost anyone (perhaps except for toolkits).

I'd argue that the fix belongs in the caller here, not the library:
"don't expect random X libraries to work when you call them with the
display lock held".  Though I do agree that such a change could break
code which worked with Xlib, which does seem problematic.

- Josh Triplett