evntsqd.m

[Jamey Sharp]

On Tue, Dec 20, 2005 at 09:34:36AM -0500, Stuart Anderson wrote:
> Jamey,
> 	There are a few people around that have some experince w/ this
> test suite,but they are only on this list, and not monitoring the
> bugzilla, thus part of the reason for having a different system here.

Sure, that's reasonable.

> Please start a seperate thread on the list for each of these other bugs,
> and let's hope that not everyone has checked out for the holidays yet.

I'll do that shortly.

> So backing up a step or two, you are proposing a change to the strategy
> which is documented in the .m file. Discussion should occur at that level
> first, and once the strategy is agreed upon, then the actual code change.
> For simple stuff, proposing both parts at the same time is probably
> fine.
> 
> At a minimum, the strategy "pseudo-code" should be updated as well.

Is it really a strategy change to first check whether the test can
possibly produce useful results? Maybe I don't understand the scope of
the strategy texts.

> Does any one have any objections to the proposed change? or any
> suggestions how this could be handled w/o loosing the test? Does
> anything break by Xlib not actually queuing anything?

In my opinion the fundamental implementation problem is that the
XTestDiscard function makes some assumptions about the implementation of
Xlib that aren't justified by the spec -- notably, that output buffers
can have stuff deleted out of them. I think it should be perfectly
legitimate to, say, implement Xlib using fdopen to get stdio buffering
instead of having to implement it internally. AFAIK the same problem
would arise on such an implementation.

The test itself remains valid: there's still client-side buffering even
in Xlib/XCB or the stdio example, and that buffer isn't being flushed by
the call to XEventsQueued, so there are no side effects on the server
until the XSync call.

Does a second connection help here? Can you reliably check for the
server side-effects from a different client?

> Is a spec change needed to clarify that XEventsQueued() may never see
> anything in the queue?

Note that XEventsQueued is behaving exactly as always. It's just that
XTestDiscard has no effect on my version of Xlib.

--Jamey
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.x.org/archives/xorg-test/attachments/20051220/48a063a7/attachment.pgp