Testing whether Xlib threading support works in XORG-RELEASE-1branch...

[Keith Packard]

Around 4 o'clock on Mar 23, Roland Mainz wrote:

> BTW: A quick test with Xfree4.3.0 shows that even an application which
> only has one thread hangs in libGTK initalisation thanks to a deadlock
> in the Xinput extension library... it seems that the original authors of
> the threading support in Xlib assumed that a mutex can be recursively
> entered... ;-(

Yeah, I know -- I've broken this myself in several libraries, it's hard to 
test when you don't have any multi-threaded applications around.  Xine is 
multi-threaded and seems to work fine, but it's not Gtk+ and doesn't use 
Xinput afaik.

One good piece of news is that the threading stuff may be completely 
replaced by a model which has been "proven" correct in XCB (we've got a Z 
specification for the library and a formal proof for that).  So, when we 
replace the insides of Xlib with XCB (in process), at least the lowest 
level library bits will be much nicer looking and far more likely to be 
correct.  Xlib itself was hacked to support threading, and I've never 
liked how that went.

I'd love to test this stuff; we're trying to get an initial X.org release 
ready by next Tuesday though, and the amount of rock-fetching needed 
before then is rather large.

-keith


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
Url : http://freedesktop.org/pipermail/release-wranglers/attachments/20040322/93855289/attachment.pgp