Adam Jackson <a...@nwnk.net> writes: > ISTR the server currently has a "write then read" strategy for deadlock > avoidance (and accidentally better reliability when low on memory). Do > we want to do that here too, and if we do, do we split this into > walking the list twice to enforce the order, in case a caller has both > flags set?
There's no strong requirement, other than the server isn't allowed to block on writes. I think what you're remembering is that X clients are kindly requested to read when writes would block. Writing before reading encourages deadlock in general, but does reduce buffering requirements. We could enforce write-before-read ordering here by removing the read flag from the mask when the write flag was set so that subsystems using both read and write wouldn't wake while waiting for write space. However, given that one use of this function is for nested X servers where the socket in question is an X client connection which requires read-before-write behavior, I think it's pretty clear that we'll want to leave this kind of policy decision to the subsystem using the fds. -- -keith
signature.asc
Description: PGP signature
_______________________________________________ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel
