Use XPath position() function instead of xsl:number, to evaluate array indices while generating code for easier readability.

API/ABI change: XCBSendRequest callers must pad to 4-byte boundaries now. When not in RAW mode, a null pointer for iov_base is replaced by up to 3 padding bytes.

Separate notion of request-completed from current-request, and mark requests completed more aggressively. Detects some usage errors that are otherwise undetectable.

Buffer a couple CARD32s on the stack instead of using an extra iovec. Also a bugfix: do not hold a reference to longlen after it goes out of scope.

Bugfix: how about *not* leaking all pending_replies when no reply matches, as often happens with Xlib?

Rename struct XCBReplyData to reader_list to follow my naming conventions and better reflect its purpose.

Replace current_reply generic queue with hand-written version. No generic queues remain so delete their implementation.

There's no more race condition between event and reply handling.
The *RequestBlind and *RequestChecked functions are not yet implemented.

Move _xcb_read_block to xcb_in.c and make it static. Change calls in xcb_conn.c to _xcb_in_read_block instead.

Move _xcb_readn to xcb_in.c and make it static. Minor change to _xcb_read_block to not depend on _xcb_readn.

Simplify: Always use writev. (In _xcb_out_flush, convert the output queue to a single iovec if needed.)

This requires dynamically allocating memory in XCBSendRequest, but this
malloc/free pair turns out to cause a 30% speed hit for the 'x11perf -noop'
test -- so for the moment I use alloca where available and fall back to malloc
on other platforms. Later I think I'll change the contract of XCBSendRequest
so the caller is responsible for memory allocation, because the caller ought
to always be able to stack-allocate here.

Add XCBGetQueuedRequestRead for Xlib that does no syscalls, just returns whatever XCB already knows about.

Minor performance fix: Change the contract on XCBGetRequestSent so that it does not waste time re-locking. This is for Xlib, and Xlib has already locked.

Move request_written update back where it was for now: doing it early can cause XCBWaitForReply to wrongly believe that the request has been flushed. Eventually, we should fix bug #6021.

Remove _xcb_assert_valid_sequence. One test is trivially true, and the other may be temporarily violated without anything bad happening.

Move _xcb_write and _xcb_writev to xcb_out.c and make them static, since only _xcb_out_write calls them.

More return value changes, and make _xcb_in_read_packet static since it is not called from outside xcb_in.c.