[Tclxml-users] Re: -final 0 bug

[Mats B. <ma...@pr...>]

Ted Nolan wrote:
> You may recall that back in December, I had found a bug in using the
> pure Tcl version of tclxml to parse XML streams with the "-final 0" option.
> 
> I think someone suggested diking out the Rose patch for bug #596959, but
> there were others saying that would break something else, and there was
> no real resolution to the issue.

I believe that this patch is not correct, but I don't want to remove
it from cvs since this is bad practise. 
Look at the parts of code tagged with "Mats" where I cache unmatched
xml to be prepended to the next xml chunk. This also works for
arbitrary chopped off xml:  <stream><|...|junk var= 'u|...|ndef'/> ...
you get it.

	# This RE only fixes chopped inside tags, not chopped text.
	if {[regexp {^([^<]*)(<[^>]*$)} [lindex $sgml end] x text rest]} {
	    set sgml [lreplace $sgml end end $text]
	    # Mats: unmatched stuff means that it is chopped off. Cache it for next round.
	    set state(leftover) $rest
	}

You may check out my copy of TclXML included in my application at:
"http://coccinella.sourceforge.net"

One additional complication may arise if you get xml chunks from events
(socket) and there are any update in callbacks that triggers a new
parse operation. The xml parser is, of course, not reentrant, but
serialization must be enforced. See code below:

proc wrapper::parsereentrant {p xml} {
    
    variable refcount
    variable stack
    incr refcount
    if {$refcount == 1} {
	
	# This is the main entry: do parse original xml.
	$p parse $xml
	
	# Parse everything on the stack (until empty?).
	while {[string length $stack] > 0} {
	    set tmpstack $stack
	    set stack ""
	    $p parse $tmpstack
	}
    } else {
	
	# Reentry, put on stack for delayed execution.
	append stack $xml
    }
    incr refcount -1
    return {}
}

Mats