RE: xml- named processing instruction eaten

Well, one small victory. I noticed the xml declaration was getting stripped at that step, so I prepended it to the output assuming some correlation and sure enough that cured my ill.
 


From: matt.garrish@bell.net
To: xproc-dev@w3.org
Date: Thu, 19 May 2011 19:05:09 -0400
Subject: xml- named processing instruction eaten






I ran into a strange bug I thought I’d throw out here.
 
I grabbed the new version of Oxygen that came out today, which is now using “xml-model” PIs for validation. In the misguided hope of being seen as cool for having the shiniest new PIs on the block, I updated a stylesheet in which I was adding the old “oxygen” PIs to use the new form (being too lazy to add them manually each time). As soon as I did that, the first processing instruction in my pipeline output got truncated to “<?xml-model?>” (calling calabash 0.9.32 from the command line on a WinXP box, not from in Oxygen).
 
After much sniffing around, the PI loses its value after a p:exec step is called. I dumped the output from that step directly to a file (modified the program to dual output) and compared it to the output from the pipeline (no additional following steps) and the raw output from the program is unaffected by this truncation; it’s only the data in the pipeline that exhibits the behaviour. I can only reasonably assume the truncating occurs when the data is read back, as a result, as I skipped the unwrap and the c:result also has the PI truncated.
 
The odd thing is it only occurs if the first PI after the xml declaration begins with “xml” (xml-model, xml-stylesheet, etc.), and only to the first PI. If I put any non-xml named processing instruction first, the xml-* PIs are unaffected (and neither is the leading non-xml PI).
 
Anyway, I don’t know what to make of this one at this point. If anyone has any insights, though, I’d appreciate it (back to “uncool” <?oxygen?> PIs for me, though).
 
Matt