On 03/23/19 16:59, Tom Lane wrote:
> Unicode-code-point numbers. I removed that, made some other changes to
> bring the patch more in line with PG coding style, and pushed it.
Thanks! It looks good. I'm content with the extra PI checking being gone.
The magic Unicode-code-point numbers come straight from the XML standard;
I couldn't make that stuff up. :)
> > You're not really validating that the input
> is something that libxml would accept, unless its processing of XML PIs
> is far stupider than I would expect it to be.
Out of curiosity, what further processing would you expect libxml to do?
XML parsers are supposed to be transparent PI-preservers, except in the
rare case of seeing a PI that actually means something to the embedding
application, which isn't going to be the case for a database simply
implementing an XML data type.
The standard literally requires that the target must be a NAME, and
can't match [Xx][Mm][Ll], and if there's whitespace and anything after
that, there can't be an embedded ?> ... and that's it.
Regards,
-Chap