Date: Fri, 8 Mar 2019 09:20:30 -0300
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Pavel Stehule <pavel.stehule@gmail.com>
Cc: Chapman Flack <chap@anastigmatix.net>,
	Markus Winand <markus.winand@winand.at>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: PostgreSQL vs SQL/XML Standards
Message-ID: <20190308122030.GA8440@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: 
 <CAFj8pRCtkpkWusVUw0MuAYA8Yz9zSxjFeCQAQoqVJbk9L2FJxA@mail.gmail.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On 2019-Mar-08, Pavel Stehule wrote:

> looks like error in xmlXPathCompiledEval function, that produce little bit
> broken result for XML_DOCUMENT_NODE type. I hadn't this problem with
> libxml2 2.7.6 64bit, but I seen this issue on same version on 32bit.
> 
> Currently I had not fresh 32 bit system to check it.
> 
> I found a workaround - in this case copy (and release xmlNode) is not
> necessary.
> 
> please, apply attached patch.

Wow :-(  At this point I'm wondering if this should be put in back
branches as well ... I mean, distill part of commit 251cf2e27bec that
doesn't affect the behavior of text nodes, and put it on all branches
together with your fix?

Another thought: should we refuse to work on known-broken libxml2
versions?  Seems like this bug could affect other parts of code too -- I
see that xmlXPathCompiledEval() is called in file places (including two
in contrib/xml2).

Third thought: an alternative might be to create a wrapper for
xmlXPathCompiledEval that detects NULL content and fills in a pointer
that xmlFreeNode can free.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services