On Jul26, 2011, at 17:07 , Tom Lane wrote:
> Florian Pflug <fgp@phlo.org> writes:
>> What about the suggested upgrade of the elog(ERROR) in xml_errorHandler
>> to elog(FATAL)? Shall I do that too, or leave it for now?
>
> No objection here --- I had considered writing it that way myself.
> I refrained for fear of making a bad situation even worse, but if
> other people think FATAL would be the safer way, fine.
Patch attached.
pg_xml_init() now checks the error context after setting it, and
ereport(ERROR)s if the check fails. I've made it so that the errhint()
which suggest that compiling against libxml <= 2.7.3 but running
against > 2.7.3 might be the reason is only emitted if we actually
compiled against an older version. This is meant to avoid confusion
should there ever be another reason for that check to fail.
I've also changed the elog(ERROR) to elog(FATAL) in xml_errorHandler().
I've pondered whether to add a check to configure which verifies that
the headers match the libxml version exactly at compile time. In the end,
I didn't do that, for two reasons. First, there isn't anything wrong with
using older headers together with a newer libxml, so long as both versions
are either <= 2.7.3 or > 2.7.3. And second, with such a check in place,
the only way to exercise libxml's promised ABI compatibility is to upgrade
the libxml 2package after compiling postgres. That, however, is unlikely
to happen except on production servers, and so by adding the check we'd
increase the chance of ABI-compatibility problems remaining undetected
during development and testing.
best regards,
Florian Pflug