gh-115133: test_xml_etree.py: Fix for Expat >=2.6.0 with reparse deferral (fixes #115133)

[hartwork]

Fixes #115133

• Issue: test.test_xml_etree*.XMLPullParserTest.test_simple_xml fails with (system) expat 2.6.0 #115133

[serhiy-storchaka]

It rather defeats the purpose of the test.

The purpose of the incremental parsing is that you can process the XML file without reading all data and building the full tree. This test tests that you get the parsed data as fast as the corresponding input data is available. The modified test only tests that it is available after feeding all input data and closing the parser. But you do not need the incremental parsing for this.

[mgorny]

I wonder how this affects stuff like XMPP where you actually need incremental parsing for things to work at all.

[Snild-Sony]

This test tests that you get the parsed data as fast as the corresponding input data is available. The modified test only tests that it is available after feeding all input data and closing the parser. But you do not need the incremental parsing for this.

I'm not sure how to guarantee that one gets the data "as fast as the input", but it should at least be possible to assert that some parsing happens before close(). Would that be close enough?

I wonder how this affects stuff like XMPP where you actually need incremental parsing for things to work at all.

Probably not well, at least in general. With XMPP specifically, you may get lucky and always send a whole packet for parsing at once (or just generally have small tokens), in which case there will be no incomplete tokens and therefore no need for reparse deferral. But it will be very implementation-dependent.

For such a special usecase, this should work as a "flush" operation:

XML_SetReparseDeferralEnabled(parser, XML_FALSE);
XML_Parse(parser, NULL, 0, XML_FALSE); // or the last data of your "batch", if you know that ahead of time
XML_SetReparseDeferralEnabled(parser, XML_TRUE);

[hardfalcon]

In case it's relevant to anyone: Python 3.11.8, built with the patch from this pull request against expat 2.6.0, and also using expat 2.6.0 after installing said Python package, is working just fine for me, both for running Gajim (an XMPP client written in Python) and for running a matrix-synapse server.

[serhiy-storchaka]

Thank you for your PR, but it was fixed by #115164 which keeps some of more strict testing.