import-annotation=no not working for subobjects

[D-Groenewegen]

Setup

Current SMW sandbox:

• SMW version: 4.1.2
• MW version: 1.35.11
• PHP version: 7.4.33
• DB: 10.5.19-MariaDB-0+deb11u2
• Elasticsearch 6.8.23

Also tested on MW 1.39.2 / SMW 4.1.1 / PHP 8.0.30 (litespeed) / MariaDB 10.6.16-MariaDB-0ubuntu0.22.04.1, which is what led me to reproduce the issue on SMW's Sandbox.

Issue

The import-annotation parameter (https://www.semantic-mediawiki.org/wiki/Help:Plainlist_format/parameter_import-annotation) was introduced to either disallow (default) or allow annotations on a result template to be applied to the page holding the plainlist query.

But tests show that at least subobject annotations on the template are applied regardless (I didn't test #set). I have tried using true/false instead as well as leaving out the parameter, but the annotations are set anyway.

Reproduced

• SMW Sandbox

References

• Can Subobjects be set using format=template? #5522 - where interestingly, the default does appear to be working (for subobjects).
• Docu of "import-annotation" option is missing from #smwdoc for "template" and "plainlist" format #3564
• Support recursive annotation per query, refs #1255, #1068 #1257
• Move definition of ask param import-annotation to SMWQueryProcessor #2735

[D-Groenewegen]

Notes

I did a minor test and can confirm that at least the SMW::off/SMW::on markers are placed:

• ResultPrinter::handleNonFileResult(), where ( !$this->isHTML ) && ( $this->hasTemplates ), calls RecursiveTextProcessor::recursivePreprocess().
• If import-annotation=no, then RecursiveTextProcessor::recursivePreprocess() returns: '[[SMW::off]]' . $this->parser->replaceVariables( $text ) . '[[SMW::on]]' (and goes for the other option if import-annotation=yes).

And yet subobjects are being set. Why?

The problem here I think is that the act of calling replaceVariables() is enough to trigger something in the parser that causes them to be set regardless. It loos like it invokes braceSubstitution(), which in turns invokes callParserFunction() and I guess that ParserFunctionFactory is the re-entry point from MW into SMW.

The markers don't have any impact on them because the damage is already done.

P. S. Are the on/pff markers Parsoid-proof ? In #4727 from 2020, about the removal of the InternalParseBeforeLinks hook, MWJames stated that "context that relies on those markers will be affected". If so, a redesign seems warranted.