Our database representation of ASTs does not use assignment patterns, instead encoding the relevant information directly in the associated function/loop/assignment. We convert from an AST with assignment patterns to one without during parsing, so the extractor does not expect any assignment patterns to be present in the AST.
Due to a bug in the parser, this can currently happen for malformed programs. While we should fix that bug once it gets fixed in Acorn, it also makes sense for the extractor to be more robust, so this PR teaches the `ASTExtractor` pass to raise a parse error when it encounters an assignment pattern, and all other passes to simply ignore them.
The `queries.xml` file defines which extractor the `codeql test` runner will use
to extract databases for the tests. In the future one will be able to write this
information in `qlpack.yml`, but we can't do that immediately because the
_existing_ CodeQL tooling would refuse to parse a `qlpack.yml` that has the new
field in it.
Adding a queries.xml file means that the normalization of file names in the test
output changes even with the old QLTest, so there are a number of consequential
updates of expected output files.
Previously, we would consider an HTML file to be a dependent of all scripts embedded in it. Now we instead consider each JavaScript toplevel inside the HTML file to be a dependent, which is more sensible anyway.
These were originally meant to give you the term that is textually matched right before/right after the receiver. When I introduced support for lookbehinds, I changed the behaviour to give you the term that is _operationally_ matched before/after the receiver (remember that lookbehinds are implemented by reverse-matching).
However, I think that's rarely ever what you want, and is wrong for the only two uses of these predicates, where it's the textual matching order that we are after, not the operational order.
Consequently, I've changed the semantics back and updated the comments to hopefully clarify the intention.
