- Users specifying extra tracing configurations may now use the `GetRegisteredMatchers(languageId)` Lua function to retrieve the existing table of matchers registered to a given language.

- The `Experimental` flag has been removed from all packaging and related commands.

- The RA pretty-printer omits names of internal RA nodes and pretty-prints

binary unions with nested internal unions as n-ary unions. VS Code extension

v1.11.0 or newer is required to compute join order badness metrics in VS Code

for the new RA format.

- The Python extractor will no longer extract dependencies by default. See https://github.blog/changelog/2023-07-12-code-scanning-with-codeql-no-longer-installs-python-dependencies-automatically-for-new-users/ for more context. In versions until 2.17.0, it will be possible to restore the old behavior by setting `CODEQL_EXTRACTOR_PYTHON_FORCE_ENABLE_LIBRARY_EXTRACTION_UNTIL_2_17_0=1`.

- The `--ram` option to `codeql database run-queries` and other

commands that execute queries is now interpreted more strictly.

Previously it was mostly a rough hint for how much memory to use,

and the actual memory footprint of the CodeQL process could be

hundreds of megabytes higher. From this release, CodeQL tries harder

to keep its _total_ memory consumption during evaluation below the

given limit.

The new behavior yields more predictable memory use, but since it

works by allocating less RAM, it can lead to more use of _disk_

storage for intermediate results compared to earlier releases with

the same `--ram` value, and consequently a slight performance

loss. In rare cases, for large databases, analysis may fail with a

Java `OutOfMemoryError`.

The cure for this is to increase `--ram` to be closer to the amount

of memory actually available for CodeQL. As a rule of thumb, it will

usually be possible to increase the value of `--ram` by 700 MB or

more, without actually using more resources than release 2.15.x

would with the old setting. An exact amount cannot stated, however,

since the actual memory footprint in earlier releases depended on

factors such as the size of the databases that were not fully taken

into account.

If you use the CodeQL Action, you do not need to do anything unless

you have manually overridden the Action's RAM setting. The Action

will automatically select a `--ram` setting that matches the version

of the CLI it uses.

This affected the human-readable summary produced by `codeql database analyze`

- Java 21 is now fully supported, including support for new language features such as pattern switches and record patterns.