The test generated warnings which were not checked. The XML file was pointing to the wrong assembly and the linker raised IL2072 warning. Instead, the test verified the suppression of warning IL2067, which was not raised by the linker in the first place.

Co-authored-by: Jeremi Kurdek <[email protected]>

The new default for trimming is equivalent to TrimmerDefaultAction=link.
Since we're trying to stay away from the term link and TrimmerDefaultAction
isn't a very clear name, this change repurposes TrimMode.

The two new TrimModes are 'full' and 'partial'. The new default is equivalent
to 'TrimMode=full'. These new modes are also recognized for earlier TFMs, but
the defaults are only changed for net7.0+.

Move CompilerGenerated* classes to the DataFlow directory. They sort of belong there anyway and they should be "shared" with AOT (and the entire DataFlow directory is "Shared").

Change how interprocedural state method tracking is initiated. For the parent method we track the "body", for everything else we add the method itself (and get the body inside the tracking). In AOT, getting a body of a method is more complicated and so having it in one place helps with factoring. Linker doesn't care either way so I made it symetric with AOT.

InterproceduralStateLattice in AOT has state (IL Provider) so it has to be instantiated for each scanner separately. Made a symetric change in linker.

Adds a test for nested local functions and data flow.

…0708.7 (#2885)

[main] Update dependencies from dotnet/runtime

* Add a test for unsupported branch instruction crash

* Make the test pass for now

…ode doesn't handle it yet (throws NIE). (#2889)

…2894)

This also makes small tweaks to help with linker->AOT sync.

This is the Roslyn analyzer counterpart to
bc46e44.

It adds detection of captured locals, and conservatively analyzes
them with all values that it encounters in the method.  The
analyzer makes no attempt to warn on reflection access to
compiler-generated fields, because the compilation strategy for
these is not known by the analyzer.

Unlike the linker, this doesn't do conservative analysis for
hoisted locals in state machine methods. The analyzer already
understands control flow within state machines the same way it
does for normal methods, so we just continue using that behavior
here. We could consider improving the linker's analysis to match,
but it would require a deeper understanding of the state machine
methods.