C#: SSA refactorings

[hvitved]

SSA refactorings in preparation for a shared SSA library implementation.

The first commit removes the class ImplicitUntrackedDefinition, which was used to generate SSA definitions for variables not amenable to SSA analysis. The class was never actually used.

The second commit unifies how SSA definitions are constructed, by collapsing the existing TSsaExplicitDef, TSsaImplicitEntryDef, TSsaImplicitCallDef, and TSsaImplicitQualifierDef IPA injectors into a single TWriteDef injector. This has the benefit that only variableWrite() is used to construct SSA definitions (a predicate similar to that will be part of the API for a shared implementation), and a common liveness analysis is applied in all cases (though in some corner cases this is less precise).

Liveness of writes to captured variables, where reads happen inside capturing callables, is ensured by adding a new (pseudo) read of kind CapturedVarCallRead at calls that may reach a read. This means that the common liveness analysis can be applied here as well. For example, in

int M()
{
   var captured = 0;
   int Inner() => captured + 1;
   return Inner();
}

a pseudo read of captured is added at the call to Inner, which makes the write captured = 0 live.

https://jenkins.internal.semmle.com/job/Changes/job/CSharp-Differences/751/

[tamasvajk]

Found some minor typos, and added a couple of questions.

[tamasvajk]

implicitEntryDefinition covers captured variables, but the writes of captured variables might not be actually executed. So wouldn't it be more accurate to set certain to false?

[hvitved]

It doesn't matter, actually, as there cannot be any prior definitions.

[tamasvajk]

I think this case is for ref locals. Why do we need strictcount(v.getAnAccess()) > 1?

[hvitved]

Yes, it is for ref locals. The check is there to avoid SSA definitions for cases like this: https://github.com/github/codeql/blob/main/csharp/ql/test/library-tests/dataflow/ssa/Consistency.cs#L55. Before the refactoring, this was taken care of by https://github.com/github/codeql/pull/4923/files#diff-7a5016d3998d3aa9a71ee689d871333e9bffa227aae862f5d5183a131fdc4f40L1800-L1810, but since we are now using a common liveness analysis, I applied this (less precise) check instead.

[tamasvajk]

LGTM