Dataflow: Add documentation for language maintainers.

[aschackmull]

Consensus seem to have emerged to put this document in docs/ql-libraries/dataflow.

[yoff]

Some "limitations" that should perhaps be documented: Do not add abstract predicates to Node, the library extends this type and will fail to implement them. Also, it seems that making Node an abstract class creates a non-monotonic recursion.

[hvitved]

Excellent to have this @aschackmull 🎉

[hvitved]

Some "limitations" that should perhaps be documented: Do not add abstract predicates to Node, the library extends this type and will fail to implement them. Also, it seems that making Node an abstract class creates a non-monotonic recursion.

I can recommend having an internal abstract NodeImpl class for defining predicates by dispatch: https://github.com/github/codeql/blob/master/csharp/ql/src/semmle/code/csharp/dataflow/internal/DataFlowPrivate.qll#L19. This also allows us to help the optimizer by telling it that e.g. getEnclosingCallable is functional, if that is not obvious from the implementation.

[intrigus-lgtm]

I'm curious, what is an IPA type?
Googling did not give me an answer unfortunately.

[MathiasVP]

I'm curious, what is an IPA type?

It's jargon for an algebraic datatype: https://help.semmle.com/QL/ql-handbook/types.html#algebraic-datatypes.

[aschackmull]

Some "limitations" that should perhaps be documented: Do not add abstract predicates to Node, the library extends this type and will fail to implement them. Also, it seems that making Node an abstract class creates a non-monotonic recursion.

I'm not sure that this is an important detail about the library itself, rather given that Node is exposed to the user of the library, it is generally a bad idea to make it abstract or give it abstract predicates as users are likely to e.g. define their sources and sinks as subset of Node by subclassing. So this is just the general rule-of-thumb for QL libraries that public classes never should be abstract unless it is their main use-case to be extended.

[aschackmull]

I think all current comments should now be addressed.

[hvitved]

Small typo