Python: Avoid bad join in AstExtended::AstNode::containsInScope

[hvitved]

I noticed this bad join order when running UnreachableCode.ql:

[2021-08-04 11:04:54] (3288s) Tuple counts for AstExtended::AstNode::containsInScope_dispred#bf/2@06c1ea:
                      485725      ~0%      {1} r1 = SCAN py_stmts OUTPUT In.0 'this'
                      485725      ~0%      {2} r2 = JOIN r1 WITH AstExtended::AstNode::getScope_dispred#ff ON FIRST 1 OUTPUT Rhs.1, Lhs.0 'this'
                      10971648538 ~0%      {2} r3 = JOIN r2 WITH AstExtended::AstNode::getScope_dispred#ff_10#join_rhs ON FIRST 1 OUTPUT Lhs.1 'this', Rhs.1 'inner'
                      5089967     ~0%      {2} r4 = JOIN r3 WITH AstExtended::AstNode::contains_dispred#bf ON FIRST 2 OUTPUT Lhs.0 'this', Lhs.1 'inner'
                      5046621     ~1%      {2} r5 = r4 AND NOT @py_scope#f(Lhs.1 'inner')
                                           return r5

https://jenkins.internal.semmle.com/job/Changes/job/Python-Differences/654/:

	D0(s)	D1(s)	D1/D0
g/openstack/nova	2054	2085	1.01509
g/django/django	1026	1033	1.00682
g/python/cpython	1777	1785	1.00450
g/ytdl-org/youtube-dl	437	438	1.00229
g/pallets/flask	607	603	0.99341
g/saltstack/salt	2008	1986	0.98904
g/FreeCAD/FreeCAD	5770	4405	0.76343

[yoff]

Thanks for the find and the solution! 💪

Out of curiosity: Is it correct that it might still choose to evaluate not inner instanceof Scope before this.contains(inner) and could we avoid that with binding directives?

[hvitved]

Out of curiosity: Is it correct that it might still choose to evaluate not inner instanceof Scope before this.contains(inner) and could we avoid that with binding directives?

That shouldn't matter much, what matters is that we get rid of the self-join of AstExtended::AstNode::getScope.