Trampolines not created for summarized methods

[khatchad]

We've been working on wala/ML#127, but I'm beginning to realize that this problem isn't specific to the Python front-end. We should have this problem for every summarized instance method, regardless of language.

We'll keep wala/ML#127 open, because, in that project, there is quite a bit of injection into the IR that goes on for Python, so it may make sense to keep that up to workaround this problem on that end. But, I opened up the issue here because I believe that there is a fundamental problem with how WALA handles summarized instance methods.

[khatchad]

From a simple search, it seems that core WALA uses trampolines to execute lambdas. There may also be some other cases, but lambdas seem like the main one.

[khatchad]

We implemented a workaround for this in wala/ML#127 (now closed) by injecting trampolines on the Python front-end side. The fix works but revealed a follow-up discrepancy: the artificially generated trampolines are missing super invocations that real (non-summarized) trampolines include (tracked as wala/ML#381).

The wala/ML#127 fix could serve as a reference for what WALA should do natively for summarized instance methods across all language front-ends. The TestIssue127 test class has paired tests comparing synthetic vs real trampolines (e.g., testIssue127e vs testIssue127eReal) that demonstrate the problem and verify the workaround. The corresponding XML summaries are in issue127*.xml.