Evaluation did not show any improvements, so this is a clean-up more than an optimisation...

I don't understand the rationale behind this. Can you please highlight what problem this solves?

Doing a quick eval of superflous on django, the first result I got looks like something we would want to keep no matter what.

I don't understand the rationale behind this. Can you please highlight what problem this solves?

Now that we are using the variable capture library, these assignments are modelled twice:

• as regular assignments by our standard modelling
• as stores into the capturing heap by the variable capture library

this is at best unnecessary and at worst a performance problem (the DCA run indicates that it is mostly the former, but it could be the latter in pathological cases).

Doing a quick eval of superflous on django, the first result I got looks like something we would want to keep no matter what.

They should all be relevant and something we want to keep modelling. But once should be enough.

Now that we are using the variable capture library, these assignments are modelled twice:

• as regular assignments by our standard modelling
• as stores into the capturing heap by the variable capture library

ok, so for the code below you want to remove the local flow from 1 to x, since flow for x will be handled by the variable-capture library.

Do we have cases where this also affects correctness of our analysis? that is, without this change I assume we would have flow from 1 to the last use of x -- will that still be the case?

def outer():
    x = 1
    def inner():
        nonlocal x
        x = 2
        print(x)
    print(x) # <-- flow from 1 to this x is ok
    inner()
    print(x) # <-- flow from 1 to this x is not what happens in execution

AHA! I had forgotten the detail that our simple local flow step is defined as:

so there are "new" local flow steps added through the variable capture library; do these cover all the flow we had before? -- did you make any queries to verify that we did not lose any flow?

(what I could see happening is that if we lose local flow, our type-tracking won't be as good, which will certainly be a problem)