Track locals()

[jayvdb]

Fixes #136
Fixes #333

[jayvdb]

Note this also corrects handling of locals() which only sees variables preceeding it, while the current approach marks all variables as used. This is new test case test_unusedVariableAfterLocals in the patch.

[jayvdb]

this could be changed to

if not binding.used:
    binding.used = (self.scope, node)

Then at the end of the scope, any bindings used only by locals() could be a different type of message.

See #333 (comment)

[asmeurer]

IMHO locals is enough of a meta-programming technique that pyflakes should warn about it (same as import *). Otherwise locals() breaks all unused variable checks that would occur before it is called.

Having it warn but as a distinct error message that could be silenced by flake8 or whatever seems like the best solution.

[jayvdb]

I agree that locals() should be reported, but only if variables would otherwise be unused, like we did for * imports.
It is not meta programming when used only to capture running state , e.g. for logging. And def check_locals(): return locals()

I would prefer to tackle that new error report in a separate PR, and limit this one to only correct reporting of locals() with bug fixes where pyflakes was wrong or inconsistent.

[asmeurer]

What exactly is the import * change you are referring to? Did that behavior change recently?

[jayvdb]

No, I am referring to 1.1.0 change, 0532189

[asmeurer]

I see. I tested master and it sill warns about everything with import *, but sometimes has a better error message. Does this do the same thing for locals? The test makes it look like there are exceptions for locals, where using locals in the code makes it stop warning about unused variables. They aren't exactly the same thing because import * defines names and locals could both use or define a name (the latter being more metaprogramattic than the former).