Name clashes between arguments and memoising procedure

[egnha]

The problem: Argument names of a memoised function (say, f) can be hijacked by names in the body or enclosing environment of the memoising function (memoise(f)).

Example:

memoise(function(hash) hash)("Am I OK?")
#> [1] "9ec45ba0998d8bc3150c..." (output truncated)

This is not a problem that is likely to be encountered for exotic names, like `_f`. For ordinary syntactic names, like hash or args, the likelihood of a name clash is still very low, but not zero. Nevertheless, it might be worthwhile to reduce this likelihood to zero, as this seems possible with a few small changes to memoise() (and, accordingly, has_cache()).

A solution:

1. To fix clashes with names in the enclosing environment — invoke such names by explicit reference to bindings in the enclosing environment.
2. To fix clashes with names assigned in the memoised function body — call the underlying (non-memoised) function f in the calling environment, rather than in the function's execution environment, as currently done.

Implementing 1) amounts to changing `_f` to encl$`_f`, etc., where encl gets parent.env(environment()). Implementing 2) amounts to replacing .init_call in memoise.R with eval.parent(`[[<-`(match.call(), 1L, encl$`_f`)).

I have implement such fixes in commits 018ef0a, 7a46d42, e32fb1b. memoise() is even a bit simpler, now, because bquote()'ing is no longer necessary. All existing tests pass, in addition to a test that verifies the absence of name clashes.

With such changes, argument names won't be hijacked by internal names:

memoise(function(hash) hash)("OK!")
#> [1] "OK!"

[jimhester]

Thanks, this is now fixed, see b0da939 and your previous commits.

[egnha]

Thanks for the merge and for fixing the test (7bd7b75).