Ignore masked references to FileAttachment.

[mbostock]

We currently consider all calls to FileAttachment to be file references, even if the FileAttachment symbol is masked in the current scope. I feel this is overzealous and inconsistent: if we don’t consider the reference to FileAttachment to be an external reference (i.e., a cell input), then we shouldn’t consider the call to be a file reference, either.

This does mean that you can circumvent the static enforcement of file references:

(FileAttachment => FileAttachment("alphabet" + ".csv"))(FileAttachment)

But that’s already possible today:

(f => f("alphabet" + ".csv"))(FileAttachment)

Deferring review request until Monday. 🤙

[j-f1]

This does mean that you can circumvent the static enforcement of file references:

Would it make sense to forbid the use of the FileAttachment identifier anywhere outside of a CallExpression’s callee then?

[mbostock]

Would it make sense to forbid the use of the FileAttachment identifier anywhere outside of a CallExpression’s callee then?

I don’t see that being necessary (or particularly helpful). The goal isn’t to strictly prevent dynamic invocation, but to have predictable, consistent semantics about what a “well-formed” file reference looks like.