Imported-via feature

[robinp]

For languages that have blanket-import-everything-from-module [1], the UI could indicate what are the actually used bindings at a blanket-import.

Reverse, given a usage, the UI could indicate (via highlight, and/or in backrefs) which import was used to bring it in scope.

See inspiration in google/haskell-indexer#100.

[1]: Haskell for sure. ML's open structure syntax? Java's import static foo.*?

[kamahen]

Python:

from os.path import *
print(sep)

In this, sep comes from the "import" line. Should the ref resolve to the "*" in that line or should it resolve to the place in file os/path.py that defines sep?

As a related issue:

import os.path.sep as os_path_sep
print(os_path_sep)

Should os_path_sep in the 2nd line resolve to the os_path_sep in the 1st line? From the language definition, the 1st line acts like a defines/binding (because it adds an entry to the variables' dictionary); but it could be argued that it's more natural when clicking on the 2nd os_path_sep to resolve that to the definition of sep in os/path.py; or it could be both.

UI design is hard (and so is schema design).

[robinp]

@kamahen this exact issue seems to be addressed by (see the exact comment) kythe/kythe#4021 (comment), and the more lengthy kythe/kythe#3934 (comment): the os_path_sep should

• both have a define/binding to a local variable/import, and then usage refer that one,

• and also a ref to the semantic entity of the os.path.sep module,

• and the two semantic entities (the local and the global module one) connected via aliases edge.

That one is pretty sure. Now, for the import * one, there's a tradeoff:

• The semantically nice thing is, I think, that the * anchor should do again the same (emit both defines/binding and a ref), but for all the imported names.

• But that might not be practical - if a module exports hundreds of names, but only one is actually used, why emit the other 99? Adds lots of noise. So for this feature to be practical, I believe the indexer should strive to emit only the used names.

(Note that #10 covers this usecase in more detail)

[robinp]

With all the new info about aliased imports, I now think this feature is deprecated, and the indexers should rather emit explicit edges for used imports even in blanket / implicit imports.