Given the increasing use of long from typing import foo, bar, ... import sequences, it's becoming more desirable to address individual components of the import node. Unfortunately, the ast.alias node doesn't contain source location metadata (e.g. lineno, col_offset).

This metadata would be comparatively easy to add, backwards compatible, and would enable more precise diagnostics e.g. lints.

Ah and one other question: Is this normally the sort of thing that would get backported? It should be very straightforward to do so, at least for 3.9 given the support for the new parser.

I'm okay with this, unless someone has any input on why the alias node should not have line/column info (and did not have it in the first place).

Ah and one other question: Is this normally the sort of thing that would get backported? It should be very straightforward to do so, at least for 3.9 given the support for the new parser.

Unfortunately no, is technically a new feature.

New changeset 75a06f0 by Matthew Suozzo in branch 'master':
bpo-43798: Add source location attributes to alias (GH-25324)
75a06f0

FYI, probably unavoidable, but this appears to have broken pytest pytest-dev/pytest#8539

FYI, probably unavoidable, but this appears to have broken pytest pytest-dev/pytest#8539

What's the problem? alias objects *have* lineno argument:

>>> print(ast.dump(ast.parse("from x import y"), include_attributes=True, indent=4))
Module(
    body=[
        ImportFrom(
            module='x',
            names=[
                alias(
                    name='y',
                    lineno=1,
                    col_offset=14,
                    end_lineno=1,
                    end_col_offset=15)],
            level=0,
            lineno=1,
            col_offset=0,
            end_lineno=1,
            end_col_offset=15)],
    type_ignores=[])

@pablo: the problem (referenced in the pytest issue) is that ast.alias has new required arguments now (see also the adjusted test in
75a06f0#diff-3f516b60719dd445d33225e4f316b36e85c9c51a843a0147349d11a005c55937R1060-R1061). That's expected/wanted though (assuming that it should not have "defaults" for B/C), and there's a patch for the issue at pytest-dev/pytest#8540.

Ah, it was not clear from the issue that pytest was constructing nodes by hand. It makes sense now. Thanks!