x.py tidy forces `use path::{...}` to come after `use path::m::{...}`; makes commits noisy.

[pnkfelix]

(This may be a rustfmt bug, rather than a rustc bootstrap issue. But I just wanted to file it somewhere before I forgot about it.)

Today, if you have:

use rustc_middle::middle::cstore::CrateDepKind;
use rustc_middle::middle::cstore::LinkagePreference::{self, RequireDynamic, RequireStatic};

and you want to import rustc_middle::middle::cstore::CStore, you could write

use rustc_middle::middle::cstore::CrateDepKind;
use rustc_middle::middle::cstore::CStore; // <--- new line
use rustc_middle::middle::cstore::LinkagePreference::{self, RequireDynamic, RequireStatic};

But one might rightly think that's overly verbose; why not use a {...} block on the first use?

But if you try that, and write:

use rustc_middle::middle::cstore::{CrateDepKind, CStore}; // <--- changed line
use rustc_middle::middle::cstore::LinkagePreference::{self, RequireDynamic, RequireStatic};

then x.py tidy complains, and says it wants you to rewrite it as:

use rustc_middle::middle::cstore::LinkagePreference::{self, RequireDynamic, RequireStatic};
use rustc_middle::middle::cstore::{CrateDepKind, CStore}; // <--- changed and *moved* line

This seems wrong to me.

if I had to guess, I'd infer that the rustfmt algorithm is using a naive alphabetical ordering on the lines, rather than considering the higher level module tree structure. Looking at it from the view point of putting nodes closer to the tree's root first, everything for use path::{...} should come before use path::m::{...}, right?

[jyn514]

@calebcartwright could you move this to rustfmt's issue tracker?

[calebcartwright]

I don't have sufficient permissions on this repo to be able to transfer an issue out, but AFAIK this is operating as expected given that the Rust Style Guide that defines the rules rustfmt implements requires the sort order be ascii-betical

https://github.com/rust-dev-tools/fmt-rfcs/blob/master/guide/items.md#ordering-of-imports