Make --crate-file-name obey --crate-type

[JustAPerson]

Before, the --crate-file-name flag only checked crate attributes for
possible crate types. Now, if any type is specified by one or more
--crate-type flags, only the filenames for those types will be
emitted, and any types specified by crate attributes will be ignored.

[alexcrichton]

I'm a little confused by this description, Here's what I get today:

$ cat foo.rs
#![crate_type = "dylib"]

fn main() {}
$ rustc --crate-file-name foo.rs --crate-type=rlib 
libfoo-aab0fa9e-0.0.dylib
libfoo-aab0fa9e-0.0.rlib

This makes me think that crate-file-name is reading both the compiler flags and the attributes (hence I'm confused about the first sentence of the PR description).

What's the intended behavior you're looking for here? I wouldn't want --crate-file-name to only look at the compiler flags when the rest of the compiler looks at both. Having compiler flags override attributes was brought up in #11573.

[JustAPerson]

Oh my. I am mistaken there. I was adding some debugging statements to the relevant code and when I tested it with both flags + attributes I thought I only saw the filenames from the attributes. I think I accidentally just used a flag for a crate type that was already specified by attribute. Sorry about that :(

My intent is the same as issue #11573. I was motivated to attempt this because when writing a Makefile for a project, I noticed that after updating a library it began to build twice unnecessarily. A new crate_type attribute had been added, and now --crate-file-name returned two names, causing make to generate two dependencies where there had previously only been one. I at first filtered only the filename I wanted, but later remembered the --crate-type and wanted to see if that would solve my problem. I had the same expectations expressed in that issue.

[alexcrichton]

I do agree that implementing #11573 would make makefiles a little easier, and would make compilation probably a little more intuitive.

Fixing the bug would likely involve modifying collect_crate_types and then updating the relevant documentation about linkage. For now though, I don't think that we want to modify just crate-file-name and not compilation, so I'm going to close this.

Feel free to reopen if you're willing to take a stab at #11573 though!

[JustAPerson]

I have modified both collect_crate_types and the relevant documentation. This addresses both my original concern and #11573.

You can see my changes here:
https://github.com/JustAPerson/rust/compare/crate-file-name
(the diff is slightly more complicated than necessary due to indenting some of the code).

I don't think that I can reopen this pull request myself, and it appears that GitHub is not registering the new commits to my branch here because this is closed. At your discretion, I can either update the original post for this pull request once reopened or create a new one.

[alexcrichton]

That looks good to me! Can you add a test for this change as well? You'd probably want to add a run-make test in the src/test/run-make directory.

[JustAPerson]

@alexcrichton,
Hiya, I've added a run-make test now. Sorry for the delay!