Multiple emit targets don’t ignore -o option, contrary to the warning

[nagisa]

$ ls -alh
…
drwxr-xr-x 1 nagisa nagisa  26 Dec 22 11:32 src/
…
$ rustc src/lib.rs --emit=llvm-ir,llvm-bc,obj,asm,link --crate-type=rlib -o rdd
warning: ignoring specified output filename because multiple outputs were requested
$ ls -alh
total 136K
…
-rw-r--r-- 1 nagisa nagisa  86K Dec 22 16:41 liblib.rlib
-rw-r--r-- 1 nagisa nagisa 5.1K Dec 22 16:41 rdd.bc
-rw-r--r-- 1 nagisa nagisa  17K Dec 22 16:41 rdd.ll
-rw-r--r-- 1 nagisa nagisa 9.7K Dec 22 16:41 rdd.o
-rw-r--r-- 1 nagisa nagisa  16K Dec 22 16:41 rdd.s
drwxr-xr-x 1 nagisa nagisa  26 Dec 22 11:32 src/
…

Although compiler warns that output filename was ignored, only the link target ignores it. All other targets generate file with provided output filename.

[thorncp]

6e7968b suggests that this behavior is intended. Specifically:

The -o option has also been redefined to be used for all flavors of outputs.
This means that we no longer ignore it for libraries.

I think the warning message is misleading, and is probably not even necessary. What do we think about removing this warning entirely, and updating rustc -h and the man page to reflect this behavior?

[nagisa]

That commit message contradicts itself further down (and still doesn’t match the current behaviour):

If the -o flag is specified, and only one output type is specified, the
output will be emitted at this location. If more than one output type is
specified, then the filename of -o is ignored, and all output goes in the
directory that -o specifies.

The emphasis mine. What’s the expected behaviour here @alexcrichton?

[brson]

I don't recall what was intended but it does seem that the current behavior is inconsistent and the warning message is wrong.

Based on the current behavior my preference is that if there are multiple output types that we name them all based on the provided file stem, meaning just change the library output to conform to the existing behavior.

@alexcrichton What do you think?

[nagisa]

This is possibly a good time and place to brainstorm about this, since this is one of the cornerstones of compiler usage.

This is what sounds most sensible to me:

• --out-dir never ignored;
• -o is alias for --crate-name.

$ # This is what it would look like
$ rustc --crate-type=rlib --out-dir=target/ -o foobar test.rs --emit=[everything] && ls target/
foobar.o foobar.s foobar.ll foobar.bc libfoobar.rlib

It does not depend on any combination of flags and works predictably (though very incompatible to other compilers) well.

[alexcrichton]

My thinking is the same as @brson's, the intended operation is something like:

// If foo.rs only generates one file, it will be called precisely `foo`
$ rustc foo.rs -o foo
// Use `foo` as a filestem for all outputs, ignore the precise filename "foo"
$ rustc --emit=[lots] foo.rs -o foo

The --out-dir options always takes precedence and will ignore -o entirely if specified, I believe. The major purpose of the -o flag is to compile executables such as rustc foo.rs -o bar.

[nagisa]

I submitted a rust-lang/rfcs#595 regarding this.

[wthrowe]

Since we can't actually change this behavior anymore since it would be breaking, would a pull request that just gets rid of the incorrect warning be accepted?

[brson]

Yes, let's change this so that, only if link is one of the requested, it says: "warning: ignoring output name requested with -o for "link" output because multiple outputs were requested".

In Rust 2 we should consider changing it so all output types are affected by the rename.

[Manishearth]

Willing to help anyone who wants to work on this

[hgallagher1993]

Ya I'd take it

[brson]

@hgallagher1993 You got it!

[brson]

@hgallagher1993 How's it going?

[hgallagher1993]

@brson I actually haven’t had a chance to even look at it, long hours in the office all last week. It's going to have my undivided attention for the next few days 😃 I'm probably going to need a bit of this as well @Manishearth

[brson]

OK, thanks for the status update!

[hgallagher1993]

@Manishearth I've been looking at this task and I was just wondering could I get some direction on how to start this...still completely new to the compiler 😄

[Manishearth]

So the warning is here:

rust/src/librustc_driver/driver.rs

Line 1275 in 6cc49e5

sess.warn("ignoring specified output filename because multiple outputs were \

RIght now we warn if the number of unnamed output types is more than one. The actual condition is "if the number of output types is more than one AND if one of the unnamed emit options is link"

The emit options are stored in a hashmap in sess.opts.output_types. The current code just looks for hashmap entries that exist but have a None value. What we need to look for is an entry where the key is Exe and the value is None.

Something like && sess.opts.output_types.get(Exe).is_none in that if should fix it. You will need to update the text to that from #20130 (comment) too.

[kjaleshire]

I believe this issue is resolved. See #38840 .