Make -o do something reasonable or remove it

[brson]

For libraries we ignore the filename completely. This would seem to imply that the flag should just specify the output directory. Locking in this strange behavior for Rust 1.0 would be unfortunate. Nominating.

[metajack]

We have --out-dir for libraries. Probably it would be could enough to fail the compile if --rlib, --dylib, or --lib appears along with -o.

[brson]

Sometimes rustc doesn't know whether it's compiling an exe or library until it reads the source code, so simply looking at the command line flags would be insufficient. Sometimes rustc outputs multiple files. My inclination is to always make the naming of the output file up to rustc and remove -o completely.

[thestinger]

Could we replace it with a --name switch to set the library or binary name?

[metajack]

@thestinger Would this override the name part of the crate_id? Ie, binary output would be ${NAME}.exe but library output would be lib${NAME}-${HASH}-${VERSION}.dll?

[thestinger]

Yes, in fact I thought that's already what -o did. It just needs the right switch name.

[brson]

or maybe --crate_id

[pnkfelix]

Okay, so the task here is to remove -o and replace the one case where it currently works with some cmdline option for overriding the crate_id, as discussed above.

[pnkfelix]

Accepted for P-backcompat-libs (aka tools)

[Florob]

FWIW, I'd love to have an option to specify the actual file name for libraries. My particular use-case is for writing Lua modules, which need to be a shared libraries with the module name as basename (i.e. having a "lib" prefix, including the hash, and including a version is undesirable).