Consider moving dynamic_lib to extra

[brson]

My current guidlines for putting things in std is: 'does it have to go in std'? Usually this criteria is met because it is used by the runtime or because it logically belongs with other code that the runtime depends on. I don't think dynamic_lib meets this criteria at this point.

[bblum]

I mostly agree but its present use of atomically would have to turn into a comment-time assertion.

[mstewartgallus]

In the future I hope that some extra safety can be added by integrating dynamic_lib with Rust's name mangling system. As well, a case could be made for dynamic_lib belonging in the std::os module (once it's not unstable anymore.) Other than that, I see no reason for it to be in std.

[pnkfelix]

Visiting for triage from a long-long delayed triage email.

Sounds like there isn't strong support for this to remain in std.

Since its presence in std presents a backwards compatibility risk, nominating for P-backcompat-lang.

[flaper87]

I'll work on this. Moving it to libextra sounds like the first step out of std for dynamic_lib

[flaper87]

FWIW, the current parts of rust using it are:

Grep output:

src/librustdoc/plugins.rs:    priv dylibs: ~[dl::DynamicLibrary],
src/librustdoc/plugins.rs:        let lib_result = dl::DynamicLibrary::open(Some(&x));
src/libstd/unstable/dynamic_lib.rs:pub struct DynamicLibrary { priv handle: *libc::c_void }
src/libstd/unstable/dynamic_lib.rs:impl Drop for DynamicLibrary {
src/libstd/unstable/dynamic_lib.rs:impl DynamicLibrary {
src/libstd/unstable/dynamic_lib.rs:    pub fn open(filename: Option<&path::Path>) -> Result<DynamicLibrary, ~str> {
src/libstd/unstable/dynamic_lib.rs:                Ok(handle) => Ok(DynamicLibrary { handle: handle })
src/libstd/unstable/dynamic_lib.rs:        let libm = match DynamicLibrary::open(None) {
src/libstd/unstable/dynamic_lib.rs:        match DynamicLibrary::open(Some(&path)) {
src/libsyntax/ext/expand.rs:use std::unstable::dynamic_lib::DynamicLibrary;
src/libsyntax/ext/expand.rs:    let lib = match DynamicLibrary::open(Some(&path)) {
src/libsyntax/ext/base.rs:use std::unstable::dynamic_lib::DynamicLibrary;
src/libsyntax/ext/base.rs:    macro_crates: ~[DynamicLibrary],
src/libsyntax/ext/base.rs:    pub fn insert_macro_crate(&mut self, lib: DynamicLibrary) {

Is it ok to have things in libsyntax that depend on libextra ? I guess some things there will have to be cleaned up

[pnkfelix]

oh that's right, I guess the dynamic syntax extensions that landed recently probably depend on this.

Hmm. Now I'm not sure whether we can pull it out of std or not. We probably can.

[flaper87]

Perhaps we should revisit the API, clean it up and pull the module out of unstable, though!

[flaper87]

@pnkfelix so, what's the final decision here?

[pnkfelix]

@flaper87 I'll put it on the mtg agenda for Tuesday.

[pnkfelix]

@flaper87 we confirmed at the weekly mtg that it is fine for rustc to depend on another library external to libstd, and that we (by which I guess I mean "you" ;) ) should continue the effort to extract dynamic_lib out to a separate library.

As you are probably aware, we're in the process of decomposing libextra into a collection of separate crates. So it would probably be best to do that for dynamic_lib if we're going to do anything at all with it.

[pnkfelix]

Apparently stability attributes would prevent people from accidentally depending on this, even if it stays in libstd (which I suspect it won't).

P-low.

[treeman]

Visiting for triage.

Do we still want to move this out of std? If so, shall we make a new dynamic_lib library or move it somewhere else?

[steveklabnik]

I'd imagine it would have the same thing as semver and everything else: another crate in the rust-lang org.

(and I would vote to do that)