Don't distribute stage1 artifacts

[alexcrichton]

For what I mean by "stage1 artifact", see this comment. The gist of it is that we don't need to distribute things built by the stage1 compiler, only things built by the stage2 compiler.

The tricky part will be figuring out where the host stage2 artifacts go. We could place them in $prefix/lib, but there's a danger of pollution. We could also only place them in $prefix/lib/rustlib/$host/lib, but then there's a danger of the rustc executable not being able to load these libraries seamlessly via rpath.

Solving this should deal with #13421, #11195, and #13733

[alexcrichton]

cc #14650

[alexcrichton]

This will also cut down our distribution size by 60MB

The FHS says:

Applications may use a single subdirectory under /usr/lib. If an application uses a subdirectory, all architecture-dependent data exclusively used by the application must be placed within that subdirectory.

Also observed by gcc, clang, ghc, mono, ocaml, and probably everyone else.

Another way would be to rename the base libraries to libfoo-stageN-... and shovel everything in $prefix/lib.

[alexcrichton]

cc @brson, maybe a good thing to tackle when dealing with the rust/cargo combined installer as well.

[alexcrichton]

cc #16029, this would fix that issue as well.

[pnkfelix]

Cc me

[oblitum]

cc #17289, also to be fixed by this.

[brson]

@alexcrichton Putting them in $prefix/lib doesn't seem worse than the situation now, where we are already using that directory. The only concern I see there is if we were to appropriate rustlib as a sort of GAC that cargo also uses to install libraries, in which case we probably would not want to just throw them into $prefix/lib, but split the rust host libs across $prefix/lib and $prefix/lib/rustlib.

FWIW, I expect this to be solved as part of #11145, but restructuring the bootstrap seems pretty far out of scope at the moment. @alexcrichton are you thinking that we would sort of reorganize the artifacts on installation, but not for the bootstrap?

[alexcrichton]

Yeah I do think that #11145 would solve this if we tackled it. I suppose my worry now about $prefix/lib is that lots of libs are installed there, and if we added a usual -L to it we may pick up some extra ones unintentionally.

I suppose we could reorganize for installation but not for bootstrap, but I do agree that it would be a little confusing. If we were to do this near-term, I would propose something like:

1. Replace all stage1 artifacts with stage2 artifacts.
2. Remove all stage2 artifacts in $prefix/lib from $prefix/lib/rustlib/$target/lib.
3. Implicitly add $prefix/lib to the set of -L paths.

This would allow us to deduplicate our distribution strategy (only distributing one set of artifacts) and would be roughly equivalent to what the bootstrap process looks like today, just not exactly so.

[steveklabnik]

Traige: as far as I know, this is still an issue today.

[oblitum]

cc #16029

[alexcrichton]

We've changed enough about our distribution that I don't believe this is relevant any more, so closing.