Rustpkg isn't checking for freshness of crate files that aren't main.rs

[tiffany352]

When I go to build something with rustpkg after modifying any file other than main.rs, despite main.rs having mod foo; where the file I'm editing is foo.rs, it doesn't check if that file has been updated when I build.

$ rustpkg --version
WARNING: The Rust package manager is experimental and may be unstable
rustpkg 0.8-pre (4c6bf48 2013-09-17 20:35:54 -0700)
host: x86_64-unknown-linux-gnu

[clonejo]

Still present in

$ rustpkg --version
WARNING: The Rust package manager is experimental and may be unstable
rustpkg 0.9-pre (e0c01ca 2013-11-02 12:11:18 -0700)
host: x86_64-unknown-linux-gnu

Same issue for using rustpkg build with lib.rs, also rustpkg test with test.rs.

[sfackler]

I've been looking into this and it looks really painful for rustpkg to actually check the file modification times. These files are actually loaded by the parser (!) and there's no record saved of what files were used to build the AST for a crate.

Rustpkg could walk the AST and duplicate the parser's logic to track down those external modules, but that seems sketchy, especially since I don't think it's possible to determine if the current module was an external or internal one at that point. What I think makes more sense is to hash the AST itself and compare that to the old version. This is nice since it takes care of #6522 as well.

[huonw]

Doesn't the parser store the file from which things came as part of the span?

[sfackler]

Ah, it does look like you can get a filename out of a Span and the CodeMap in the Session. You would have to be careful to avoid the synthetic files like "<std-macros>". I do kind of like using the AST hash since it doesn't require a recompile of the library if you only changed tests for example.

[alexcrichton]

rustpkg is gone.