On Sun, Sep 28, 2008 at 9:09 PM, Chaddaï Fouché wrote:

2008/9/28 Casey Rodarmor :

...

The comparison I was thinking of was with python, where a module can be placed in an arbitrary directory, and then accessed using a relative path. If I have 'stuff.py' that contains class 'Foo', I can move it to 'hello/module/stuff.py', and then import it is hello.module.stuff.

And if you have something in hello/ it must refer to the same module by module.stuff instead. I'm not sure it's clearer...

Yeah, I can see where you're coming from.

Anyway the hierarchical modules in Haskell means that every code must import the same module with the same name which I find saner (though I could wish the import/export syntax would be more advanced to allow for shorter imports when I have a lot of imports in the same hierarchy part).

...

A python module doesn't need to know where it lives.

Well, it doesn't need to, but everyone else needs to, in Haskell, the module need to know where it lives, but everyone else don't care.

...

I often find myself working on computers where I don't have administrative privileges, which means I might not be able to install libraries in the 'right' places. This approach also makes it simple to create self-contained projects that can easily be moved from machine to machine, where only the local directory structure is important. Would there be any downside to this in haskell?

Note that if the modules are in the same project, they're all in the same hierarchy (at least that's how I see a project) and so their names are coherent with each other and with the relative directory hierarchy, there is no problem there. Now, I think your preoccupation has more to do with packaging several libraries you're not sure will be available on the final machine with your project, and reasonably, you don't want to put them all in the same base directory (yeah, it would be a mess). I would argue that Haskell being a compiled language, its distribution problematics aren't exactly the same as Python... If you're just distributing an application you presumably would just compile it on your development setup and then distribute the executable.

Very true, but I still really like source distributions, even for compiled languages. Even though it can mean headaches, a source distribution is sometimes more portable than a binary distribution. (Usually because a clever end user can fiddle a little bit and get it working, whereas an incompatible binary is a useless blob.)

But if you really want to distribute the code source in a self sufficient package (assuming GHC on the final machine I guess ?), you should use Cabal : it automates dependency checking, compilation and installation with several source directory and configurable target installation directory. It's pretty easy to use too (mainly I suggest you do copy/paste of a good project file on Hackage).

I'll definitely check out Cabal. I'm a student, and so I'm always developing on personal and school computers, as well coding with random project partners (who always seem to be running equally random OSs). And then, to top it off, I've got to turn in assignments without being exactly sure where the grader is going to run them. I'm *always* wanting to use some library in a project, and (apt-get|port|emerge) my_sweet_library is never an option :-( Thus my concern with portable source packages. :-) These days I usually write crazy makefiles that try to do their best to compile and link in all the libraries needed, wherever they might find themselves.

-- Jedaï