Add -dev suffix to requirements on README

[cscherrer]

The readme page suggests installing libblas and liblapack. But there are no such packages! I assume you probably meant to have a -dev suffix on each of these. Yes?

[cartazio]

I think so :)
it also varies with the distro right?

[cscherrer]

Probably. But you say apt-get install, so I assumed you were thinking Ubuntu. I'd guess any Debian-based distro probably uses the same or very similar names, but I'm not sure.

[cartazio]

well, yeah, but i'm on a mac usually :)

if we're going to make it accurate, having a table of "distro names X pkg names to install" is probably the right way to roll

[cscherrer]

Looks like it's *-dev on Debian and *-devel on Redhat. Not sure about others.

[Rufflewind]

Note that some distributions don't make the distinction between dev and non-dev packages.
On Arch Linux, they are just blas and lapack.

[cartazio]

@Rufflewind you make a good point (hence my "we should list all the options if we're going to list some of them" stance)

I see you started your own blas binding! Can I invite you to contrib to hblas? (though if you have your own design, thats cool too). (i'll soon be giving a higher level api for hblas via my numerical project, wellposed/numerical-core is the core piece for now)

[Rufflewind]

I was writing a C binding generator in Haskell to handle Fortran call conventions and name mangling. It became redundant when I discovered the CBlas wrapper.

I'm currently porting some numerical code to Haskell, so I'd interested in writing FFI wrappers for Blas functions.

[cartazio]

@Rufflewind well, theres a lot of easy wrapping to be done on the blas front, youre welcome to help out, the pullrequests by @archblob are some decent examples of what those wrappers might look like.

If you're really comfy with lapack, i'd love some help figuring out how to give those nice high level wrappers that are still full powered.

I should also mention i'm amidst the release engineering for a nicer array layer that will interop with hblas, you can see it at https://github.com/wellposed/numerical-core currently

[Rufflewind]

I noticed you don't seem to be using hsc or c2hs to generate the FFI bindings. Wouldn't these tools make writing them a bit easier?

[cartazio]

@Rufflewind lets have this chat on a new ticket perhaps?
no one has convinced me that using either will make things easier, I still have to write down the name and the type signature either way! and then I immediately have a more complex configuration process. Right now theres zero configuration time logic