ByteString: More efficient IO read/write

[konsumlamm]

Implement hPut, hGet, hGetContents using block IO and change readFile, writeFile, appendFile to use them.

Implements #112.

The code currently compiles, but it doesn't work, since primBS2FPtr, primFPtr2BS & primFPtrLen2BS aren't implemented correctly (_primitive "I" doesn't work, since while ByteString and ForeignPtr have essentially the same representation, they have different tags). I'm not sure what the best way forward is here. Ideally there would be conversion between bytestrings and forptrs without having to copy the payload, but I'm not sure that's possible. Another idea is to add mallocByteString and withByteString (so that we don't have to go through ForeignPtr), but that would mean code duplication.

[augustss]

I think the right thing is to use the same tag for both. The tags only differ because of serialization. But that's better done with some subfield of a foreign pointer. I think it's a straightforward change. I'll do it later today.

[augustss]

I've pushed a change that gets rid of T_BSTR.

I'm on vacation in New Zealand, so responses are a bit slow. :)

[konsumlamm]

Unfortunately, I think this just moved the problem. It's still not possible to simply convert between a foreign pointer and a bytestring using I, since the latter needs the fptype to be set to FP_BSTR. Looking at addForPtr, I think something similar should work here, but since the foreign pointer and the bytestring need to share the finalizer, the tag can't be stored there. I think it would've been possible before, but maybe I'm missing something.

[augustss]

It can probably be fixed by just removing the sanity checks. The fptype doesn't matter, except for serialization.

[konsumlamm]

It can probably be fixed by just removing the sanity checks. The fptype doesn't matter, except for serialization.

Ok, I'll try that.

[konsumlamm]

I added two new primitives, fp2bs & bs2fp. fp2bs takes a foreign pointer and an int and returns the pointer as a bytestring (setting the fptype and the size), while bs2fp simply returns the bytestring forptr unchanged. I'm not entirely sure if this is sound, but at least I haven't encountered any segfault yet.

[augustss]

Just setting the type to bytestring isn't sound unless the pointer was originally a bytestring. A bytestring has a size that is used in serialization. The serialization code needs a sanity check for NOSIZE.

I'll look at it some more tonight.

[konsumlamm]

Just setting the type to bytestring isn't sound unless the pointer was originally a bytestring. A bytestring has a size that is used in serialization. The serialization code needs a sanity check for NOSIZE.

fp2bs also sets the size.

[augustss]

Having fp2bs and bs2fp is a good idea.