Re: [Haskell-cafe] Records (was Re: [Haskell] Improvements to GHC)

On 11/17/05, Cale Gibbard wrote:

On 17/11/05, Sebastian Sylvan wrote:

...

On 11/17/05, Cale Gibbard wrote:

...

On 17/11/05, Sebastian Sylvan wrote:

...

On 11/17/05, Greg Woodhouse wrote:

...

Isn't there a potential for confusion with function composition (f . g)?

That being said, I like this idea (I just need to think it through a bit).

I've been wanting this for ages. It's SO much better than the current horribly broken records we have. There could be confusion with function composition, but there's no ambiguity (compositon have spaces around the dot, while record accessors do not). Personally I think that the dot is way to good of a symbol to be "wasted" on function composition. I mean, how often do you really use function composition in a way which doesn't obfuscate your code? I use ($) way more often than (.). Some people do use it more often than I do, but I find that in most cases except simple "pipelined" functions it only makes the code harder to read. I'd rather function composition was left out of the prelude alltogether (or defined as (#) or something).

Anyway. The current records system is a wart.

Actually, I didn't mention this in the other post, but why not the other way around? Make record selection (#) or (!) (though the latter gets in the way of array access), and leave (.) for function composition. Personally, I'd like something which looked like an arrow for record selection, but most of the good 2-character ones are unavailable. (~>) is a bit hard to type and looks wrong in some fonts. There's a "triangle" which is not taken, and isn't so hard to type (|>).

I never really understood the attachment to (.) for record selection. There's no reason that we have to make things look like Java and C.

This is going to be highly fuzzy and completely subjective. Here it goes.

I find that for selections (records, or qualified modules etc.) I want the operator to be small and so that the important "word groups" become the module or the record. When I read the following two variants myPoint#x myPoint.x

I think both of those look crowded -- smashing operator punctuation up against symbols basically never looks good to me. The right amount of spacing isn't generally available without proper typesetting, but a full space is a lot closer than no space at all.

Why not "myPoint # x" and "myPoint . x"?

Well, again this is just preference, but to me I'd like selectors to not have space between the record and the label, they still need to be "connected", but with a symbol which is small enought to help you easily see what's what.

...

I definatly prefer the latter. In the first one the operator is so large that it makes "myPoint" and "x" blend together as you read it (step away from the monitor and squint and you'll see what I mean), whereas in the second example the operator is small and makes the two operands naturally separate slightly when reading it, which makes it easier to tell which identifier is accessed. Also, it's certainly not a BAD thing if Haskell uses the same operators as other languages.

With function composition, though, the operator is just as important to identify when reading as the operands are. So I don't think a "big" operator is a problem there - likewise I have no problems with ($) being "large".

How about (¤)? It looks like a ring to me, I'm not sure where that's located on a EN keyboard, but it's not terribly inconvenient on my SE keyboard. f ¤ g looks better than "f . g" for function composition, if you ask me.

That symbol actually does look better, but isn't on any English keyboards to the best of my knowledge. I can get it in my setup with <compose-key> o x, but not many people have a compose key assigned. Also, this may just be a bug, but currently, ghc gives a lexical error if I try to use that symbol anywhere, probably just since it's not an ASCII character.

Hmm. On my keyboard it's Shift+4. Strange that it's not available on other keyboards. As far as I know that symbol means nothing particularly "swedish". In fact, I have no idea what it means at all =)

...

That's my subjective view on why the dot-operator is so darn nice, anyway.

Oh and to answer to your other post. I realise that function composition is a fundamental operation, but it's so fundamental that it's quite useless for most real-world cases unless your willing to seriously ubfuscate your code.

I disagree, there are plenty of cases where it's just what you want, and saves you from introducing a lambda term for nothing. This occurs very often in parameters to higher order functions. A simple example would be something like "filter (not . null)", or "any ((`elem` consumers) . schVertex)". More sophisticated examples come up all the time, and often the functions being composed have some parameters applied to them. I disagree that it's just for obfuscation.

Well, what I said was that for any more complex cases than that it ends up being hard to read. In a lot of examples when peopl abuse function compostion you really have to think for quite a while before you can figure out what it does, whereas if they had just written out everything in a few more steps it would've been immediatly clear. "filter (not # null)" isn't in any way worse than "filter (not . null)" IMO. Also, like I said earlier, I would even prefer a larger operator for compositon since in that case you want spaces around the operator (at least that's how I always use it in maths) and the dot is perhaps a bit too small in that case (easy to miss it on a glance, misstaking it for application). How about (<>) for the ring operator? That's almost a circle. "filter (not <> null)" looks pretty nice IMO. At any rate, I think that the dot is such a good symbol that I'd prefer to give it to a primitive operator (such as module or record selection) rather than something which is user-defined (but in the prelude). Plus, if the record system was halfway decent, I'm pretty sure record selection would occur a few orders of magnitude more often than function composition. So from a pure "popularity" standpoint record selection would win (assuming it's agreed that three uses for the dot operator is too much and one needs to go, that is). /S -- Sebastian Sylvan +46(0)736-818655 UIN: 44640862