Hugs-Users

Download

hugs-users@haskell.org

November 2018

1 participants
2 discussions

Trex label names: leaky abstraction
by Anthony Clayden 09 Nov '18

09 Nov '18

Gaster & Jones 1996 says "We consider two rows to be equivalent if they include the same fields regardless of the order in which they are listed." [section 3.2] But this is tricky: Figure 6 shows there is an ordering on labels, so that in effect records are compiled to tuples with left-to-right ordering of label names alphabetically. The User Manual says "The order of fields in a record pattern *is* significant because it determines the order---from left to right---in which they are matched." In particular, a pattern match on a record with a field value `undefined` might succeed if some other field label is mentioned first in the pattern; but throw an error if it's the undefined field's label mentioned first. It's not difficult to expose this behaviour. Consider > import Hugs.Trex > > x5y = (x = 5, y = 'y'); x7y = (x = 7, y = undefined) > z5y = (z = 5, y = 'y'); z7y = (z = 7, y = undefined) > > x5y == x7y -- returns False > z5y == z7y -- throws error Prelude.undefined The tuples with label `z` are compiled to put that field second, regardless of the order of appearing in the expression. Then (==) applies by comparing field values in left-to-right alphabetic ordering. Field `x` is alphabetically before `y`; and the `x` values differ; so the comparison returns False without comparing the `y` values. Then in comparing the field `z` records, the `y` fields are compared first (ignoring that the `z` values differ), exposing that one `y` field is `undefined`. This is similar to comparing tuples > (5, 'y') == (7, undefined) -- returns False > ('y', 5) == (undefined, 7) -- throws error There seems to be some infelicity when building records: > f ... rho ... = (y = 'y' | rho) This throws strange errors depending whether `rho` includes labels that are alphabetically before `y`. Bug report incoming ... AntC

1 0

[haskell-beginners] Lifting over record syntax
by Anthony Clayden 04 Nov '18

04 Nov '18

musing on this https://mail.haskell.org/pipermail/beginners/2018-October/018363.html to which my reply talked mainly about Trex. Does it make sense to regard a Trex record (or any free-standing anonymous/extensible record) as a container, with values 'lifted' into the structure, and an interface given by the label names, _not_ by order of fields? That is, "container" in the Monad sense? Does it further make sense if all the values are already lifted into some Functor (like Maybe or (Either e) ) to hoist the Functor out over the record structure? I'm not sure I'm up with the program for Functor/Applicative/Monad everywhere. If a record structure is a Functor, what's the type constructor? `Rec`? If record structure is an Applicative, what's the constructor for `pure` that lifts into `Rec`? The `<*>` aka `ap` or bind for a Monad presumably is to take two records (with disjoint labels) and concatenate them to a single record. Given data Person = Person (Rec( name :: String, age :: Int)) we want, corresponding to the question: Person <?$> (age = 27) <*> (name = "Joe") in which operator <?$> is map-like: it takes a function to its left that expects a record of n fields, then maps over n singleton records with corresponding labels appearing in any order. I might be able to get there by generics/reflection over the field names embedded in data constructor `Person`s type, then overloading <?$> to look for each label. That is, if Hugs had any sort of generics/TypeRep. Label names in Trex are literals; there's no such thing as a label variable. (Which is why it's a tad annoying that they start lower case.) Furthermore the same label name must appear in both terms and types -- in fact labels occupy a namespace separate vs terms or types. So Trex is a long way from from generic record handling like: recAppend :: ( rho'\__x ) => Rec rho' -> Rec ( __x :: a) -> Rec ( __x :: a | rho') recAppend rho ( __x = x ) = ( __x = x | rho ) in which I've used double-underscore prefix to signify a label variable. This is intended to extend a record `rho` with a singleton record. If we try appending a record with more than one field, beware that field order is arbitrary, so this recAppend rho ( __x = x, __y = y) = ... has no principal type (a familiar difficulty). The programmer doesn't care which way round labels `__x, __y` bind, providing they're distinct, but the typing does care. Some sort of generic record extend/concatenate would be great. You might, looking at Trex syntax, think that `|` is it. This is valid: ( x = "x" | (y = 'y' | (z = 3.14))) ( x = "x", y = 'y', z = 3.14 ) -- equivalent I could put any record value in place of `(z = 3.14)`. But not in place of the `x = ..` or `y = ...`: `|` is not an operator, not commutative, not associative. Furthermore this is a place where parentheses make a difference, unlike usually in Haskell. So the following are not equivalent, indeed they're all invalid syntax ( (x = "x") | (y = 'y' | (z = 3.14))) -- x = ... has parens ( x = "x" | (y = 'y' | z = 3.14 )) -- z = ... doesn't have parens ( x = "x" | y = 'y' | (z = 3.14) ) -- z = ... in parens OK, but y = ... is not ( (x = "x") | (y = 'y') | (z = 3.14) ) -- no chance I'd like to write term `( rho1 | rho2 )` to concatenate two records. That's currently unrecognised syntax, so I think could be added as such. What would be its type/is it principal? ( rho1 | rho2 ) :: (rho1' \\ rho2') => Rec( rho1' | rho2' ) -- inventing more syntax in which constraint `(rho1' \\ rho2')` requires the two rows' labels be mutually disjoint -- read "lacks all". Ur/web has something like this. Note I'm not envisaging `|` as a genuine operator: this is still hard-wired syntax; pipe is a reserved symbol in H98 anyway. AntC

1 1