Difference between revisions of "If-then-else"

From HaskellWiki
Jump to navigation Jump to search
(summarize discussion at HaskellPrime mailing list from an advocators view)
 
(fix link to haskell prime wiki)
 
(8 intermediate revisions by 5 users not shown)
Line 1: Line 1:
 
== Replace syntactic sugar by a function ==
 
== Replace syntactic sugar by a function ==
   
 
For processing conditions, the <hask>if-then-else</hask> syntax was defined in Haskell98. However it could be simply replaced by the function <hask>if'</hask> with
For processing conditions,
 
the <hask>if-then-else</hask> syntax was defined in Haskell98.
 
However it could be simply replaced by the function <hask>if'</hask>
 
with
 
 
<haskell>
 
<haskell>
 
if' :: Bool -> a -> a -> a
 
if' :: Bool -> a -> a -> a
Line 16: Line 13:
 
=== Advantages ===
 
=== Advantages ===
   
The advantages of the function <hask>if'</hask>
+
The advantages of the function <hask>if'</hask> over the syntax <hask>if-then-else</hask> are the same like for all such alternatives. So let me repeat two important non-syntactic strengths of Haskell:
over the syntax <hask>if-then-else</hask>
 
are the same like for all such alternatives.
 
So let me repeat two important non-syntactic strengths of Haskell:
 
   
 
* types: classification, documentation
 
* types: classification, documentation
 
* higher order functions: combinators
 
* higher order functions: combinators
   
If <hask>if'</hask> would be a regular function,
+
If <hask>if'</hask> would be a regular function, each language tool can process it without hassle. Haddock can generate documentation for it, a text editor can make suggestions for values to insert, Hoogle can retrieve that function.
each language tool can process it without hassle.
 
Haddock can generate documentation for it,
 
a text editor can make suggestions for values to insert,
 
Hoogle can retrieve that function.
 
   
 
For example, the Hoogle query
 
For example, the Hoogle query
Line 42: Line 32:
 
=== Use cases ===
 
=== Use cases ===
   
Each of the following functions could be defined in terms of <hask>if'</hask>.
+
Each of the following functions could be defined in terms of <hask>if'</hask>. Actually, they do not even need to be in Prelude because they can be constructed so easily.
Actually, they do not even need to be in Prelude
 
because they can be constructed so easily.
 
   
That function is harder to explain in English,
+
That function is harder to explain in English, than by its implementation. :-)
than by its implementation. :-)
 
 
<haskell>
 
<haskell>
 
zipIf :: [Bool] -> [a] -> [a] -> [a]
 
zipIf :: [Bool] -> [a] -> [a] -> [a]
Line 53: Line 40:
 
</haskell>
 
</haskell>
   
  +
Infix version. This duplicates the ternary operator of C-like languages, and can be used: <hask>(cond ? exp1 $ exp2)</hask>. Additionally, Church booleans can be represented compactly by [[Section of an infix operator|sectioning]] on (?), i.e. <hask>(True?) = const; (False?) = flip const</hask>
Select a member of a pair.
 
This resembles the <code>cond?x:y</code> operation of the C language.
 
 
<haskell>
 
<haskell>
infixr 1 ?:
+
infixr 1 ?
(?:) :: Bool -> (a,a) -> a
+
(?) :: Bool -> a -> a -> a
(?:) = uncurry . if'
+
(?) = if'
 
</haskell>
 
</haskell>
   
From a list of expressions choose the one,
+
From a list of expressions choose the one, whose condition is true. The first parameter is the default value. It is returned if no condition applies.
whose condition is true.
 
The first parameter is the default value.
 
It is returned if no condition applies.
 
 
<haskell>
 
<haskell>
 
select :: a -> [(Bool, a)] -> a
 
select :: a -> [(Bool, a)] -> a
Line 70: Line 53:
 
</haskell>
 
</haskell>
 
See [[Case]].
 
See [[Case]].
  +
  +
Easy lifting into monads (MonadReader in this case)
  +
<haskell>
  +
ifF :: (a -> Bool) -> (a -> b) -> (a -> b) -> (a -> b)
  +
ifF = liftM3 if'
  +
</haskell>
   
   
 
=== Why add this function to Prelude? ===
 
=== Why add this function to Prelude? ===
   
  +
Actually people could define <hask>if'</hask> in each module, where they need it, or import it from a <hask>Utility</hask> module, that must be provided in each project. Both solutions are tedious and contradict to modularization and software re-usage. The central question is, whether <hask>if'</hask> is an idiom, that is so general that it should be in the Prelude, or not. I think it is, otherwise it wouldn't have get a special syntax.
Actually people could define <hask>if'</hask> in each module,
 
where they need it,
 
or import it from a <hask>Utility</hask> module,
 
that must be provided in each project.
 
Both solutions are tedious and
 
contradict to modularization and software re-usage.
 
The central question is, whether <hask>if'</hask> is an idiom,
 
that is so general that it should be in the Prelude, or not.
 
I think it is, otherwise it wouldn't have get a special syntax.
 
   
 
=== If-Then-Else vs. guards ===
 
=== If-Then-Else vs. guards ===
   
Actually <hask>if-then-else</hask> isn't used that often today.
+
Some people (any exact statistics?) prefer guards to <hask>if-then-else</hask>.
 
This practice has its own drawbacks, see [[Syntactic sugar/Cons]] and [[Things to avoid]].
Most programmers gave it up in favor of guards.
 
This practice has its own drawbacks,
 
see [[Syntactic sugar/Cons]] and [[Things to avoid]].
 
   
 
<!--
 
<!--
The default case is always built-in
+
The default case is always built-in and needs no extra compiler check. Compare
and needs no extra compiler check.
 
Compare
 
 
<haskell>
 
<haskell>
 
flipCond cond (x,y) | cond = (x,y)
 
flipCond cond (x,y) | cond = (x,y)
Line 106: Line 83:
 
=== Is If-Then-Else so important? ===
 
=== Is If-Then-Else so important? ===
   
Counting <hask>if-then-else</hask> or <hask>if'</hask>
+
Counting <hask>if-then-else</hask> or <hask>if'</hask> in today's Haskell programs isn't a good measure for the importance a <hask>if'</hask> function, because
in today's Haskell programs
 
isn't a good measure for the importance a <hask>if'</hask> function,
 
because
 
 
* frequently guards are used instead of <hask>if-then-else</hask>
 
* frequently guards are used instead of <hask>if-then-else</hask>
 
* there is no standard function, and this let people stick to work-arounds.
 
* there is no standard function, and this let people stick to work-arounds.
Line 115: Line 89:
 
=== What is so bad about the if-then-else sugar? ===
 
=== What is so bad about the if-then-else sugar? ===
   
  +
Since syntactic sugar introduces its own syntactic rules, it is hard to predict how it interferes with other syntactic constructs. This special syntax for instance led to conflicts with do notation. A [https://web.archive.org/web/20190612222820/https://prime.haskell.org/wiki/DoAndIfThenElse syntactic extension] to solve this problem is proposed for Haskell'. It is not known what conflicts this extension might cause in future.
Since syntactic sugar introduces its own syntactic rules,
 
it is hard to predict how it interferes with other syntactic constructs.
 
This special syntax for instance led to conflicts with do notation.
 
A syntactic extension to solve this problem
 
is [http://hackage.haskell.org/trac/haskell-prime/wiki/DoAndIfThenElse proposed] for Haskell'.
 
It is not known what conflicts this extension might cause in future.
 
   
   
 
=== Why breaking lots of old and unmaintained code? ===
 
=== Why breaking lots of old and unmaintained code? ===
   
  +
Haskell without <hask>if-then-else</hask> syntax makes Haskell more logical and consistent. There is no longer confusion to beginners like: "What is so special about if-then-else, that it needs a separate syntax? I though it could be simply replaced by a function. Maybe there is some subtlety that I'm not able to see right now." There is no longer confusion with the interference of
Haskell without <hask>if-then-else</hask> syntax
 
  +
<hask>if-then-else</hask> syntax with <hask>do</hask> notation. Removing <hask>if-then-else</hask> simplifies every language tool, say compiler, text editor, analyzer and so on.
makes Haskell more logical and consistent.
 
There is no longer confusion to beginners like:
 
"What is so special about if-then-else, that it needs a separate syntax?
 
I though it could be simply replaced by a function.
 
Maybe there is some subtlety that I'm not able to see right now."
 
There is no longer confusion with the interference of
 
<hask>if-then-else</hask> syntax with <hask>do</hask> notation.
 
Removing <hask>if-then-else</hask> simplifies every language tool,
 
say compiler, text editor, analyzer and so on.
 
 
If we arrive at [[Haskell two]] some day, (http://haskell.org/hawiki/HaskellTwo)
 
it will certainly be incompatible to former Haskell versions.
 
This does not mean, that old code must be thrown away.
 
There should be one tool,
 
that converts Haskell 98 and Haskell' to Haskell-2.
 
Having '''one''' tool for this purpose
 
is better than blowing '''all''' language tools with legacy code.
 
Syntactic replacements like <hask>if-then-else</hask> syntax to
 
<hask>if'</hask> function should be especially simple.
 
   
  +
If we arrive at [[Haskell two]] some day, ([https://web.archive.org/web/20061011095312/http://haskell.org/hawiki/HaskellTwo http://haskell.org/hawiki/HaskellTwo] (Web Archive)) it will certainly be incompatible to former Haskell versions. This does not mean, that old code must be thrown away.
 
There should be one tool, that converts Haskell 98 and Haskell' to Haskell-2.
  +
Having '''one''' tool for this purpose is better than blowing '''all''' language tools with legacy code. Syntactic replacements like <hask>if-then-else</hask> syntax to <hask>if'</hask> function should be especially simple.
   
 
=== Summary ===
 
=== Summary ===
Line 161: Line 115:
 
== Objections ==
 
== Objections ==
   
  +
Haskell is not intended to be a minimalistic language, but to be one that is easy to read. <hask>if-then-else</hask> resembles a phrase from English language. It shows clearly which expression is returned on a fulfilled condition, and which one is returned for an unsatisfied condition. It is thus easier to read. The special syntax saves parentheses around its arguments.
Haskell is not intended to be a minimalistic language,
 
but to be one, that is easy to read.
 
<hask>if-then-else</hask> resembles a phrase from English language.
 
It shows clearly which expression is returned on a fulfilled condition,
 
and which one is returned for an unsatisfied condition.
 
It is thus easier to read.
 
The special syntax saves parentheses around its arguments.
 
 
If properly indented, like
 
If properly indented, like
 
<haskell>
 
<haskell>
Line 182: Line 130:
   
 
[[Category:Proposals]]
 
[[Category:Proposals]]
  +
[[Category:Syntax]]

Latest revision as of 05:34, 10 December 2020

Replace syntactic sugar by a function

For processing conditions, the if-then-else syntax was defined in Haskell98. However it could be simply replaced by the function if' with

if' :: Bool -> a -> a -> a
if' True  x _ = x
if' False _ y = y

Unfortunately there is no such function in the Prelude.

Advocacy

Advantages

The advantages of the function if' over the syntax if-then-else are the same like for all such alternatives. So let me repeat two important non-syntactic strengths of Haskell:

  • types: classification, documentation
  • higher order functions: combinators

If if' would be a regular function, each language tool can process it without hassle. Haddock can generate documentation for it, a text editor can make suggestions for values to insert, Hoogle can retrieve that function.

For example, the Hoogle query

[Bool] -> [a] -> [a] -> [a]

may return

zipWith3 if'


Use cases

Each of the following functions could be defined in terms of if'. Actually, they do not even need to be in Prelude because they can be constructed so easily.

That function is harder to explain in English, than by its implementation. :-)

zipIf :: [Bool] -> [a] -> [a] -> [a]
zipIf = zipWith3 if'

Infix version. This duplicates the ternary operator of C-like languages, and can be used: (cond ? exp1 $ exp2). Additionally, Church booleans can be represented compactly by sectioning on (?), i.e. (True?) = const; (False?) = flip const

infixr 1 ?
(?) :: Bool -> a -> a -> a
(?) = if'

From a list of expressions choose the one, whose condition is true. The first parameter is the default value. It is returned if no condition applies.

select :: a -> [(Bool, a)] -> a
select = foldr (uncurry if')

See Case.

Easy lifting into monads (MonadReader in this case)

ifF :: (a -> Bool) -> (a -> b) -> (a -> b) -> (a -> b)
ifF = liftM3 if'


Why add this function to Prelude?

Actually people could define if' in each module, where they need it, or import it from a Utility module, that must be provided in each project. Both solutions are tedious and contradict to modularization and software re-usage. The central question is, whether if' is an idiom, that is so general that it should be in the Prelude, or not. I think it is, otherwise it wouldn't have get a special syntax.

If-Then-Else vs. guards

Some people (any exact statistics?) prefer guards to if-then-else. This practice has its own drawbacks, see Syntactic sugar/Cons and Things to avoid.


Is If-Then-Else so important?

Counting if-then-else or if' in today's Haskell programs isn't a good measure for the importance a if' function, because

  • frequently guards are used instead of if-then-else
  • there is no standard function, and this let people stick to work-arounds.

What is so bad about the if-then-else sugar?

Since syntactic sugar introduces its own syntactic rules, it is hard to predict how it interferes with other syntactic constructs. This special syntax for instance led to conflicts with do notation. A syntactic extension to solve this problem is proposed for Haskell'. It is not known what conflicts this extension might cause in future.


Why breaking lots of old and unmaintained code?

Haskell without if-then-else syntax makes Haskell more logical and consistent. There is no longer confusion to beginners like: "What is so special about if-then-else, that it needs a separate syntax? I though it could be simply replaced by a function. Maybe there is some subtlety that I'm not able to see right now." There is no longer confusion with the interference of if-then-else syntax with do notation. Removing if-then-else simplifies every language tool, say compiler, text editor, analyzer and so on.

If we arrive at Haskell two some day, (http://haskell.org/hawiki/HaskellTwo (Web Archive)) it will certainly be incompatible to former Haskell versions. This does not mean, that old code must be thrown away. There should be one tool, that converts Haskell 98 and Haskell' to Haskell-2. Having one tool for this purpose is better than blowing all language tools with legacy code. Syntactic replacements like if-then-else syntax to if' function should be especially simple.

Summary

  • Light proposal, compatible with Haskell 98: Add if' to the Prelude, maybe with a different name.
  • Full proposal, incompatible with Haskell 98 and Haskell': Additionally remove if-then-else syntax


See also


Objections

Haskell is not intended to be a minimalistic language, but to be one that is easy to read. if-then-else resembles a phrase from English language. It shows clearly which expression is returned on a fulfilled condition, and which one is returned for an unsatisfied condition. It is thus easier to read. The special syntax saves parentheses around its arguments. If properly indented, like

if a
  then b
  else c

or

if a
  then b else c

then there is no conflict with the do-notation.