Surprise list

From Erights

(Difference between revisions)
Jump to: navigation, search
(accumulator precedence)
(Single-letter uriGetters are a special case)
 
(One intermediate revision not shown)
Line 38: Line 38:
    
    
   # problem: <ClassCastException: String doesn't coerce to an int>
   # problem: <ClassCastException: String doesn't coerce to an int>
 +
 +
The current developer version also allows strict checking to be turned on with a pragma (experimental):
 +
 +
# E sample
 +
? pragma.enable("for-must-match")
 +
 +
? for x :int in things {
 +
>    println(x)
 +
> }
 +
# stdout: 1
 +
#
 +
 +
# problem: <ClassCastException: String doesn't coerce to an int>
==<code>def x</code> doesn't return <var>x</var>==
==<code>def x</code> doesn't return <var>x</var>==
Line 53: Line 66:
   # value: &lt;Promise&gt;
   # value: &lt;Promise&gt;
-
===Reason and Uses===
+
===Rationale===
-
If you want to pass a resolver as an argument, this syntax is helpful:
+
If you want to pass a resolver as an argument, using it as an “out parameter”, this syntax is helpful:
   x.hereIsAResolver(def y)
   x.hereIsAResolver(def y)
Line 70: Line 83:
This has been discussed in [http://www.eros-os.org/pipermail/e-lang/2005-July/010821.html an e-lang thread].
This has been discussed in [http://www.eros-os.org/pipermail/e-lang/2005-July/010821.html an e-lang thread].
-
===Reason and Uses===
+
===Rationale===
This cannot be fixed without removing [http://www.erights.org/elib/distrib/pipeline.html pipelining], eliminating one of the major benefits of the E reference model.
This cannot be fixed without removing [http://www.erights.org/elib/distrib/pipeline.html pipelining], eliminating one of the major benefits of the E reference model.
Line 92: Line 105:
     ...
     ...
-
===Reason and Uses===
+
===Rationale===
None known; this is probably an accident of the definition of the expansion of accumulator syntax.
None known; this is probably an accident of the definition of the expansion of accumulator syntax.
Line 99: Line 112:
Avoid accumulator syntax when the accumulation cannot be expressed as a single call.
Avoid accumulator syntax when the accumulation cannot be expressed as a single call.
 +
 +
 +
 +
== Shadowing function arguments ==
 +
 +
Rebinding a name within a block is an error, e.g.
 +
 +
  def foo() {
 +
  def x := 2
 +
  def x := 3
 +
  }
 +
 +
("Failed: x already in scope")
 +
 +
However, rebinding an argument does not issue any warning:
 +
 +
  def foo(x) {
 +
  ...
 +
  def x := 2
 +
  ...
 +
  println(x)
 +
  }
 +
 +
=== Rationale ===
 +
 +
The already-in-scope error is intended to catch accidentally using the same name twice, not prohibit rebinding. Generally, you can expect it to be supressed anywhere there is visible <code>{...}</code> syntax.
 +
 +
== Single-letter uriGetters are a special case ==
 +
 +
It is not possible to refer to a single-letter uriGetter in a URI literal.
 +
 +
  ? def <t> := ["foo" => "bar"]
 +
  # value: ["foo" => "bar"]
 +
 
 +
  ? interp.setExpand(true)
 +
  ? <t:foo>
 +
  # expansion: file__uriGetter.get("t:foo")
 +
 
 +
  # value: <file:/Users/kpreid/t:foo>
 +
 +
===Rationale===
 +
 +
This is a feature intended for convenient support for Windows drive letter filenames. {{XXX|Look at whether MarkM agreed to remove it.}}
 +
 +
===Alternative===
 +
 +
Use names longer than one letter.
[[Category:E language]]
[[Category:E language]]

Latest revision as of 11:28, 27 October 2011

Potentially surprising parts of the E language.

Current as of E 0.8.33o.

Contents

For-loop pattern failure is not an error

If the key or value pattern of a for loop does not match, then the loop simply skips that element of the collection. This can be useful, but is unlike all other non-explicit pattern match operations (def and parameter lists).

Examples

 # E sample
 
 ? def things := [1, "two", 3]
 # value: [1, "two", 3]
 
 ? for x :int in things {           
 >     println(x)
 > }
 # stdout: 1
 #         3
 #         
 

Alternative

Move the pattern into a def:

 # E sample
 
 ? for thing in things {
 >     def x :int := thing
 >     println(x)
 > }
 # stdout: 1
 #         
 
 # problem: <ClassCastException: String doesn't coerce to an int>

The current developer version also allows strict checking to be turned on with a pragma (experimental):

# E sample
? pragma.enable("for-must-match")

? for x :int in things {
>     println(x)
> }
# stdout: 1
#

# problem: <ClassCastException: String doesn't coerce to an int>

def x doesn't return x

The forward declaration expression, def var, does not return var but rather the Resolver for it.

Examples

 # E sample
 
 ? def x
 # value: <Resolver>
 
 ? x
 # value: <Promise>

Rationale

If you want to pass a resolver as an argument, using it as an “out parameter”, this syntax is helpful:

 x.hereIsAResolver(def y)
 ... use y

Alternative

If you want the actual promise, simply write (def x; x).

Unresolved references do not necessarily behave as their future resolved identity

— messages sent before the reference is resolved may be reacted to however the current holder of the reference “arrow-head” chooses, which does not necessarily correspond to the reference to which the unresolved reference resolves.

This has been discussed in an e-lang thread.

Rationale

This cannot be fixed without removing pipelining, eliminating one of the major benefits of the E reference model.

Examples

For now, see this lengthy example by MarkM.

Alternative

To avoid being vulnerable to this type of misbehavior, do not use a sameness test (==) or Map key lookup in order to decide on the reliability of the response to a previously sent message. This might involve using a when-catch/whenResolved construct to wait until the reference is resolved.

Accumulator operator is lowest-precedence

As you can see in the following expansion, the operator following the “_” is always lowest-precedence (as it is the one effectively rewritten into an assignment form):

   ? e`pragma.enable("accumulator"); accum 0 while (a) { _ * b + c }`
   # value: e`null
   ...
   #                            accum__1 := accum__1.multiply(b.add(c))
   ...

Rationale

None known; this is probably an accident of the definition of the expansion of accumulator syntax.

Alternative

Avoid accumulator syntax when the accumulation cannot be expressed as a single call.


Shadowing function arguments

Rebinding a name within a block is an error, e.g.

 def foo() {
 	def x := 2
 	def x := 3
 }

("Failed: x already in scope")

However, rebinding an argument does not issue any warning:

 def foo(x) {
 	...
 	def x := 2
 	...
 	println(x)
 }

Rationale

The already-in-scope error is intended to catch accidentally using the same name twice, not prohibit rebinding. Generally, you can expect it to be supressed anywhere there is visible {...} syntax.

Single-letter uriGetters are a special case

It is not possible to refer to a single-letter uriGetter in a URI literal.

 ? def <t> := ["foo" => "bar"]
 # value: ["foo" => "bar"]
 
 ? interp.setExpand(true)
 ? <t:foo>
 # expansion: file__uriGetter.get("t:foo")
 
 # value: <file:/Users/kpreid/t:foo>

Rationale

This is a feature intended for convenient support for Windows drive letter filenames. XXX Look at whether MarkM agreed to remove it.

Alternative

Use names longer than one letter.

Personal tools
more tools