After thinking about this some more for the last weeks
+1 with asterisk
from my side:
1) I am always for being as Java compatible as possible (though I see
that this might not be feasible in all cases in the future, due to Java
changing at a much faster pace and with more syntax changes now than
before; example: Java considered naming the new "var" keword "def",
which is similar to but not the same as Java-var in Groovy...)
2) I feel { { } } being interpreted as an array containing an empty
closure is confusing, i.e. not least surprise. I would rather not see it
cut it so close with regards to what the Parrot parser can handle
syntax-wise. What do others think ?
3) After introducing this syntax extension, what will be considered the
"Groovy way" of initializing an array in the future ? Is it still
final int[] a = [ 1, 1, 2, 3, 5, 8 ] as int[]
or
final int[] a = { 1, 1, 2, 3, 5, 8 }
?
In the 2nd case I would be worried that the core Groovy syntax becomes
all over the place over time, same as with the new Java lambda syntax
(though less pronounced, since using/initializing arrays is typically rare).
4) I am not too worried about the breaking edge cases, because I feel
they are quite rare in practice, the compiler catches them, and they are
easy to fix.
Cheers,
mg
On 29.04.2018 15:29, Paul King wrote:
+1
For completeness, I added some more details about the breaking changes
and workarounds into the issue - included below for easy reading.
Cheers, Paul.
=================
Groovy currently "promotes" a singleton instance of an object into an
array for assignments, e.g.:
Integer[] nums = 42
assert nums instanceof Integer[]
assert nums.size() == 1
assert nums[0] instanceof Integer
This aligns with how Groovy behaves if you try to call `.each{}` on a
non-aggregate. It treats it like a singleton collection and "iterates"
over the one item.
The existing behavior also currently works for singleton Closures:
Closure[] fns0 = { }
assert fns0 instanceof Closure[]
assert fns0.size() == 1
assert fns0[0] instanceof Closure
To add support for Java array notation, we will need to partially
disable this behavior. The proposed change involves smart parsing,
e.g. it will distinguish cases which must be an array and cases which
must be a closure but there are some degenerate edge cases which will
become breaking changes.
The case with the empty closure above will no longer work, instead you
will get this behavior, i.e. an empty array is given precedence over
an empty closure:
Closure[] fns1 = { }
assert fns1 instanceof Closure[]
assert fns1.size() == 0
To get the old behavior back you have a couple of options. Firstly,
you can provide the explicit closure argument delimiter:
Closure[] fns2 = { -> } // can't be an array
assert fns2 instanceof Closure[]
assert fns2.size() == 1
assert fns2[0] instanceof Closure
Or don't rely on singleton promotion and explicitly provide also the
array curly braces:
Closure[] fns3 = { { } }
assert fns3 instanceof Closure[]
assert fns3.size() == 1
assert fns3[0] instanceof Closure
Similarly, for the case of the identity closure:
Closure[] fns4 = { it }
Previously this worked but under this proposal will give:
groovy.lang.MissingPropertyException: No such property: it ...
Your options are to add the extra array braces as per above, or use
explicit params, e.g.:
Closure[] fns5 = { it -> it }
assert fns5 instanceof Closure[]
assert fns5.size() == 1
assert fns5[0] instanceof Closure
Alternatively, for this special case you have the following additional
option:
Closure[] fns6 = Closure.IDENTITY
assert fns6 instanceof Closure[]
assert fns6.size() == 1
assert fns6[0] instanceof Closure
There are other cases as well, e.g. this code which currently creates
a closure array containing a closure returning the integer 0:
Closure[] fns7 = { 0 }
will no longer be supported and will fail with:
org.codehaus.groovy.runtime.typehandling.GroovyCastException: Cannot
cast object '0' with class 'java.lang.Integer' to class
'groovy.lang.Closure'
The solutions are similar to previously (explicit delimiter):
Closure[] fns8 = { -> 0 }
or (explicit outer array braces):
Closure[] fns9 = { { 0 } }
On Sun, Apr 29, 2018 at 8:37 PM, Daniel.Sun <[email protected]
<mailto:[email protected]>> wrote:
Hi all,
As we all know, Java array is one of features widely applied
in Java
projects. In order to improve the compatibility with Java(Copy &
Paste). The
PR[1] will make Groovy support java-like array and make the
differences[2]
with Java less and less, e.g.
*One-Dimensional array*
```
String[] names = {'Jochen', 'Paul', 'Daniel'}
```
*Two-Dimensional array*
```
int[][] data = {
{1, 2, 3},
{4, 5, 6},
{7, 8, 9},
new int[] { 10, 11, 12 },
{13, 14, 15}
}
```
*Annotation array*
```
@PropertySources({
@PropertySource("classpath:1.properties"),
@PropertySource("file:2.properties")
})
public class Controller {}
```
*More examples*
Please see the examples on the PR page[1]
*Known breaking changes*
1. Closure array in the dynamic mode
Before
```
Closure[] y = { {-> 1 + 1 } }
assert y[0].call().call() == 2
```
After
```
Closure[] y = { {-> 1 + 1 } }
assert y[0].call() == 2
```
2. String array in the dynamic mode
Before
```
String[] a = {}
assert 1 == a.length
assert a[0].contains('closure')
```
After
```
String[] a = {}
assert 0 == a.length
```
If Groovy 3 supports Java-like array, what do you think
about the new
feature? Do you like it? We need your feedback. Thanks in advance!
[+1] I like it
[ 0] Not bad
[-1] I don't like it, because...
Cheers,
Daniel.Sun
[1] https://github.com/apache/groovy/pull/691
<https://github.com/apache/groovy/pull/691>
[2] http://groovy-lang.org/differences.html
<http://groovy-lang.org/differences.html>
--
Sent from:
http://groovy.329449.n5.nabble.com/Groovy-Users-f329450.html
<http://groovy.329449.n5.nabble.com/Groovy-Users-f329450.html>
