9:43 a.m.
On 18/08/2011 15:03, Edson Tirelli wrote:
Mark,
The [] syntax for the labels will clash with the sequencing syntax
we've been discussing. Possibly {} or a unique separator:
{else1} A()
else1 := A()
else1 ?= A()
Considering that Patterns can also take bindings, probably {} is
more distinct:
{else1} a : A()
My vote:
when
{else1} Person( name == "darth" ) // works on patterns
A()
{else2} B()
then
....
otherwise.else1
...
otherwise.else2
...
end
Will we support unlabeled "else" as well?
when
A() and B()
then
...
otherwise
...
end
If so, what will be the semantics of it? What happens if an A() is
inserted but not B()? vice-versa? What happens if C() is inserted?
no plans for
unlabelled "else". "otherwise" is something different.
Regarding inline "consequences", at the moment I am not really a
fan of it. I think it complicates the syntax unnecessarily at this
point but I can be convinced. The support to else by itself is a big
step forward as you know users frequently ask for that.
The labelled consequencei
is necessary for compact case statements,
which is very common for signal processing type problems, otherwise you
need to generate a large number of rules with small variations which is
harder to read and maintain. This "case" type of construct is at the
heart of erlang.
o : Object() from stream
(or A( field1 == "1" ) > {a1} from o
A( field1 == "2" ) > {a2} from o
B( field1 == "3 ) > {b1} from o
B( field2 == "5", field 3 == 6 ) > {b2} ) from o
http://www.erlang.org/doc/reference_manual/expressions.html#id75991
When can do this one step at a time, no need for a big bang. Each
proposal is an iterative improvement on the last. What i want to make
sure is we have done the design through to completion and considered all
edge cases, so as to avoid a dead end in the syntax should we add it
later. Last thing i want to do is think in one years time "oh it would
be nice to add this but we can't because of a syntax choice we made
earlier and we didn't want to think about more complex use cases as we
assumed they would never be necessary".
the only issue with {label} as opposed to [label] is what do we chose
for when we want inline code rather than label. I guess we can say
{label} will call a function literal if it exists, otherwise it attempts
to be evaluated as an expression.
Mark
My .02c.
Edson
2011/8/18 Mark Proctor <mproctor(a)codehaus.org
<mailto:mproctor@codehaus.org>>
We have been looking into designs around else, so here are our initial
brain storming ideas, which aims at doing more than just else, but
handling signal processing like situations. "else" is always
triggerd by
the failure of a left propagation. In effect an named "else" block is
just another terminal node that will result in an activation on the
agenda. It will have access to declarations prior to the failure of
propagation in the network.
// Possible syntaxes
[name] ( CE+ ) // no symbol
[name] | ( CE+ )
[name] < ( CE+ )
1)
when
[name1] < Person( name == "darth" ) // works on patterns
A()
then
....
then.name1
...
end
2)
when
$p : Person( )
[name1] < eval( $p.name <http://p.name> == "darth" ) // works
on evals
A()
then
....
then.name1
...
end
3)
when
[name1] < ( Person( name == "darth" ) and Address( city ==
"death
star" ) // works on groups
A()
then
....
then.name1
...
end
This could actuall be extended to have inline "then" too. In this case
when their is a success propagation on that node it will result in an
activation placed on the agenda that has access to all the prior bound
declarations.
1)
when
Person( name == "darth" ) > [name1] // works on patterns
A()
then
....
then.name1
...
end
2)
when
$p : Person( )
eval( $p.name <http://p.name> == "darth" ) > [name1] // works
on evals
A()
then
....
then.name1
...
end
3)
when
( Person( name == "darth" ) and Address( city == "death star"
) >
[name1] // works on groups
A()
then
....
then.name1
...
end
This can be used with 'or'
when
( A() > [a1] or
B() > [b1] or
C() > [c1] )
D()
then
...
then.a1
....
then.b1
....
then.c1
...
end
It's a little tricker but in theory we can do this before/afer the
'or' too
This can be used with 'or'
when
[x1] < ( A() > [a1] or
B() > [b1] or
C() > [c1] )
D()
then
...
then.a1
....
then.b1
....
then.c1
...
then.x1
....
end
We could allow [name] as just an inline creation to an activation that
always passes, which with 'or' could provide a "default".
when
[x1] < ( A() > [a1] or
B() > [b1] or
C() > [c1] or
[default] )
D()
then
Of course both could be supported at the same time
[afailed] < A() > [asuccess]
We could further allow just an inline code block, isntead of an inline
reference to a block {...code here...} instead of [name1].
We can also use this to do switch like operations, for erlang style
signal processing, although i'd like to see an improvemet to the
syntax
here, just not sure what it would be...
$o : Object() from stream
( A() > [a] from $o or
B() > [b] from $o or
C() > [c] from $o )
Where as 'or' currently works like java's "|" single operator,
i.e. all
logical branches are tested. We could add a short cut or operationr
'sor' that would work like "||", so once the first CE matches in
an 'or'
block the rest are igored. We could even consider an 'xor' ....
Finally there is no reason why we couldn't allow other CE's after
the <.
Which would provide for very rich signal processing. For instance. If
A() fails, it'll propagate to B, if B() fails it'll activate [a1]
[a1] < B() < A()
This can be nested and using using parenthesis to show groupings.
( [a1] < B() > [b2] ) < A()
Anyway more food for thought, enjoy :)
Mark
_______________________________________________
rules-dev mailing list
rules-dev(a)lists.jboss.org <mailto:rules-dev@lists.jboss.org>
https://lists.jboss.org/mailman/listinfo/rules-dev
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com <http://www.jboss.com>
_______________________________________________
rules-dev mailing list
rules-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev