Patch #1420
closed
Add saxon:closure extension function
0%
Description
SourceForge user: gschadow
Hi Michael, it worked! I could indeed add a saxon:closure
() extension function that will force it's argument
Expression to be lazily evaluated. It simply creates a
Closure from the first argument in the current
XPathContext.
In order to make it happen I had to make numerous little
additions, most of them to get the name saxon:closure
known. I used the saxon:expression extension function
as an example and did everything the way you did it for
saxon:expression.
The only real change that you may want to review is
that I made SequenceValue implement Item (simply by
adding one trivial little missing function implementation.
Also required was a little change in Value.asValue().
Attached is a ZIP file that contains the complete diff
that should take care of every change needed. It's
against 7.6 though but the diff has not much context so
I don't think much would be rejected.
Also in the ZIP file is a little test case (to invoke as a
transform over any XML input). The test-case requires a
Java extension thingie with a little side-effect, because
that's my whole point.
The point that the test illustrates is how side-effects
can be carried out with lazy evaluation ONLY if the value
is actually referenced. The real example is a database
insert that should only be done for parent-records that
have dependent children. If you run the example without
the closure() function it's output is:
next 1 for insert a
next 2 for insert b parent 1
next 3 for insert c parent 1
next 4 for insert d
next 5 for insert e
next 6 for insert f parent 5
<child id="2" parentId="1" name="b"/>
<child id="3" parentId="1" name="c"/>
<child id="6" parentId="5" name="f"/>
which is bad because it inserted "next 4 for indsert d" for
the parent named "d" that had no children. Once you
apply the patch and use the closure() function the
output is correctly:
next 1 for insert a
next 2 for insert b parent 1
next 3 for insert c parent 1
next 4 for insert e
next 5 for insert f parent 4
<child id="2" parentId="1" name="b"/>
<child id="3" parentId="1" name="c"/>
<child id="5" parentId="4" name="f"/>
Files
Updated by Anonymous over 20 years ago
SourceForge user: gschadow
Logged In: YES
user_id=575520
Oops, the diff file was somehow screwed up, I couldn't
quite apply it properly. Here is a resubmission of just that
diff file.
Updated by Anonymous over 20 years ago
SourceForge user: gschadow
Logged In: YES
user_id=575520
hmm, I'm wondering if calling this function "closure" is actually
appropriate. It's intent is more like "delay evaluation" or "force
lazy evaluation". So, lazy-evaluate() or delay-evaluation()
might be better names. You call the shots on this.
Updated by Anonymous over 20 years ago
SourceForge user: mhkay
Logged In: YES
user_id=251681
I've thought about this on and off. I can't say I'm
comfortable with it. It's going to be very hard to explain
to people what this function is designed to achieve, and I
suspect that it's going to be difficult to keep it tested
and maintained as the internal structure of the software
evolves. I can see a stronger case for the reverse function,
one which forces eager evaluation of an expression (this can
be written as a standard extension function with no special
support).
I can already see some problems, for example lazy evaluation
of an expression isn't safe if it depends on certain aspects
of the context, such as position() and last(). These
conditions could change in the future.
I would be much happier with a design that takes a proper
look at how the language can be extended cleanly to manage
side-effects, on the lines of monads in Haskell (or
otherwise!). I think many of the use-cases could be handled
more cleanly by using higher-order functions and I want to
explore extensions in that direction - though my priority at
the moment is to implement what's in the 2.0 spec, rather
than extensions to it.
I can't see why you want SequenceValue to implement Item. An
Item is defined in the language spec as a Node or an
AtomicValue, whereas a SequenceValue can hold multiple Items.
The name saxon:closure, incidentally, was used in the past
for a function that computes a transitive closure, and since
people sometimes upgrade from very old releases, I wouldn't
want to re-use the name. saxon:lazy-evaluate would be better.
Please register to edit this issue
.