You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@xalan.apache.org by Sc...@lotus.com on 2000/11/08 20:36:29 UTC
Re: [Fwd: Re: Problem with new variable evaluation?]
> Do you anticipate a lot more
> changes the rest of today?
I'm doing a fairly major change to composition of global variables and
parameters... the need to work on the same list in order for importance to
work correctly. May go forward with lazy evaluation of variables, not sure
yet...
> I'd come across the LocPathIterator bug
Sorry, which one?
> I'm currently dealing
> with a problem in TemplateList.compose(). If a wild card template is
> inserted into a chain via insertAssociationIntoList and it ends up being
> at the head of the chain, compose() doesn't update the head so that
> subsequent inserts of additional wild cards go in the wrong place. I'm
> fixing it.
Be very careful. Those lists are tricky, really tricky. I think there's a
diagnostics method you can call to dump the lists, if you haven't allready
found it.
> I'm changing Compiler, WalkerFactory, and StepPattern so
> we're colliding a bit.
What's happening with these?
-scott
Gary L Peskin
<garyp@firste To: Scott Boag <Sc...@lotus.com>
ch.com> cc:
Subject: [Fwd: Re: Problem with new variable evaluation?]
11/08/00
02:42 PM
Scott -- I'm forwarding this to you because the list seems a little slow
today.
Gary
-------- Original Message --------
Subject: Re: Problem with new variable evaluation?
Date: Wed, 08 Nov 2000 11:31:36 -0800
From: Gary L Peskin <ga...@firstech.com>
Organization: The Firstech Corporation
To: xalan-dev@xml.apache.org
References: <OF...@lotus.com>
Scott_Boag@lotus.com wrote:
>
> Hi Gary. This is fixed in the check-in I just did.
>
> I have the conformance tests up and running again, so I'll be stepping
> through those today to see if there are any more regressions.
Hi, Scott. This turns out, I think to be due to some problems caused by
my new code. I've been working pretty hard on this and keep finding
little things. I'm changing Compiler, WalkerFactory, and StepPattern so
we're colliding a bit.
I've updated from your latest commit. Do you anticipate a lot more
changes the rest of today?
I'd come across the LocPathIterator bug and it actually worked well
because I was able to tell when I wasn't using a ChildIterator but now
that you've fixed LocPathIterator, I'll have to check carefully. :)
I also have the conformance tests going, with help from Shane, and
that's how I'm finding these things. It's very helpful. Cleaning up
the compiler has caused some new bugs to surface. I'm currently dealing
with a problem in TemplateList.compose(). If a wild card template is
inserted into a chain via insertAssociationIntoList and it ends up being
at the head of the chain, compose() doesn't update the head so that
subsequent inserts of additional wild cards go in the wrong place. I'm
fixing it.
Since I haven't journeyed into these corners of Xalan before, it's been
slow going but I think the ramp-up time will be worth it. Very
interesting stuff. I have some ideas for speeding things up but first
let me get the thing working.
Gary
Re: [Fwd: Re: Problem with new variable evaluation?]
Posted by Gary L Peskin <ga...@firstech.com>.
Scott_Boag@lotus.com wrote:
> > I'd come across the LocPathIterator bug
>
> Sorry, which one?
The one you fixed (in nextNode(), I think) where the savedStart wasn't
really the savedStart if it was -1 but was getting reset to 0.
> > I'm currently dealing
> > with a problem in TemplateList.compose(). If a wild card template is
> > inserted into a chain via insertAssociationIntoList and it ends up being
> > at the head of the chain, compose() doesn't update the head so that
> > subsequent inserts of additional wild cards go in the wrong place. I'm
> > fixing it.
>
> Be very careful. Those lists are tricky, really tricky. I think there's a
> diagnostics method you can call to dump the lists, if you haven't allready
> found it.
I'm following step-by-step in the debugger and being very careful.
Also, I'll run full regression tests and get 1:1 match (or better)
before checking anything in.
>
> > I'm changing Compiler, WalkerFactory, and StepPattern so
> > we're colliding a bit.
>
> What's happening with these?
I'm improving the Compiler.getWhatToMatch() so that it really only shows
what to match. This lets me take the special casing at executeStep for
attribute/document nodes out of StepPattern and makes things cleaner and
hopefully faster. However, this is surfacing several problems. For
example, patterns like text() were going onto the wild card list and the
text list was never populated. My changes are also improving a few of
the conformance tests.
Gary