Saxon

SourceForge user: mhkay

Logged In: YES

user_id=251681

Thanks for reporting this (although you somehow failed to

notice the bright yellow box that says this isn't the right

way to report suspected bugs: I prefer to use this section

of the site to explain confirmed bugs. Even where it's a

real bug, as here, it's easier for users generally if the

entry starts with the description of the bug after analysis).

What's happening is that apply-templates makes a decision

whether to create a new stack frame for evaluating patterns

in template rules, and it's getting this decision wrong in

this case, so the variables in the pattern are being written

using slots in the same stack-frame as the template

containing the apply-templates instruction. As a

circumvention, do the call indirectly via another template

(which should define a dummy variable).

As it happens I have already redesigned this area for 8.1

and the bug went away in the process.

Test case added: match54

SourceForge user: mhkay

Logged In: YES

user_id=251681

Another related problem has been found (saved as test case

match55). This happens when the variables are not used

explicitly in the match pattern, but are introduced by the

optimizer. This example showed different syptoms: an

ArrayIndexOutOfBoundsException. This problem was still

present in the redesigned code, but has now been fixed.