Re: xslt 2.0 regex

On Sat, Mar 17, 2012 at 2:01 PM, davep <davep@xxxxxxxxxxxxx> wrote:
> On 17/03/12 17:38, Brandon Ibach wrote:
>> Ignoring that for the moment, though, Tony pointed out one consequence
>> of this, but the bigger issue is that the "|" operator in regex is
>> fairly low precedence, so you often need some parenthesis around the
>> list of alternatives to get things right.  Your NameStartChar.re has a
>> hex-char-ref-encoded "$" at the beginning, so that regex is actually
>> "$[A-Z] | _ | [a-z] | ...", which means it will match "(a dollar sign
>> followed by an upper-case English letter) or (an underscore) or (a
>> lower-case English letter) or ...".
>
> Which is correct (less the missing :), that's what they call the startChar?

Not quite.  Notice the parens I put in the "translation".  The dollar
sign is only matched when it is followed by an upper-case English
letter.  A variable name starting with anything else won't match
correctly.  You'd need parens around the list of alternatives to make
sure the "|" only applies to the upper-case-letter pattern and not the
dollar sign.  Since you re-use NameStartChar.re in NameChar.re,
however, that won't work.  Moving the dollar sign to Name.re solves
that problem.

>> All that said, I got this to work by dropping the "$" from the start
>> of NameStartChar.re and changing Name.re to:
>>
>> concat("\$(", $NameStartChar.re, ")(", $NameChar.re,")*")
>
> presumably because $ is within the a-z grouping....
> Yes it works.
>
> It should be possible (and simpler) to use Tony's idea \i\c+  since : is a
> valid variable name. Not a syntax I've used before I think?

QName is actually defined (in more EBNC-like notation) as "(NCName
':')? NCName", which allows only one ":" under specific circumstances,
so your best bet is probably something like this, based on Tony's
later version:

([\i-[:]][\c-[:]]*:)?[\i-[:]][\c-[:]]*

-Brandon :)