> An XSLT 2.0 processor running in backwards compatibility mode(due to 
> xsl:version=1.0) accepts an illegal XSLT 1.0 stylesheet, as 
> long as it is 
> valid XSLT 2.0(say, an if statement was in an expression). 
> Hence, one can 
> write an "1.0" stylesheet which works just fine with an XSLT 
> 2.0 processor, 
> but is flagged as invalid at a static stage with an XSLT 1.0 
> processor.

Yes, that's correct.
> 
> What makes me wonder is because I find it hard to to motivate 
> this behavior or 
> see it in a practical perspective.

The WG didn't feel it was reasonable to expect a 2.0 processor to have
complete knowledge of which facilities are in 1.0 and which are new in 2.0 -
especially for things where the rules are relaxed at run-time such as RTF to
node-set conversion. Therefore it's not an error to say version="1.0" and
use a 2.0 feature.

Instead the version="1.0" attribute is used to switch behavior for a few
specific constructs: it simulates the 1.0 behavior for those constructs.
> 
> From one perspective, I think I find it annoying, because one 
> can't "simulate" 
> an XSLT 1.0 processor with the xsl:version attribute, and 
> hence be guaranteed 
> a stylesheet works with a "real" 1.0 processor

That's true: the WG felt it wasn't a reasonable requirement on 2.0
processors that they implement all the rules of 1.0 alongside those of 2.0.

Michael Kay