XSLT 2.0 and XPath 2.0 Programmer's Reference, 4th Edition (465 page)

The actual string that is checked against the pattern facet is as follows:

• When converting from
  xs:string
  or
  xs:untypedAtomic
  , the value being converted.
  
• In other cases, a lexical representation of the result of converting the value to the primitive base type of the target type, where the primitive types are considered to include
  xs:integer
  ,
  xs:dayTimeDuration
  , and
  xs:yearMonthDuration
  . The lexical representation chosen is generally the same as the result of casting the value to
  xs:string
  using the rules in this chapter, except in cases where XML Schema defines a canonical lexical representation that is different from the result of XPath casting. This affects the following cases:
  
  • For
    xs:decimal
    , a whole number is represented as (say)
    12.0
    rather than
    12
    .
    
  • For
    xs:double
    and
    xs:float
    , the value is always in exponential notation with a single non-zero digit before the decimal point; for example,
    1.2E1
    rather than
    12
    .
    
  • For
    xs:time
    and
    xs:dateTime
    , the value (if it has a timezone) is adjusted to UTC, so
    12:00:00-05:00
    is represented as
    17:00:00Z
    .
    
  • For
    xs:date
    , the timezone is adjusted to be in the range −11:59 to +12:00, adding or subtracting a day as necessary.
    

Suppose, for example, that the schema defines a
my:utcTime
type as a restriction of
xs:time
, with

. This constrains the value to end with
Z
. The cast
my:utcTime(“12:00:00+01:00”)
will therefore fail. However,
my:utcTime(xs:time(“12:00:00+01:00”))
will succeed. The cast to
xs:time
succeeds because the value is a valid time, and the cast to
my:utcTime
succeeds because the canonical lexical representation of the value is
11:00:00Z
, which matches the pattern, despite the fact that the actual value is not in UTC. As I said before, the best advice one can give is to avoid this area.