Created attachment 187091 [details]
Schema document with empty branches in the pattern facets of several datatypes

When a choice in a regular expression ends with an empty branch, the empty branch is not handled correctly.  Leading empty branches appear also to have issues, but different ones.

Example:  "(a|)"

This is a choice between the expression on the left ("a") and the one on the right (""), and should match any string that matches either of them.  But when I validate strings against a type defined with this regex, a single "a" is accepted but the empty string "" is not.  Strings containing characters other than "a" and strings with length greater than 1 are correctly rejected.

Example:  "(|a)"

This is the same choice as the previous example and defines the same language.  Validating against a type with this pattern, however, libxml 20703 correctly accepts "" and "a", and correctly rejects "b" and "c", but also accepts "aa".  

The regular expressions "", "|", and "(|)" correctly accept the empty string as input and correctly reject "a" in the input.

Rewriting "(a|)" to "(a)?" is of course possible, but "(a|)" is in fact a legal regex, libxml2 correctly accepts it, and you probably want to interpret it correctly.  (Perhaps few people would write (a|) by hand, but a simple translation of the ABNF for URIs and IRIs in RFC 3986 and RFC 3987 does produce an expression of this form for the non-terminal 'relative-part'.)

I attach a schema document and an XML document with sample test strings.