[BUG] replace, tokenize and analyze-string must throw when pattern matches an empty string

[PieterLamers]

Describe the bug

fn:replace, fn:tokenize and fn:analyze-string allow to use a pattern that matches an empty string. That results in odd behaviour as in this example:

replace( '12.34' , '^\D*', '')

The provided pattern does always match because it matches the empty string.

Expected behavior

Error FORX0003 is thrown with location information

Actual

The first character is swallowed: 2.34

To Reproduce

replace( '12.34' , '^\D*','')  

or

replace( '12.34' , '^[^0-9]*','')

will return 2.34 instead of the desired 12.34

Reference

This used to have a different behaviour in earlier versions of existdb where a pattern that matched an empty string would just return the input unchanged (likely related to #3530).

fn:replace specification

XQTS 31 tests

Context (please always complete the following information):

• OS: Windows10
• eXist-db version: 5.3.0-snapshot 55e77cc
• Java Version 1.8.0.281

[line-o]

This might be a change in behaviour again.

replace("$11.23", "^[^0-9]*(.*)$", "$1")

is the correct form to write the replacement.

[line-o]

Looking at it again, I am unsure. This looks like a bug. But

• replace('12.34' , '^\D','')
• replace('12.34' , '^[^0-9]','')
  both do not match and therefore do not replace a single character at the beginning.

[line-o]

So replace($may-start-with-currency-symbol, "^\D+", "") might be the easiest solution.

[line-o]

related #3530

[line-o]

BaseX 9.5 will raise error [FORX0003] Pattern matches empty string. for the original pattern "^[^0-9]*".

[line-o]

Saxon 10.0 (HE) also throws the same error.
So all in all this is definitely a bug, because no error is thrown!

[line-o]

@PieterLamers would you or may I edit the bug description to reflect the new findings? Or should I open a separate one?

[PieterLamers]

Hi @line-o , thanks for the explanations! You are welcome to edit the ticket. I think I will simply replace the * by a + to avoid the error.