PhoneRecognizer: analysis_explanation names the previous match's region when an international number precedes a national one

[feiiiiii5]

What happens

PhoneRecognizer.analyze rebinds the region variable it is iterating over to the region detected for the current match, so that value leaks into every later match in the same pass. The result is an analysis_explanation that names a region which never matched the number.

This reproduces with the default recognizer — no supported_regions, no custom configuration, no registry:

from presidio_analyzer.predefined_recognizers.generic.phone_recognizer import PhoneRecognizer

text = "My international number is +44 1234 567890, and my US one is (415) 555-0132"
for res in PhoneRecognizer().analyze(text, ["PHONE_NUMBER"]):
    print(res.start, res.end, res.score, res.analysis_explanation.textual_explanation)

Measured on main @ f251c513, Python 3.11:

27 42 0.4 Recognized as GB region phone number, using PhoneRecognizer     <- +44 1234 567890, correct
61 75 0.4 Recognized as GB region phone number, using PhoneRecognizer     <- (415) 555-0132, wrong

The US number is explained as a GB number. Swap the two numbers in the text and both labels are correct:

 5 19 … Recognized as US region phone number    (415) 555-0132
25 40 … Recognized as GB region phone number    +44 1234 567890

So the bug is order-dependent, which is also why it is invisible today: every multi-number case in tests/test_phone_recognizer.py::test_when_phone_with_textual_explanation_then_succeed puts the national-format number first and the international one second.

Why

for region in self.supported_regions:                      # outer loop variable
    for match in phonenumbers.PhoneNumberMatcher(text, region, ...):
        try:
            parsed_number = phonenumbers.parse(text[match.start : match.end])
            region = phonenumbers.region_code_for_number(parsed_number)   # rebinds it
            results += [self._get_recognizer_result(match, text, region, nlp_artifacts)]
        except NumberParseException:
            results += [self._get_recognizer_result(match, text, region, nlp_artifacts)]

phonenumbers.parse() is called with no default_region, so a national-format number always raises NumberParseException and takes the except branch — which uses whatever region currently holds. After any international number in the same pass, that is the previous match's region rather than the region being iterated.

What is affected

Detection itself is not: spans, scores and entity types are identical before and after (verified on both orders above). Only analysis_explanation.textual_explanation is wrong. That still matters for anything that surfaces or keys on the explanation — the field exists precisely to say which region matched — and this class has needed an explanation fix before (#1330 fixed a wrong analysis_explanation here).

Expected

Each result reports the region it was actually matched under: the +44 number as GB, the (415) number as US, regardless of the order they appear in the text.

Suggested fix is a per-match matched_region so the loop variable is never overwritten; I have it ready with a regression case added to the existing parametrized test, Patch with regression test: #2270

[feiiiiii5]

Duplicate of #2268, which was filed a minute earlier and follows the repo's bug template. Closing this one so the report lives in a single place; the reproduction, both orderings and the measured output are all in #2270, which I have retargeted to #2268.

Both reports came from the same account in the same minute, so this is on me to clean up rather than leave for the maintainers to reconcile.