Guard AllNumberGroupsRemainGrouped against indexOf/find(-1) for the country-code skip (Java + C++)

[twcclegg]

Issue

https://issuetracker.google.com/issues/552025944

What's wrong

In PhoneNumberMatcher.allNumberGroupsRemainGrouped (Java), when the country calling code isn't a literal substring of the candidate text, indexOf returns -1, and:

fromIndex = normalizedCandidate.indexOf(countryCode) + countryCode.length();

leaves fromIndex at a small, wrong, non-negative value instead of "not found." This desyncs the subsequent group-matching loop, which can make a grouping check that should succeed silently return false instead — i.e. a valid phone number embedded in free text can fail to match, with no exception. The very next indexOf call in the same method (one line below, inside the loop) already guards against exactly this with if (fromIndex < 0) { return false; }; this fix applies the same guard to the country-code skip.

In practice this is probably rare, since a matched candidate's text usually does contain the country code's digits somewhere. Confirmed with a constructed case (country code 44, candidate text that doesn't contain "44") added as a new unit test.

Fix

Guard the indexOf result before using it, leaving fromIndex at its default (0, i.e. "don't skip anything") when the country code isn't found as a literal substring, rather than propagating a bogus offset.

Scope across languages

Per the contributing guide's note to check whether a fix applies in C++, Java, and JS:

• Java — fixed (this is the primary fix).
• C++ (cpp/src/phonenumbers/phonenumbermatcher.cc) — has the exact same pattern with std::string::find, and it's arguably worse there: find returns npos (size_t max) on a miss, and the unguarded + country_code.size() wraps around via unsigned overflow into a small, wrong, non-negative from_index instead of crashing loudly. Fixed with the same guard.
• JS — not applicable. The JS port has no PhoneNumberMatcher (find-numbers-in-text) implementation at all; allNumberGroupsRemainGrouped/AllNumberGroupsRemainGrouped doesn't exist there. Verified by grepping javascript/i18n/phonenumbers/ for the function and for any matcher module — only asyoutypeformatter.js and phonenumberutil.js (parse/format/validate) exist, no text-scanning matcher.

Testing

No local JDK/C++ toolchain was available to me to run the test suites, so I traced both fixes by hand and I'd appreciate a maintainer/CI double-check.

• Java: testAllNumberGroupsRemainGroupedCountryCodeNotInCandidate calls the package-private allNumberGroupsRemainGrouped directly with a PhoneNumber whose countryCodeSource is FROM_NUMBER_WITH_PLUS_SIGN (country code 44) and a candidate string that never contains the literal digits "44". Before the fix this returns false (bug); after the fix it returns true.
• C++: no test added. AllNumberGroupsRemainGrouped is in an anonymous namespace in phonenumbermatcher.cc, so unlike the Java package-private method it isn't reachable from the test binary in a different translation unit for a direct unit test, and I didn't want to change its linkage just to test this. Happy to take a suggestion from a maintainer on how this is normally exercised (e.g. via PhoneNumberMatcher::Match/Find) if a regression test is wanted here.