Changes to error offsets, fixing inconsistencies

[NWilson]

Part of #549

[jirislaby]

If I pass ".*.*$*X" to pcre2_compile(), I get erroroffset at 6 (X) now. It used to be 5 (*). Is this expected?

[NWilson]

In Perl, if I try the regex:

my $regex = qr/.*.*$*X/;

The error message is $* is no longer supported as of Perl 5.30 which contains no character offset.

However if I try the (similar) my $regex = qr/.*.*\K*X/ the error message is:

\K* matches null string many times in regex; marked by <-- HERE in m/.*.*\K* <-- HERE X/

Perl's error offset is consistently pointing to the caret position in between the * and X.

In pcre2test the same test gives:

$ ./pcre2test
PCRE2 version 10.48-DEV 2025-10-21 (8-bit)
  re> /.*.*$*X/
Failed: error 109 at offset 6: quantifier does not follow a repeatable item
        here: .*.*$* |<--| X

I am satisfied that this is consistent with the way Perl reports error messages, with the offset just to the right of the error.

The change was intentional. I apologise if this has caused problems for your application.

In a future version of PCRE2, I plan to add a function to retrieve the "error direction" to improve on the current "error offset". The error direction would indicate to text-based applications whether the error is <-- HERE to the left of the offset, or HERE --> to the right. I have considered adding a start/end position to the offset (for highlighting), but I have not heard any demand for that yet.

[jirislaby]

The change was intentional. I apologise if this has caused problems for your application.

Fine by me. Except the reporting has to be different for 10.47+ and earlier versions...