Conflict-marker diagnostic falsely rejects heredoc content

[fglock]

Summary

PerlOnJava incorrectly reports a version-control conflict marker when a Perl heredoc contains a legitimate run of = characters. This is a parser regression that breaks Text::FormatTable v1.03.

CPAN evidence

• CPAN run: 20260918-141920-96054
• Distribution: Text::FormatTable v1.03
• System Perl: PASS — 5 tests
• PerlOnJava: FAIL during compilation
• Both JVM and interpreter backends reproduce the failure

The upstream test contains a heredoc with expected table output. One line consists of a run of equals signs:

my $shouldbe = <<'END';
 a| b     c
=================
...
END

PerlOnJava aborts before the test can run:

Version control conflict marker at test.pl line 21, near "============="

The same test passes under standard Perl.

Root cause

The conflict-marker diagnostic added by commit 2d3c36bb2355 treats the equals-sign content inside the heredoc as a CONFLICT_MARKER token. The diagnostic is intended for actual source-level merge markers such as a standalone ======= line, but it is being applied to literal heredoc content where the text must remain part of the string.

The failure is therefore independent of Text::FormatTable behavior; the module's test data happens to contain the triggering text.

Reproduction

Run the Text::FormatTable v1.03 test suite with PerlOnJava. The test.pl program fails at the heredoc line containing the table rule. A minimal equivalent reproducer is:

my $text = <<'END';
=======
END

This must compile as a string literal and must not raise a conflict-marker diagnostic. The archived CPAN run reproduces the failure on both execution backends.

Expected behavior

Conflict-marker diagnostics should not inspect or tokenize the contents of quoted strings, including heredocs. Literal sequences of seven or more <, =, or > characters inside a string must remain string data.

Actual source-level conflict markers should continue to receive the intended diagnostic when they occur in a context where Perl treats them as source rather than string content.

Requested fix

• Restrict conflict-marker detection to valid source-level marker positions outside quoted strings and heredocs.
• Preserve the existing diagnostics for genuine conflict markers.
• Add a focused parser regression test covering equals-sign runs in heredocs and quoted strings.
• Rerun Text::FormatTable v1.03 and the existing conflict-marker diagnostic tests on both JVM and interpreter backends.