[akashdeshmukh-git]

What was wrong

When pcre2_match() fails early (e.g. PCRE2_ERROR_BADOFFSET, UTF validation failure), match_data->code is not assigned. Calling pcre2_substring_length_byname(), pcre2_substring_copy_byname(), or pcre2_substring_get_byname() subsequently dereferences a NULL code pointer in pcre2_substring_nametable_scan(), causing a segmentation fault or assertion failure.

How to reproduce

1. Call pcre2_match() with arguments causing an early error return (such as start_offset > length).
2. Call pcre2_substring_length_byname() with the resulting match data block.
3. Observe segmentation fault in pcre2_substring_nametable_scan().

What changed

In pcre2_substring_copy_byname(), pcre2_substring_get_byname(), and pcre2_substring_length_byname(), check match_data->rc < 0 before scanning the nametable and return the match error code immediately, matching the behavior of the �ynumber functions. If match_data->code is NULL, return PCRE2_ERROR_NULL.

How it was tested

Added unit test in src/pcre2test_inc.h asserting that calling copy_byname, get_byname, and length_byname after an early match failure returns PCRE2_ERROR_BADOFFSET without crashing.

Fixes #1049

[NWilson]

Thanks for opening a PR!

It looks like you've somehow committed a whole-file edit with the wrong line endings.

[akashdeshmukh-git]

Apologies for the noise! The line endings were inadvertently converted to CRLF on Windows. I have pushed a clean commit with unix LF line endings so only the intended 35 lines are changed.

[NWilson]

Thank you. The check for rc is good.

However, I would prefer the nullness checks to be done as preconditions (assertion-only).

/* Preconditions on argument values */

...

/* Preconditions on argument state */

PCRE2_ASSERT(match_data->code != NULL);

...

[akashdeshmukh-git]

Thanks, done in 38321c5. The NULL checks are now PCRE2_ASSERT(match_data->code != NULL) under a "Preconditions on argument state" comment. I placed it after the rc < 0 check rather than before, because an early match failure (e.g. bad offset) leaves code unset with a negative rc, which is the case this PR fixes.