[diracdeltafunk]

Description

vec_negate, which subtractInPlace uses to negate each column of the matrix being subtracted, looped while p->next != NULL instead of p != NULL. So it skipped the last stored entry of each column and crashed on an empty column. This broke subtraction of sparse mutable matrices:

k = ZZ/101;
A = mutableMatrix(matrix(k, {{7},{11}}), Dense => false);
B = mutableMatrix(matrix(k, {{2},{3}}), Dense => false);
A - B
A - mutableMatrix(k, 2, 1, Dense => false)

Before: A - B was {{9}, {8}}, and subtracting the zero matrix crashed M2. After: {{5}, {8}} and {{7}, {11}}.

Related issues and pull requests

Fixes #4707.

Testing

Built M2-unit-tests on macOS arm64. Added two engine tests: one for
all coefficients being subtracted and one for an empty column. Against
the original code, the first failed and the second exited with SIGSEGV.
With the fix, both pass via CTest (2/2).

Also checked the example above in M2 and ran all 277 engine unit tests. The second commit adds the new test to the autotools build's Makefile.files; I only built with CMake.

Release notes

Fix incorrect sparse matrix subtraction and a crash on empty columns.

AI disclosure

• 4 — AI produced most or all of the content, which I reviewed thoroughly.

Tools and models used: OpenAI Codex (GPT-6).
How AI was used, and on which parts: Codex investigated the bug, wrote
the fix and tests, ran the checks, and drafted this description.
How I verified the result: I looked at the code

License

• My changes use the existing license of the affected files or package.

Checklist

• This pull request targets the development branch.
• I have reviewed my own changes.
• Documentation and tests are updated as needed.

[dimpase]

isn't NULL to be avoided in C++, and nullptr used instead ? https://en.cppreference.com/cpp/language/nullptr

[dimpase]

ok, but the NULL question ought to be sorted

[diracdeltafunk]

Thanks, changed to nullptr -- I think you're right that this is what you're supposed to do in C++.

Claude identified some more NULLs that could be nullptrs in the codebase. I can ask Claude to hunt these down for another PR.