The series is part of Johannes Schindelin's effort to bring Git for Windows' Coverity reports down to zero. The cover letter lists several earlier Coverity-related series already merged. This one has five patches: guarding get_path_prefix in archive-tar against zero-length input, using size_t in http's fwrite_sha1_file, rejecting huge byte counts in the wrapper read/write helpers, and adding signed_mult_overflows and mult_overflows helpers to git-compat-util.
Phillip Wood reviewed the series in detail, and Schindelin has answered several points. On the length check, Schindelin agreed that a length cannot be negative and called his approach overzealous. He replaced the logic with manual overflow checks that are easier to review, including a plain comparison against INT_MAX that returns NULL. He also acknowledged that the GCC-only logic had not been a good idea.
On the new multiplication-overflow macros, Schindelin conceded that Wood's review had effectively shown the approach would not work. He said he had assumed the behavior came from the C standard when it was introduced in C23. He dropped the macros from v2.
The wrapper patch changed the most. Schindelin said write_in_full() returns the same count the caller passed in on success, modulo truncation to ssize_t, which is not useful, so it now returns 0 on success. Many read_in_full() callers only care whether the read succeeded, so they move to a new read_in_full_or_error(), which returns 0 on success and -1 on error. His message is cut off before it finishes describing what happens to callers that do need the byte count.
Harald Nordgren, Patrick Steinhardt, Junio C Hamano and Simon Richter also took part in the thread. The excerpts do not show their positions. The reworked v2 still needs review.