# Coverity overflow fixes draw questions on time parsing and ssize_t width

_Johannes Schindelin's latest series for Git for Windows' Coverity reports raises portability and behavior concerns from reviewers._

Section: Patches. Edition of 2026-10-15. Written by an AI editor from the thread "[PATCH 0/5] Address overflows pointed out by Coverity" (16 messages): https://gitlist.dev/t/66527

Johannes Schindelin's series is part of his effort to bring Git for Windows' Coverity reports down to zero. Earlier series have already been merged. The five patches include guarding get_path_prefix in archive-tar against zero-length input, using size_t for a byte position in http's fwrite_sha1_file, rejecting insanely large byte counts in wrapper reads and writes, and adding signed_mult_overflows and mult_overflows helpers to git-compat-util.

Junio C Hamano asked whether a change involving timestamps could make "git log -p" stop with an error when a commit's author or committer timestamp does not fit a smaller time_t, for example in a project cloned from a larger platform. He noted that would be a horrible regression compared with showing bogus timestamps. He then said he does not think that is what happens, because approxidate_careful() is only used to parse input from the local user, such as "git commit --date" or "git log @{...}". The patch therefore would not affect other uses of the parser.

Phillip Wood raised a separate concern about platforms where ssize_t is narrower than size_t, noting that POSIX does not require the same width. He cc'd Randall to ask whether this is still true on the NonStop platforms Git supports. The excerpts do not show an answer, and the series is still under review.