The cover letter points to two earlier series. jk/xdiff-size_t replaced mmbuffer_t, which used a long for its size, with mmfile_t, which uses size_t. kn/safer-mem-arithmetic-in-xdiff computes the merge buffer size as a size_t with overflow checks. Because the second targets the maint branches, which lack the first, it also makes xdl_do_merge() fail when the result exceeds LONG_MAX, so the size cannot be silently truncated on LLP64 platforms. On master that check is no longer needed for mmfile_t.

The first patch removes the check. The commit message notes that the merge result size was still passed to odb_write_object() and write_buf_to_worktree(), which accepted unsigned long, so the remaining patches change those types.

The second patch changes odb_write_object() and odb_write_object_ext() to use size_t, since nearly everything they call, including hash_object_file(), format_object_header(), convert_object_file(), the write_object callback and git_zstream, already does, and many callers pass a size_t such as strbuf.len.

The third patch changes write_buf_to_worktree() in notes-merge.c, whose two callers both pass a size_t. According to the message, since c6a4629e32 the first caller reads the note into a size_t, so on 64-bit Windows a delete/modify conflict on a note of 4 GiB or more would write a truncated note into NOTES_MERGE_WORKTREE.

No review replies are shown yet in the excerpts.