[PATCH 0/4] plugging some mmap() leaks

From: Jeff King <peff@peff.net>
Date: Mar 5, 2026, 23:03 UTC
Message-ID: <20260305230315.GA2354983@coredump.intra.peff.net>
In-Reply-To: <20260305220214.GB736322@coredump.intra.peff.net>

On Thu, Mar 05, 2026 at 05:02:14PM -0500, Jeff King wrote:

> Anyway, I think the solution is probably something like the patch above,
> though probably it needs to cover the case where new_pack is NULL.

So here is a more polished version. I decided to try running the whole test suite with leak-checking and NO_MMAP, and it turned up one other case. This series fixes that, too, and then turns on the flag for all leak-checking builds.

  [1/4]: check_connected(): delay opening new_pack
  [2/4]: check_connected(): fix leak of pack-index mmap
  [3/4]: pack-revindex: avoid double-loading .rev files
  [4/4]: Makefile: turn on NO_MMAP when building with LSan

 Makefile        |  1 +
 connected.c     | 38 +++++++++++++++++++-------------------
 pack-revindex.c |  4 ++++
 3 files changed, 24 insertions(+), 19 deletions(-)

-Peff