While working on 64-bit block status, I tried using nbdsh to run a
known-failing h.block_status() in a loop, and was surprised that
memory ran away, even though I remember having worked on plugging
python memory leaks a while back.  So here's a couple more places that
need plugging.
Patch 1 is ready to go. Patch 2 helps, but is incomplete; I'm posting
it now because it's my bedtime.
Eric Blake (2):
  api: Avoid memory leak on certain strict_mode failures
  python: Fix memory leak on callback error
 generator/Python.ml |  4 ++--
 lib/rw.c            |  4 ++--
 tests/errors.c      | 45 ++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 48 insertions(+), 5 deletions(-)
-- 
2.33.1