Fix ref counting bugs in race condition correction code.

When we attempt to insert a key/value pair into the store, we have
to allow for the possibility that a racing thread may have already
inserted an equivalent key/value. We have special code in place to
handle this eventuality; if we spot an existing entry, we take the
existing one in preference to our new key/value pair.

This means that fz_store_item needs to take a new reference to any
existing thing it finds before returning it.

Currently the only store user that is exposed to this possibility
is pdf_image; it spots an existing tile being returned, and was
inadvertently double freeing the key.

0 files changed, 0 insertions, 0 deletions