On Sat, Apr 22, 2017 at 1:46 AM, Jeff King <peff@xxxxxxxx> wrote: > We generally disallow null sha1s from entering the index, > due to 4337b5856 (do not write null sha1s to on-disk index, > 2012-07-28). However, we loosened that in 83bd7437c > (write_index: optionally allow broken null sha1s, > 2013-08-27) so that tools like filter-branch could be used > to repair broken history. Uh oh.. cache-tree. > However, we should make sure that these broken entries do > not get propagated into new trees. For most entries, we'd > catch them with the missing-object check (since presumably > the null sha1 does not exist in our object database). But > gitlink entries do not need reachability, so we may blindly > copy the entry into a bogus tree. Phew.. not another bug of mine :D > When merged to pu, this fixes the existing test breakage in t7009 when > GIT_TEST_SPLIT_INDEX is used (because the split index didn't rewrite the > whole index, "git rm --cached" didn't always barf). Latest 'pu' has your patch, but t7009 still fails on me (with "invalid object" error), more on this later.. > But I think it's worth doing on its own merits, as demonstrated by the new tests. Agreed. The patch looks correct. Just checking, since cache-tree helps speed up many operations, dropping cache-tree can have some performance implication. But this must be an error case (null sha1) and we will not run into it often to worry about unnecessary dropping, correct? > The one thing I haven't figured out it is why the new test in t7009 > fails with the split-index. And even more curiously, the new tests in > t1601 _don't_ fail with it, even if I instrument the fake index to have > more entries (making it more likely to split). back to t7009 failure. I'll see if I can look more into this this weekend. If split-index somehow produces these null sha1, then I probably have a problem. Thanks for looking at it anyway. One bug down. Thousands to go... BTW, I ran t7009 with valgrind and it reported this. Is it something we should be worried about? I vaguely recall you're doing something with prio-queue... ==4246== Source and destination overlap in memcpy(0x5952990, 0x5952990, 16) ==4246== at 0x4C2EACD: memcpy@@GLIBC_2.14 (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so) ==4246== by 0x545D05: swap (prio-queue.c:15) ==4246== by 0x545D72: prio_queue_reverse (prio-queue.c:25) ==4246== by 0x4CBC0C: sort_in_topological_order (commit.c:723) ==4246== by 0x574C97: prepare_revision_walk (revision.c:2858) ==4246== by 0x48A2BA: cmd_rev_list (rev-list.c:385) ==4246== by 0x405A6F: run_builtin (git.c:371) ==4246== by 0x405CDC: handle_builtin (git.c:572) ==4246== by 0x405E51: run_argv (git.c:624) ==4246== by 0x405FF3: cmd_main (git.c:701) ==4246== by 0x4A48CE: main (common-main.c:43) -- Duy