Hi Marek,

I was wrong. The validation function overlooked the fact that the initial allocation of second block for the directory can be eventually returned back to the available list when the directory is expanded during bucket splitting. So indeed, the correct
address check is dbf->dir[dir_index] >= dbf->header->block_size. I have fixed this in commit 64ef7e82. Please confirm that it fixes the testsuite. Thanks a lot for your help!

Thanks for testing it. Test is failing for me with same gdbm and qsf version (with no downstream patches in both cases).

It happened to me occasionally that it passed. But really rarely - for the first time I tried it:-) And then it failed in all my attempts...

So please try it repeatedly. (sorry, I know that it's not a specific instruction)

I'm trying it in Fedora Rawhide (gcc-8.2.1,glibc-2.28,systemd-239). Where did you tried it?

If its different distribution. Different compiler or libraries could cause the issue. So I can try to prepare a reproducer in Fedora for you... (could it be in Docker container?)

Hi,
updated gdbm 1.15 (and later) fails qsf testsuite. More info can be found in [1] (maybe also related to debian bug report [2])

Fedora qsf maintainer bisected the gdbm code and found, that the issue is caused by commit http://git.savannah.gnu.org/cgit/gdbm.git/commit/?id=655cd193549e20ea8a8e77125adec7c5909c067e

Is it possible that '2*' in gdbm_dir_entry_valid_p function is wrong? With 'dbf->dir[dir_index] >= dbf->header->block_size;' test suite is passing again.

Thanks,
Marek

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1607290
[2] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=904006