large_file: need 4+ GB file.  (Could be sparse, though.)


Add tests for filesystem types other than ext2.  Adding support for the others
shouldn't be too hard.  Obviously need to search for mke2fs, e2fsck, tune2fs
and adapt acordingly.  Probably need to change how scribble.sh creates
fragmented files, too (i.e. re-taylor for the allocation policy of whatever
fs).  Should rename fs-dependent scripts to have e2- prefix or something.  The
fs image file naming scheme will need tweaking.


Test on big-endian machine, preferably both powerpc and non-powerpc (see
comment near top of e2dump.c).


Create a fs image with really long file name (including several leading
directories), and compile with checkergcc or other buffer-overrun checker.
(ElectricFence probably wouldn't work, since it traps only malloc, and
probably won't catch overrunning stack arrays.)  There are several places in
defrag that use fixed-size buffers for formatting strings.


Create filesystems that use ext2 compression.  (Unfortunately this requires
appropriate kernel support whenever the images are regenerated using top.sh.)


Try using an inode priority file.  (Hmm, how do we check that the priority
requests were honoured?  Still, at least we'd have the tar & fs image cmp
tests.)


Go for code coverage generally.  (See gcov in the gcc documentation.)
