It feels like diff can do a better job on not committed changes.
$ dd if=/dev/random of=small.txt bs=1024000000 count=1 # modify the file
$ dvc add small.txt
$ dvc diff -t small.txt HEAD # IT SUPPOSE TO BE DIFFERENT FROM HEAD
dvc diff from 2dae4a2 to 2dae4a2 # IT IS NOT :(
$ dvc diff -t small.txt HEAD^
dvc diff from 8a1aa99 to 2dae4a2 . # IT IS NOT DIFFERENT EVEN WITH OLDER VERSION:(
diff for 'small.txt'
-small.txt with md5 b4301ef665d0343e028fcd5821654edc
+small.txt with md5 b4301ef665d0343e028fcd5821654edc
file size was not changed
$ git add small.txt.dvc
$ git commit -m change
$ dvc diff -t small.txt HEAD^ . # NOW YOU CAN SEE THE DIFFERENCE
dvc diff from 2dae4a2 to 12ed32b
diff for 'small.txt'
-small.txt with md5 b4301ef665d0343e028fcd5821654edc
+small.txt with md5 4be464ee7907d9d5e558133907bb3510
added file with size 1.0 GB
It feels like
diffcan do a better job on not committed changes.