How to use "cmp" to compare two binaries and find all the byte offsets where they differ?

I think cmp -l file1 file2 might do what you want. From the manpage:

-l  --verbose
      Output byte numbers and values of all differing bytes.

The output is a table of the offset, the byte value in file1 and the value in file2 for all differing bytes. It looks like this:

4531  66  63
4532  63  65
4533  64  67
4580  72  40
4581  40  55
[...]

So the first difference is at offset 4531, where file1's decimal octal byte value is 66 and file2's is 63.

Method that works for byte addition / deletion

diff <(od -An -tx1 -w1 -v file1) \
     <(od -An -tx1 -w1 -v file2)

Generate a test case with a single removal of byte 64:

for i in `seq 128`; do printf "%02x" "$i"; done | xxd -r -p > file1
for i in `seq 128`; do if [ "$i" -ne 64 ]; then printf "%02x" $i; fi; done | xxd -r -p > file2

Output:

64d63
<  40

If you also want to see the ASCII version of the character:

bdiff() (
  f() (
    od -An -tx1c -w1 -v "$1" | paste -d '' - -
  )
  diff <(f "$1") <(f "$2")
)

bdiff file1 file2

Output:

64d63
<   40   @

Tested on Ubuntu 16.04.

I prefer od over xxd because:

• it is POSIX, xxd is not (comes with Vim)
• has the -An to remove the address column without awk.

Command explanation:

• -An removes the address column. This is important otherwise all lines would differ after a byte addition / removal.
• -w1 puts one byte per line, so that diff can consume it. It is crucial to have one byte per line, or else every line after a deletion would become out of phase and differ. Unfortunately, this is not POSIX, but present in GNU.
• -tx1 is the representation you want, change to any possible value, as long as you keep 1 byte per line.
• -v prevents asterisk repetition abbreviation * which might interfere with the diff
• paste -d '' - - joins every two lines. We need it because the hex and ASCII go into separate adjacent lines. Taken from: Concatenating every other line with the next
• we use parenthesis () to define bdiff instead of {} to limit the scope of the inner function f, see also: How to define a function inside another function in bash

See also:

• https://superuser.com/questions/125376/how-do-i-compare-binary-files-in-linux
• https://unix.stackexchange.com/questions/59849/diff-binary-files-of-different-sizes

The more efficient workaround I've found is to translate binary files to some form of text using od.

Then any flavour of diff works fine.