This script will do the job, at least as long escape sequences are as expected. Call it e.g. as git-diff-word-context -U 5. It accepts all arguments git diff would accept and intercepts -U (respectively --unified=) to set the number of characters.
It just utilises grep to fulfil the task:
#!/usr...
The solution is generic enough. You can just tell Git what a word ist. Have a look at option --word-diff-regex for git diff and friends.
To be more specific: For example, --word-diff-regex=. will treat each character as a word and, correspondingly, show differences character by character.
Knu, I added two characters to the script to fullfill the by character requirement.
To be honest, I would limit words to not cross lines (i.e. to use --color-words='[^\n]') instead. I just used . in the updated answer because you did the same in your question, thus I imply it's your intention.
You already can do. Just call the script like this: git-diff-word-context --color-words=<arg>. This will overwrite script default. The script just passes all arguments to git diff with the exception of -U. Or you use it as a filter: git show HEAD~5 --color-words=<arg> | git-diff-word-context.
I have found a bug if changed words (currently characters) overlaps with context. I will fix it in the answer. Do you want to get the changes listed one by one or combined in that case? Example with `-U5`: `TestTextByMe` -> `TestTEXTByME` a) One change, i.e. both changes combined.
Discussion between doak and Knu
Discussion between doak and Knu