On Sun, Sep 20, 2020 at 08:54:53AM -0400, Thomas Guyot wrote:
> Considering the issue with MacOS I'm starting to think the best solution
> is to not use any heuristic and read passed-in files directly. That
> said, I don't think it makes much change either way (if I resurrect
> Brian's patch is will probably end up being a hybrid between the two as
> both read the pipe at the same place and my approach was simpler further
> down).
>
> I'm not sure which way I prefer to start first - will you accept a patch
> that reads passed in files as-is if I I start with this one?
I think the ideal is:
- implement a command-line option to read the content of paths on the
command-line literally (i.e., reading from pipes, dereferencing
symlinks, etc)
- make sure we have the inverse option (which you should get for free
in step 1 if you use parse_options)
- flip the default to do literal reads
We'd sometimes wait several versions before that last step to give
people time to adjust scripts, etc. But in this case, I suspect it would
be OK to just flip it immediately. We don't consider "git diff" itself
part of the stable plumbing, and the --no-index part of it I would
consider even less stable. And AFAICT most people consider the current
behavior a bug because it doesn't behave like other diff tools.
-Peff
