On Tue, Aug 30, 2022 at 05:35:31PM +0200, Stefan Sperling wrote:
> On Tue, Aug 30, 2022 at 04:28:07PM +0100, Tom Jones wrote:
> > On Tue, Aug 30, 2022 at 05:19:29PM +0200, Stefan Sperling wrote:
> > > On Tue, Aug 30, 2022 at 02:47:58PM +0100, Tom Jones wrote:
> > > > Prior to this change ed script output was in the wrong order, i.e. in
> > > > the order diff_result provides and changes where missing added or
> > > > changed lines. ed edits need to be in reverse order to keep the edited
> > > > file in sync 
> > > 
> > > As it is, this patch breaks merging on Got, which relies on the
> > > existing output. The existing output is based on the 'ed-style'
> > > script which Caldera diff3(1) generates internally. As you've
> > > discovered this seems to not match traditional diff(1) ed output,
> > > which is unfortunate, and I was unaware of this when I added the
> > > ed output module for the purpose of supporting Got's merging code.
> > > 
> > > So we will need both variants. Should we add a flag?
> > > Or would providing two different ed-output modules be better?
> > 
> > Does the got version include the changed/added lines? The diff.git one
> > didn't, which made me think it didn't work.
> > 
> > In addition to ed script diff also has forward ed (-f) with mangled commands
> > and reverse ed (-n).
> > 
> >      -f --forward-ed
> >              Identical output to that of the -e flag, but in reverse order.
> >              It cannot be digested by ed(1).
> > 
> >      -n      Produces a script similar to that of -e, but in the opposite
> >              order and with a count of changed lines on each insert or delete
> >              command.  This is the form used by rcsdiff.
> > 
> > Is one of these the input that diff3 requires?
> 
> I am not sure which it is.
> I just worked towards what the diff3 code expects, and tested my changes
> by running the Got test suite. The diff3 code used by Got is the same as
> used by OpenRCS and OpenCVS. FreeBSD has this code in usr.bin/diff3/diff3.c.

Weird.

I have done some work on diff3 in FreeBSD recently. I'll have a look
into this and respond with a suggestion for how to handle this, it might
be that diff3 is happy just getting the edit ranges from the diff
program and doesn't validate the actual ed script..

- Tom