[PATCH] merge: add resolve --prep, which outputs conflicted comparison files

[Durham Goode]

On 2/24/17 4:04 PM, Jun Wu wrote:
> Excerpts from Durham Goode's message of 2017-02-24 15:42:34 -0800:
>> Maybe I don't understand your proposal.  The current merge-tools are
>> invoked once per file.  Would this python merge-tool be invoked in some
>> other way where it's given the full conflict state at the beginning then
>> it can do what it wants?  So we'd special case python merge tools to be
>
> Actually it's a good question. I think it's more consistent if being called
> per-file. But I see that'll be problematic about "when is it the last file?"
> That probably requires an extra parameter about how many files are left.
>
>> a little different?  We could just define a internal merge tool called
>> 'internal:json' or something and skip the whole extensibility thing.
>
> That's possible. Although I'm not sure if it's a good idea or not. Since we
> have ":dump" already, maybe name it something like ":dump-json". I guess the
> most difficulty part to reach agreement is to define the json schema - the
> merge state includes things like executable bit, symlink, etc. So I'd rather
> to put this outside core from the beginning.

If it ends up outside core, it's such a minor feature we'll likely never 
do the work to move it in to core. I don't think the json schema will be 
that big of a deal. The kinds of data we need are pretty clear: file 
content at various versions, file flag at various versions, kind of 
conflict.

>
>> Also, how does the merge tool communicate to the caller the json
>> results? Write it to a file? Or does the merge tool have the ability to
>> print to stdout?
>
> It has ui and repo.wvfs so it could do whatever it wants.
>
> Therefore the signature could be something like:
>
>   def mergefunc(ui, repo, basefctx, localfctx, otherfctx, index, total):
>       raise ... # merge failure
>       return None # alternative, merge failure
>       return fctx-like # merge success
>
> If we want to do the ":dump-json" thing, it could be like:
>
>   state = {}
>   def merge(....):
>       if index == 0:
>           state.clear()
>       state[basefctx.path()] = ....
>       if index == total:
>           ui.write(json.dumps(state))
>       return None # report failure

What if we drop the --prep and just introduce '--tool 
internal:dump-json' as the switch that gets us to this code path.  So 
that gets it in core without needing much of a rewrite. If you're 
worried about deciding on a schema, we can mark it as experimental in 
the documentation.