[PATCH 3 of 4] diff: accept "dict" as diff option

[Yuya Nishihara]

On Sun, 17 Jun 2018 17:47:11 +0200, Boris FELD wrote:
> On 15/06/2018 13:44, Yuya Nishihara wrote:
> > On Fri, 15 Jun 2018 10:48:34 +0200, Boris FELD wrote:
> >>> I'd rather force callers of ctx.diff() to build a diffopts object.
> >> Do you see a way to solves the import cycle?
> > Which one?
> 
> +  Import cycle: mercurial.obsutil -> mercurial.patch -> 
> mercurial.scmutil -> mercurial.obsutil
> +  Import cycle: mercurial.copies -> mercurial.scmutil -> 
> mercurial.obsutil -> mercurial.patch -> mercurial.copies
> +  Import cycle: mercurial.obsolete -> mercurial.obsutil -> 
> mercurial.patch -> mercurial.scmutil -> mercurial.obsolete

IIUC, the cycle is caused by the following function call:

  obsolete.createmarkers() -> obsutil.geteffectflag() -> _cmpdiff()

and I think createmarkers() is more like a utility function than a storage
function. So, if we had a higher-level utility module for obsolete users,
which isn't imported from scmutil, maybe we can move createmarkers() and
geteffectflag() to that module. We'll also have to move scmutil.cleanupnodes()
to that module.

  <new utility module> -> scmutil / patch

Another option will be just delaying the import of the patch module, and add
TODO comment.