[PATCH STABLE] merge: mark .hgsubstate as possibly dirty before submerge for consistency

[FUJIWARA Katsunori]

At Thu, 29 Jan 2015 16:00:46 -0600,
Matt Mackall wrote:
> 
> On Fri, 2015-01-30 at 05:15 +0900, FUJIWARA Katsunori wrote:
> > # HG changeset patch
> > # User FUJIWARA Katsunori <foozy at lares.dti.ne.jp>
> > # Date 1422561545 -32400
> > #      Fri Jan 30 04:59:05 2015 +0900
> > # Branch stable
> > # Node ID 6becb9dbca25057c6186e255a48dd2c2ce5701a5
> > # Parent  8a544fb645bb00995f5a0836b3a8aab0852c3e49
> > merge: mark .hgsubstate as possibly dirty before submerge for consistency
> 
> Queued for stable, thanks.

I just focused on ".hgsubstate", because I accidentally met a test
failure for ".hgsubstate" inconsistency.

But after posting this patch, I hit on that this may not be safety
enough, because other normal files can also satisfy these conditions
at any failures (also other than "submerge": for example, I/O error at
writing merge status out)

In fact, should we do "normallookup" also for files to be
updated("g")/merged("m") for consistency, as a prevent for failures ?

> -- 
> Mathematics is the supreme nostalgia of our time.
> 
> 
> 


----------------------------------------------------------------------
[FUJIWARA Katsunori]                             foozy at lares.dti.ne.jp