[PATCH 6 of 6 V2] dirstate: read from pending file under HG_PENDING mode if it exists

[FUJIWARA Katsunori]

At Tue, 13 Oct 2015 12:30:44 -0700,
Pierre-Yves David wrote:
> 
> On 10/13/2015 10:54 AM, FUJIWARA Katsunori wrote:
> > # HG changeset patch
> > # User FUJIWARA Katsunori <foozy at lares.dti.ne.jp>
> > # Date 1444758557 -32400
> > #      Wed Oct 14 02:49:17 2015 +0900
> > # Node ID 68fdb7150daad2ea8747b65145d00acab9be01de
> > # Parent  f452ff0cefe2755caff781f95bd135038e3d4776
> > dirstate: read from pending file under HG_PENDING mode if it exists
> 
> This series is pushed to the clowcopter. Does this conclude your 
> dirstate run?

Even after this series, there are some situations that in-memory
changes (not only for dirstate, but also for changelog and so on)
aren't visible to external hook/commit log editor process.

I have some patches to make in-memory changes visible to all related
external process (even though they don't change dirstate itself).


> Do we have any mechanism to clean up a stalled .pending file? this 
> should like something we should guard ourself against.

Oh, I overlooked the case that temporary files ('.pending' or so) are
accidentally left.

How about this ?

  (1) check existence of 'journal.backupfiles' before starting
      transaction (maybe in 'transaction.__init__' ?)

  (2) unlink files registered in it, if it exists (= meaning failure
      of cleanup at preceding transaction)

These functions can be easily factored out from 'rollback' and
'_playback' in transaction.py.

BTW, this method requires some additional preparations, because
current unlinking 'journal.backupfiles' logic in transaction.py
prevents this file from being used as "failure of cleanup at preceding
transaction" mark, AFAIK.


> -- 
> Pierre-Yves David
> 

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