Fwd: Re: hg convert from cvs bug ?

[Mads Kiilerich]

  Henrik,

In this case 6019e6517f95 "convert: better support for CVS branchpoints 
(issue1447)" looks like a regression. As long as branches isn't 
considered the conversion was better before the change.

Would it be possible to optimize for both cases, or would it make sense 
to add an option for restoring the old behavior?

Or is it just a bug in the branchpoint detection?

/Mads


-------- Original Message --------
Subject: 	Re: hg convert from cvs bug ?
Date: 	Thu, 01 Jul 2010 17:12:34 +0200
From: 	Mads Kiilerich <mads at kiilerich.com>
To: 	Dongsheng Song <dongsheng.song at gmail.com>
CC: 	mercurial at selenic.com



  Dongsheng Song wrote, On 07/01/2010 09:12 AM:
>  When I try to convert FreeTDS, I found hg convert not use any
>  cvsps.fuzz setting:

fuzz works fine - but it only handles time fuzz.

>  For example:
>
>  $ hg log
>  ...
>  changeset:   2:449bdfd47711
>  branch:      camber
>  user:        brianb
>  date:        Fri Oct 12 23:28:40 2001 +0000
>  summary:     initial checkin
>
>  changeset:   1:e7896f690dbf
>  branch:      camber
>  user:        brianb
>  date:        Fri Oct 12 23:28:40 2001 +0000
>  summary:     initial checkin
>
>  changeset:   0:556e7c897126
>  branch:      camber
>  user:        brianb
>  date:        Fri Oct 12 23:28:40 2001 +0000
>  summary:     initial checkin
>
>  Why these commit of 'initial checkin'  not in the same changeset ?
>
>  Here is my convert script:
>
>  export VCS_ROOT=/home/dongsheng/vcs
>  export
>  CVS_ROOT=:pserver:anonymous at freetds.cvs.sourceforge.net:/cvsroot/freetds
>
>  cd ${VCS_ROOT}/cvs&&  cvs -d ${CVS_ROOT} co -d freetds freetds
>  cd ${VCS_ROOT}/hg&&  hg convert --datesort ${VCS_ROOT}/cvs/freetds freetds

The reason is that the logentries have different branchpoints associated
(for example alternating BRANCH0_60 and none) and they thus are
considered to belong to different changesets.

See http://selenic.com/hg/file/34e33d50c26b/hgext/convert/cvsps.py#l496

/Mads