[mpich-devel] old old MPICH history now on line

[Rob Latham]

On 07/15/2015 04:01 PM, Dave Goodell (dgoodell) wrote:
> On Jul 15, 2015, at 3:35 PM, Rob Latham <robl at mcs.anl.gov> wrote:
>
>> Did you ever run 'git blame' to find out why some part of MPICH does what it does, only to find the commit responsible was 6a1cbdcfc42 "Added all components in the right places."?  I'm personally surprised how often I care about changes made more than 8 years ago.
>>
>> If you, like me, sometimes find yourself needing more history, you can now -- ta-da -- consult http://git.mpich.org/mpich-CVS.git/
>
> Rob,
>
> Nice!  I don't generally need MPICH history that often these days, but still a win in the ongoing struggle to extirpate CVS, not to mention making the CVS history available to folks without MCS accounts.
>
> If you want to be really slick, you could try grafting in this old history with git-replace in a local repo and then publishing those graft instructions: https://www.kernel.org/pub/software/scm/git/docs/git-replace.html
>

wes, I considered rebasing MPICH onto this old history, but decided 
against breaking every single feature branch in the world.

Dave, thanks for the pointer to git replace.

a sufficiently new 'git replace' has --graft, which  sounds like what I 
want.  If you only have git-1.9.1 (as on my Ubuntu-14.04 machine), you 
have to simply stomp on the initial "put everything in its place" 
commit.  But what to stomp it with?

There are two post-conversion commits to the CVS repository.  They were 
made in 2009 so they are easy to spot.  Instead, I replaced it with 
7e0e4d706c:

$ git replace 6a1cbdcfc 7e0e4d706c32cd73e

And now take a look at an old file.  Let's say 
src/mpid/common/sock/poll/sock_wait.i (which I was hacking on this morning):

f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 567) #undef FUNCNAME
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 568) #define FUNCNAME 
MPIDU_Socki_handl
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 569) #undef FCNAME
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 570) #define FCNAME 
MPIU_QUOTE(FUNCNAME
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 571) static int 
MPIDU_Socki_handle_read
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 572) {
94b5535c6 (Rob Latham    2015-07-16 10:44:16 -0500 573)     ssize_t nb;
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 574)     int 
mpi_errno = MPI_SUCCESS;
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 575) 
MPIDI_STATE_DECL(MPID_STATE_RE
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 576) 
MPIDI_STATE_DECL(MPID_STATE_RE
f96fd2d95 (toonen        2004-03-12 05:36:31 +0000 577) 
MPIDI_STATE_DECL(MPID_STATE_MP

Awesome!  we have been stuffing a ssize_t value into an int for 11 years 
until I fixed that just this morning.

The data pack-rat in me is immensely pleased to have full history now.

==rob

-- 
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA