Merge on push?

[John D. Mitchell]

On Wednesday 2008.06.11, at 01:12 , Martin Geisler wrote:
[...]
> Now r2 doesn't correspond to the view developer B had in his working
> copy when he committed! And he tested a revision (r2 - changes in r1)
> which cannot easily be reproduced. That is quite messy...

Yes and it can get very dangerous when doing deployments out of the  
repositories.  It subverts [sorry, couldn't resist] the protections  
given by the SCM system by bastardizing the situation halfway back to  
the old manual version control model.

All of the uses of this that I've seen (suffered with :-) in practice  
are basically very clear signs that the SCM processes are broken. [In  
fact, IMHO, the "necessity" of this in Subversion is also proof that  
svn's model is broken but that's a separate discussion.]  There seems  
to be this belief that powerful SCM processes must a priori be big,  
complex, hard to manage, etc. but that's a false conclusion based on  
bad SCM tools and the experience too many developers have had  
suffering under big, complex, hard to manage, stupid SCM processes  
created by people who didn't know what the hell they were doing and/or  
being driven to such processes by business people who micromanaged/ 
dictated such stupidity.

Take care,
John