Re: [Axiom-developer] patches

[Ralf Hemmecke]

On 03/28/2006 01:15 PM, Page, Bill wrote:
> On Tuesday, March 28, 2006 5:40 AM Ralf Hemmecke wrote:
> > Why don't you use tla?
> >
> > Make a new branch from axiom--main--1, hack at the new branch 
> > and if it is finished, tell Tim that there is something new
> > available ready for merge. Tim then could investigate that
> > branch, test and check documentation, and if he finds that
> > it is all fine just do a star-merge to axiom--main--1.
> > Would that be bad?

> As far as I am concerned that would be Excellent! :)

> Unfortunately skill and confidence in these tools is lacking.
> I think it needs someone (or a small group of Axiom developers)
> to take the lead, write **simple** documentation and demonstrate
> exactly how this works.

Well, it took me some time to figure out a way how that could be done.
Bill, you know that I am currently in the process of translating 
Broadbery's Java code to Aldor so that "Aldor for Axiom" could be built 
without the need to install Java.

I also volunteer to write the documentation on the wiki related to this 
tla stuff. Once one understands the basic things, it becomes easy. But I 
wanted to commit to a local archive (without internet access) and then 
merge with the one at axiom-developer.org. That was a bit trickier to 
find out. But I think I have an acceptable way (simply create a local 
personal archive).

> I have not used 'tla star-merge', but I have used the similar
> function of 'darcs pull'.

I have also quickly read through darcs. The idea that in darcs an 
archive is just a collection of patches, sounds rather simple. 
Star-merge in GNU arch is what it says. You branch from a certain 
revision. Then two development lines appear on two different branches 
(which have a common ancestor) where each of the branch maintainers can 
merge from the other branch. Star-merge keeps track of which patches 
have already been applied so that one patch is not applied several 
times. That is all.

> It seems to me that these commands
> save an enormous amount of time and protect against the common
> errors of less automated methods of handling patches. The
> decentralized model based on pulling (merging) patches from
> separate archives also seems like a good idea. This allows the
> work of maintaining an archive to be shared in a flexible
> manner. Plus the 'darcs send' command can be used to create
> and send patches by email with very little effort.

I also think that we should use the version control tools more 
excessively, but unfortunately it takes a lot of time until one gets 
accustomed to their different principles.

If you are accustomed to centralised development like in CVS, then it's 
at first a bit hard to get the idea of how else development could be 
done. Nobody writes, for example, that it is perfect, if you create your 
one archive, JUST FOR YOURSELF, work on it and from time to time (if 
things are reasonably stable), merge back to the version you branched from.

So there is one suggestion from my side. The version axiom--main--1 
should be made read-only. That is declared to be THE stable branch and 
only Tim (or any person that he trusts) has write access to it.

For anything else, people should branch from that one, create there own 
private (or public on axiom-developer.org) archive, produce new code, 
tell Tim, so that Tim could decide whether the changes should go into 
axiom--main--1.

People (like me) feel more confident if they know that they don't 
destroy the work of others if they start hacking on a new idea.

You hear/read more of that probably soon.

All the best
Ralf