Re: [Gnu-arch-users] Re: Future of GNU Arch, bazaar and bazaar-ng ... ?

[Magnus Therning]

On Tue, Aug 23, 2005 at 08:45:07AM +0800, Lalo Martins wrote:
>And so says Matthieu MOY on 22/08/05 21:39...
>> darcs has finer grain than this. You can commit just a hunk for a file,
>> i.e. if the diff is
>> 
>> @@...@@
>>  foo
>> -bar
>> +baz
>>  foobar
>> @@...@@
>>  XXX
>> -YYY
>>  ZZZ
>> 
>> you can commit only the "-YYY" if you whish.
>
>IMHO, hunk-level granularity for commits is a serious mistake.
>
>Since you're committing a version of the code that you never actually
>had on your machine, you can't possibly have tested it.
>
>I think it encourages errors.

It is however a quite common use case, especially among Linux kernel
developers:

 1. X submits a patch to the mailing list.
 2. Linus, or someone else, likes the patch, but thinks it's too big.
 3. X has to split up the patch in smaller parts.

It'd be nice if it were possible to do the following:

 1. Branch off just before the large patch.
 2. Apply the patch.
 3. Split up the patch by repeatedly committing a few hunks until the
    entire patch is commited on the branch.
 4. Repost to the kernel mailing list the new series of patches.

BTW, this sort of thing has been refered to as shelve/unshelve on the
Bazaar-NG mailing list.

I think your comment on testing is very valid, but it can be addressed
through the way shelve/unshelve is implemented. In arch it could be done
by having the user undo local changes and then selectively redo parts of
what was undone. Each part that is redone should be tested and committed
before undoing the next part.

/M

-- 
Magnus Therning                    (OpenPGP: 0xAB4DFBA4)
address@hidden
http://therning.org/magnus

Software is not manufactured, it is something you write and publish.
Keep Europe free from software patents, we do not want censorship
by patent law on written works.

Tragedy purges the mind of trivia.
     -- George Gilder

pgpqpMQvsVXDX.pgp
Description: PGP signature