Darcs really ought to know the difference between a patch that has been
explicitly requested, and one that it has selected because of a depedency. 
Doing so would allow us to tell the user (for example) that we are pushing a
patch because we depend on it, even though it wasn't selected with --match.

See issue333

This is actually a pretty important thing to work on.  It may require an
overhaul of the SelectChanges code

See also
--------
- issue125: I'm resurrecting this because I realise now that it's not *quite*
the same thing as issue125.
- issue762: filed by Tim Chevalier has a rather nice explanation of what we want
(see msg4035)
- issue1123: is vaguely related
- http://wiki.darcs.net/PatchSelectionMechanism

We've recently had a discussion on darcs-users, and a concrete user 
interface has surfaced:

http://lists.osuosl.org/pipermail/darcs-users/2010-August/024817.html

Before
------
 $ darcs rollback -p Hardlink
  Mon Aug  2 04:54:45 BST 2010  Alexey Levan <exlevan at gmail.com>
    * Minimize the number of packed inventories
  Shall I rollback this patch? (1/3)  [ynW...], or ? for more options: y
  Mon Aug  2 03:49:14 BST 2010  Alexey Levan <exlevan at gmail.com>
    * Use cache while getting a packed repository
  Shall I rollback this patch? (2/3)  [ynW...], or ? for more options: y
  Sun Aug  1 06:03:49 BST 2010  Alexey Levan <exlevan at gmail.com>
    * Hardlink files while getting a packed repository
  Shall I rollback this patch? (3/3)  [ynW...], or ? for more options: 

After
-----
  $ darcs rollback -p Hardlink
  Sun Aug  1 06:03:49 BST 2010  Alexey Levan <exlevan@gmail.com>
    * Hardlink files while getting a packed repository

  There are 2 patches that depend on this patch. Hit 'i' to see them.
  Shall I rollback this patch? (3/3) [ynW...], or ? for more options: y
  OK, will also roll back its 2 dependants

There are a couple of details to sort out, such as how you count the 
dependencies (given that a patch may depend on more than one matching 
patch) and how this interacts with patch navigation.  But I suspect 
whoever implements this will only really make progress by trying and 
seeing.

> After
> -----
>   $ darcs rollback -p Hardlink
>   Sun Aug  1 06:03:49 BST 2010  Alexey Levan <exlevan@gmail.com>
>     * Hardlink files while getting a packed repository
> 
>   There are 2 patches that depend on this patch. Hit 'i' to see them.
>   Shall I rollback this patch? (3/3) [ynW...], or ? for more options: y
>   OK, will also roll back its 2 dependants
> 
> There are a couple of details to sort out, such as how you count the 
> dependencies (given that a patch may depend on more than one matching 
> patch) and how this interacts with patch navigation.  But I suspect 
> whoever implements this will only really make progress by trying and 
> seeing.

I like this new direction-- thanks for the team for working on it. My
suggested refinement: Might the prompt also notify you that
dependencies must also be rolled back?

Perhaps this:

 There are 2 patches that depend on this patch and will also be rolled
 back if this patch is selected. Hit 'i' to see them.
 Shall I rollback this patch? (3/3) [ynW...], or ? for more options: y
 OK, will also roll back its 2 dependants

OTOH, the original proposed prompt was shorter, and perhaps it's a
detail you expect people to learn as they use the software a little bit,
and then don't need the more verbose prompt anymore.

    Mark

On 10 Aug 2010, at 18:00, Mark Stosberg wrote:

> OTOH, the original proposed prompt was shorter, and perhaps it's a
> detail you expect people to learn as they use the software a little bit,
> and then don't need the more verbose prompt anymore.

I'm all for interfaces that make it obvious to beginners what features are available (or easy to find out what they are), without drowning experienced users in noise.

-- 
http://www.patchspace.co.uk/
http://www.linkedin.com/in/ashleymoran