When recording, the new code that announces that patches are skipped due to
dependencies sometimes gives false positives.  To reproduce this, make
several independent changes to a file, run an interactive record and hit
'f' on the first change.  This should select all the changes, and does, but
you get a confusing message claiming some changes have been skipped due to
dependencies.

This is actually observed on the darcs-conflicts branch, but I expect it's
also present in darcs-unstable.
-- 
David Roundy
Department of Physics
Oregon State University

On Thu, Dec 06, 2007 at 00:11:50 -0000, David Roundy wrote:
> When recording, the new code that announces that patches are skipped due to
> dependencies sometimes gives false positives.  To reproduce this, make
> several independent changes to a file, run an interactive record and hit
> 'f' on the first change.  This should select all the changes, and does, but
> you get a confusing message claiming some changes have been skipped due to
> dependencies.

Hmm, that is indeed confusing.  Is it just the messages that are at
fault or is there a deeper rework of the code needed.  I am guessing
that what we really need are 4 different messages:

- Skipping the next N patches as requested
- Skipping the next N patches due to dependencies
- Taking the next N patches as requested
- Taking the next N patches due to dependencies

On Thu, Dec 06, 2007 at 11:58:58AM +0000, Eric Y. Kow wrote:
> On Thu, Dec 06, 2007 at 00:11:50 -0000, David Roundy wrote:
> > When recording, the new code that announces that patches are skipped due to
> > dependencies sometimes gives false positives.  To reproduce this, make
> > several independent changes to a file, run an interactive record and hit
> > 'f' on the first change.  This should select all the changes, and does, but
> > you get a confusing message claiming some changes have been skipped due to
> > dependencies.
> 
> Hmm, that is indeed confusing.  Is it just the messages that are at
> fault or is there a deeper rework of the code needed.  I am guessing
> that what we really need are 4 different messages:
>
> - Skipping the next N patches as requested
> - Skipping the next N patches due to dependencies
> - Taking the next N patches as requested
> - Taking the next N patches due to dependencies

Just switching from "Skipping" to "Taking" when appropriate will go a long
way towards fixing this, and I recommend just that change for now, as some
rework of the patch selection may be needed for darcs-2, due to changing
semantics.

But yes, if we change PatchChoices to hold a five possible states per patch
we could do this ("yes", "no", "maybe", "yes, due to dependencies" and "no,
due to dependencies").  We might be able to concatenate this into four
states, since "maybe" always defaults one direction.
-- 
David Roundy
Department of Physics
Oregon State University

This is resolved by Kevin Quick's recent patch.  Thanks Kevin!

Hi Kevin,

Could you have a second look at this?  I don't think this is quite the correct
interaction... (below, icons is a directory with a lot of files).

adddir ./icons
Shall I record this change? (1/?)  [ynWsfvpxdaqjkc], or ? for help: s
hunk ./Makefile 6
+pack : reg2008-pack.tex
+       latex2html -no_subdir -init_file pack2html.pl -split 4 $<
+       cp my-$(basename $(<F)).css $(basename $(<F)).css
+
Shall I record this change? (2/199)  [ynWsfvpxdaqjk], or ? for help: f
Included record of 191 changes due to dependencies.

Can you run the record with the -v flag to verify that the 191 skips  
are not just Makefile changes?

I'll see if I can reproduce this here.  Basically after each response  
it re-filters the possible and not-possible set of patches before  
issuing the next prompt, so the 's'kip on the adddir should have  
removed anything dependent on adddir at that point and not later than  
that.   I assume that sincd the icons directory was new that all the  
files in it were adds as well; this may indicate some indistinct  
patch dependency relationship between an added directory and its  
contents.

-KQ

On 30 Jan 2008, at 4:59 AM, Eric Kow wrote:

>
>
> Hi Kevin,
>
> Could you have a second look at this?  I don't think this is quite  
> the correct
> interaction... (below, icons is a directory with a lot of files).
>
> adddir ./icons
> Shall I record this change? (1/?)  [ynWsfvpxdaqjkc], or ? for help: s
> hunk ./Makefile 6
> +pack : reg2008-pack.tex
> +       latex2html -no_subdir -init_file pack2html.pl -split 4 $<
> +       cp my-$(basename $(<F)).css $(basename $(<F)).css
> +
> Shall I record this change? (2/199)  [ynWsfvpxdaqjk], or ? for help: f
> Included record of 191 changes due to dependencies.
>
> ----------
> status: resolved-in-unstable -> chatting
>
> __________________________________
> Darcs bug tracker <bugs@darcs.net>
> <http://bugs.darcs.net/issue568>
> __________________________________
> _______________________________________________
> darcs-devel mailing list
> darcs-devel@darcs.net
> http://lists.osuosl.org/mailman/listinfo/darcs-devel

> Can you run the record with the -v flag to verify that the 191 skips
> are not just Makefile changes?

Well, I'm positive about that.  They come from a bunch of addfiles and
the Makefile changes contains only the 6 lines (also, I had elided a
small bit where I hit 'c' which is why you see patch counts).

Hope this helps, and thanks!

Resolved by

Thu Mar 20 12:06:56 EDT 2008  David Roundy <droundy@darcs.net>
  * resolve issue568: don't explain why patches are skipped.
  This isn't an optimal fix, but it removes the incorrect claim that patches
  were skipped due to dependencies, and that's enough for now.

although apparently the automatic trigger misfired.