cat <<EOF > f
aa
bb
EOF
darcs rec -alm init
darcs repl -f aa bb f
darcs revert

The dialog of revert will ask the wrong questions.
If the hunk is reverted the replace will be inplicitly reverted too,
but darcs will proclaim that is just skipped a patch.

The expected dialog should ask for the replace first, and skip
the hunk if the replace was not reverted.

I don't know if this is relevant, but 'darcs replace' on an Added but
not yet Recorded file doesn't create a Replace patch.

After some thinking I see I may be wrong on a few things.

To create a Replace patch for a not yet recorded file would really be
nonsense, since the file is empty in pristine, and any replace would be
mute.

The revert dialog is harder. The message "Skipped revert of 1 patch." is
confusing, but I looked in SelectChanges.lhs, and it didn't look trivial to
change.

The thing is that in most other dialogs the questions are asked in the
other dependency order, so that a No can force later patches to become No,
but a Yes can only force previous (probably Wait) patches to become Yes.
Thus the skip wording works in the common case, although it is possible to
answer Y and N to a series of patches, walk back with K, and watch darcs
"skip X patches" with various Y and N answers.

So the question becomes: is this the best way to ask questions about
reverts? I have a faint memory of this being discussed before.

I'll change this issue into a wish for the dialog to be less confusing, or
for the Revert dialog to be reversed (although I'm not sure that will be an
improvement).

I think this is a UI bug if we're saying 'skipped revert of N patches due to
dependencies' when we really mean 'forcing revert of N patches due to dependencies'.

My guess is that we're wrong in the same way for all the undo operations, eg
obliterate, unrecord and rollback.

I suggest we actually sit down with some examples and work out for each
interactive command if the skipped job message is the correct one, and if not
what we should say instead.

This may give us some clarity about how to fix it.

(actually, Tommy's title was easier to understand)

For reference: note that we once tried to do this, but ran into issue568.

Here is one proposed mockup (not with actual dependencies, but just to
give an idea).  It's been tweaked by Florent a bit:

Thu Jun 10 09:41:51 BST 2010  Florent Becker <florent.becker@ens-lyon.org>
  * haddocks for SelectChanges
Shall I unpull this patch? (1/?) [yN...], or ? for more options: n

Skipping the following patch due to dependencies
Mon May 31 07:56:30 BST 2010  Florent Becker <florent.becker@ens-lyon.org>
  * better message when skipping already decided patches
I cannot unpull this patch; something else depends on it.

Tue Jun  8 07:28:02 BST 2010  Petr Rockai <me@mornfall.net>
  * Fix typo in the BSD version of date arithmetic (testsuite).
Shall I unpull this patch? (3/?) [yN...], or ? for more options: n

Skipping the following 2 patches due to dependencies:
Tue Jun  8 07:16:31 BST 2010  Petr Rockai <me@mornfall.net>
  * Let's try to work with BSD date as well.
I must unpull this patch; it depends on something we unpull
Mon Jun  7 23:32:57 BST 2010  Petr Rockai <me@mornfall.net>
  * Fix a race condition in the match-date test.
I cannot unpull this patch; something depends on it

Mon Jun  7 19:51:16 BST 2010  Petr Rockai <me@mornfall.net>
  * Skip issue1763 test (non-ascii filenames) on win32.
Shall I unpull this patch? (6/?) [yN...] or ? for more options