view tests/test-push-checkheads-unpushed-D3.t @ 3846:f9dad99a90d5

evolve: create a new commit instead of amending one of the divergents

This patch changes the behavior of evolve command while resolving
content-divergence to create a new commit instead of amending one of the
divergent ones.

In past, I have made this change, backed out this change and now today again I
am doing this change, so let's dive in some history.

Using cmdrewrite.amend() was never a good option as that requires hack to delete
the evolvestate and also gives us less control over things. We can't make the
commit on top of different parents as that of content-divergent ones. Due to all
these, I first made this change to create a new commit instead of amending one.

But, after few days, there was flakiness observed in the tests and turned out
that we need to do some dirstate dance as repo.dirstate.setparents() does not
always fix the dirstate. That flakiness was a blocker for progress at that time
and we decided to switch to amend back so that we can have things working with
some hacks and we can later fix the implementation part.

Now, yesterday while tackling resolving content-divergence of a stack which is
as follows:

C1 C2
 | |
B1 B2
 | |
A1 A2
 \/
base

where, A1-A2, B1-B2, C1-C2 are content-divergent with each other. Now we can
resolve A1-A2 very well because they have the same parent and let's say that
resolution leads to A3.
Now, we want to resolve B1-B2 and make the new resolution commit on top of A3 so
that we can end up something like:

C3
 |
B3
 |
A3
 |
base

however, amending one of the divergent changesets, it's not possible to create a
commit on a different parent like A3 here without some relocation. We should
prevent relocation as that may leads to some conflicts and should change the
parent before committing.

So, looking ahead, we can't move with using amend as still using that we will
need some relocation hacks making code ugly and prone to bad behaviors, bugs.
Let's change back to creating a new commit so that we can move forward in a good
way.

About repo.dirstate.setparents() not setting the dirstate, I have researched
yesterday night about how we can do that and found out that we can use
cmdrewrite._uncommitdirstate() here. Expect upcoming patches to improve the
documentation of that function.

There are lot of test changes because of change in hash but there is no behavior
change. The only behavior change is in test-evolve-abort-contentdiv.t which is
nice because creating a new commit helps us in stripping that while aborting.
We have a lot of testing of content-divergence and no behavior change gives
enough confidence for making this change.

I reviewed the patch carefully to make sure there is no behavior change and I
suggest reviewer to do the same.

====================================
Testing head checking code: Case D-3
====================================

Mercurial checks for the introduction of new heads on push. Evolution comes
into play to detect if existing branches on the server are being replaced by
some of the new one we push.

This case is part of a series of tests checking this behavior.

Category D: remote head is "obs-affected" locally, but result is not part of the push
TestCase 3: multi-changeset branch, split on multiple new others, only one of them is pushed

.. old-state:
..
.. * 2 changesets branch
..
.. new-state:
..
.. * 2 new branches, each superseding one changeset in the old one.
..
.. expected-result:
..
.. * pushing only one of the resulting branch (either of them)
.. * push denied
..
.. graph-summary:
..
.. B'◔⇢ø B
..   | |
.. A | ø⇠◔ A'
..   | |/
..    \|
..     ●

  $ . $TESTDIR/testlib/push-checkheads-util.sh

Test setup
----------

  $ mkdir D3
  $ cd D3
  $ setuprepos
  creating basic server and client repo
  updating to branch default
  2 files updated, 0 files merged, 0 files removed, 0 files unresolved
  $ cd server
  $ mkcommit B0
  $ hg up 0
  0 files updated, 0 files merged, 2 files removed, 0 files unresolved
  $ cd ../client
  $ hg pull
  pulling from $TESTTMP/D3/server (glob)
  searching for changes
  adding changesets
  adding manifests
  adding file changes
  added 1 changesets with 1 changes to 1 files
  new changesets d73caddc5533
  (run 'hg update' to get a working copy)
  $ hg up 0
  0 files updated, 0 files merged, 1 files removed, 0 files unresolved
  $ mkcommit A1
  created new head
  $ hg up '0'
  0 files updated, 0 files merged, 1 files removed, 0 files unresolved
  $ mkcommit B1
  created new head
  $ hg debugobsolete `getid "desc(A0)" ` `getid "desc(A1)"`
  obsoleted 1 changesets
  1 new orphan changesets
  $ hg debugobsolete `getid "desc(B0)" ` `getid "desc(B1)"`
  obsoleted 1 changesets
  $ hg log -G --hidden
  @  25c56d33e4c4 (draft): B1
  |
  | o  f6082bc4ffef (draft): A1
  |/
  | x  d73caddc5533 (draft): B0
  | |
  | x  8aaa48160adc (draft): A0
  |/
  o  1e4be0697311 (public): root
  

Actual testing
--------------

  $ hg push --rev 'desc(A1)'
  pushing to $TESTTMP/D3/server (glob)
  searching for changes
  abort: push creates new remote head f6082bc4ffef!
  (merge or see 'hg help push' for details about pushing new heads)
  [255]
  $ hg push --rev 'desc(B1)'
  pushing to $TESTTMP/D3/server (glob)
  searching for changes
  abort: push creates new remote head 25c56d33e4c4!
  (merge or see 'hg help push' for details about pushing new heads)
  [255]

Extra testing
-------------

In this case, even a bare push is creating more heads

  $ hg push
  pushing to $TESTTMP/D3/server (glob)
  searching for changes
  abort: push creates new remote head 25c56d33e4c4!
  (merge or see 'hg help push' for details about pushing new heads)
  [255]

  $ cd ../..