tests/test-rebase-brute-force.t

revlog: add a C implementation of `headrevsdiff`

Python implementation of `headrevsdiff` can be very slow in the worst
case compared with the `heads` computation it replaces, since the
latter is done in C.

Even the average case of this Python implementation is still
noticeable in the profiles.

This patch makes the computation much much faster by doing it in C.

  $ cat >> $HGRCPATH <<EOF
  > [extensions]
  > drawdag=$TESTDIR/drawdag.py
  > bruterebase=$TESTDIR/bruterebase.py
  > [experimental]
  > evolution.createmarkers=True
  > evolution.allowunstable=True
  > EOF
  $ init() {
  >   N=`expr ${N:-0} + 1`
  >   cd $TESTTMP && hg init repo$N && cd repo$N
  >   hg debugdrawdag
  > }

Source looks like "N"

  $ init <<'EOS'
  > C D
  > |\|
  > A B Z
  > EOS

  $ hg debugbruterebase 'all()-Z' Z
     A: A':Z
     B: B':Z
    AB: A':Z B':Z
     C: ABORT: cannot rebase 3:a35c07e8a2a4 without moving at least one of its parents
    AC: A':Z C':A'B
    BC: B':Z C':B'A
   ABC: A':Z B':Z C':A'B'
     D: D':Z
    AD: A':Z D':Z
    BD: B':Z D':B'
   ABD: A':Z B':Z D':B'
    CD: ABORT: cannot rebase 3:a35c07e8a2a4 without moving at least one of its parents
   ACD: A':Z C':A'B D':Z
   BCD: B':Z C':B'A D':B'
  ABCD: A':Z B':Z C':A'B' D':B'

Moving backwards

  $ init <<'EOS'
  > C
  > |\
  > A B
  > |
  > Z
  > EOS
  $ hg debugbruterebase 'all()-Z' Z
    B: B':Z
    A: 
   BA: B':Z
    C: ABORT: cannot rebase 3:b8d7149b562b without moving at least one of its parents
   BC: B':Z C':B'A
   AC: 
  BAC: B':Z C':B'A