[Bug 3841] New: "hg annotate" against merged revision shows wrong revision

[mercurial-bugs at selenic.com]

http://bz.selenic.com/show_bug.cgi?id=3841

          Priority: normal
            Bug ID: 3841
                CC: mercurial-devel at selenic.com
          Assignee: bugzilla at selenic.com
           Summary: "hg annotate" against merged revision shows wrong
                    revision
          Severity: bug
    Classification: Unclassified
                OS: All
          Reporter: foozy at lares.dti.ne.jp
          Hardware: All
            Status: UNCONFIRMED
           Version: 2.5.1
         Component: Mercurial
           Product: Mercurial

Below graph is a part of filelog tree of "mercurial/commands.py" in
the repository of Mercurial itself.

  A .. B ....... D -- E ... F -- G
         \          /          /
          \        /          /
           ..... C -----------

  A: df78d8ccac4c
  B: 574869103985
  C: 938dd667ca21
  D: 1c0c413cccdd
  E: c6b912f8b5b2
  F: 1e84f1014f33
  G: 8db4d406b3d3

"def clone(....)" line in "mercurial/commands.py" is modified at rev
"A", and no other revision between "A" to "G" modifies that line: "G"
doesn't, too.

Then, "hg annotate" should show rev "A" for "def clone(....)" line as
source revision against every revisions between "A" to "G", but it
shows not "A" but "C" against "G".

I found the cause of this issue in the "hist" cache management in
"filectx.annotate()".

For consistent annotation, "hist[C]" should be kept until building
"hist[G]" up, in this case. But in filelog tree above, "G" merges "F"
and its ancestor "C", and this causes:

  - "hist[C]" is purged just after building "hist[E]" up unexpectedly,

  - "hist[C]" is re-built up just before building "hist[G]" up, and

  - "hist[C]" is built up incorrectly, because "pcache[C]" is empty at
    that time: this causes mis-recognition as "all lines are new at C"

I already confirmed that changing "hist" or "pcache" purge policy can
fix this issue. For example:

      ====================
      --- a/mercurial/context.py    Sat Feb 09 17:54:01 2013 +0000
      +++ b/mercurial/context.py    Tue Feb 26 01:58:13 2013 +0900
      @@ -719,7 +719,7 @@
                               needed[p] -= 1

                       hist[f] = curr
      -                pcache[f] = []
      +                del pcache[f]

               return zip(hist[base][0], hist[base][1].splitlines(True))

      ====================

But, such changes seem to decrease performance (by scanning same
revisions again and again) and/or resource efficiency (by keeping
useless entries).

Are there any other ideas ?

-- 
You are receiving this mail because:
You are on the CC list for the bug.