bzr log lies/is inconsistent about changed files

[Frits Jalvingh]

On our 2a format repo:
jal@odeon:~/bzr/vp-trunk$ bzr log viewpoint/src/JSprinter/JSprinterGen.java
------------------------------------------------------------
revno: 236 [merge]
committer: Wouter
branch nick: vp-trunk
timestamp: Fri 2009-11-13 15:03:23 +0100
message:
  Proficon integratie.
------------------------------------------------------------
revno: 4
committer: Frits Jalvingh <email address hidden>
branch nick: vp-trunk
timestamp: Thu 2009-08-27 20:30:46 +0200
message:
  Initial commit of vp-trunk
------------------------------------------------------------
revno: 2
committer: Frits Jalvingh <email address hidden>
branch nick: vp-3.2-hot
timestamp: Thu 2009-08-27 20:20:17 +0200
message:
  Initial commit of vp-3.2-hot
------------------------------------------------------------
revno: 1
committer: Frits Jalvingh <email address hidden>
branch nick: vp-3.1-hot
timestamp: Thu 2009-08-27 20:16:59 +0200
message:
  Initial commit of 3.1-hot
------------------------------------------------------------
Use --include-merges or -n0 to see merged revisions.

Then try:
jal@odeon:~/bzr/vp-trunk$ bzr log -v -r 236 -n0
------------------------------------------------------------
revno: 236 [merge]
committer: Wouter
branch nick: vp-trunk
timestamp: Fri 2009-11-13 15:03:23 +0100
message:
  Proficon integratie.
modified:
  viewpoint/src/iRed/Svb/Maintenance/ComplexAggregatie.java
    ------------------------------------------------------------
    revno: 130.113.20
    committer: Wouter
    branch nick: vp-3.3-hot
    timestamp: Fri 2009-11-13 14:56:13 +0100
    message:
      Refactor & toevoegen commentaar.
    modified:
      viewpoint/src/iRed/Svb/Maintenance/ComplexAggregatie.java

This is inconsistent: log lies about what changed in some contexts. What is worse, it looks like that commit 236 somehow touched almost *all* files in the repo: asking the log for lots of files show those as changed in that commit too.

This looks a bit like the effect we had when we hit bug 415508; but the repository was recreated from scratch since then and nothing from the failed repository was used in the new one. And regardless of the reason *why* this occured- if this commit has a lot of changed texts attached to it it should show them, so the error becomes obvious quicker also.

[Frits Jalvingh (fjalvingh)]

Version used:

jal@odeon:~/bzr/vp-trunk$ bzr version
Bazaar (bzr) 2.1.0
  Python interpreter: /usr/bin/python 2.6.4
  Python standard library: /usr/lib/python2.6
  Platform: Linux-2.6.31-20-generic-x86_64-with-Ubuntu-9.10-karmic
  bzrlib: /usr/lib/python2.6/dist-packages/bzrlib
  Bazaar configuration: /home/jal/.bazaar
  Bazaar log file: /home/jal/.bzr.log

Copyright 2005, 2006, 2007, 2008, 2009 Canonical Ltd.
http://bazaar-vcs.org/

bzr comes with ABSOLUTELY NO WARRANTY. bzr is free software, and
you may use, modify and redistribute it under the terms of the GNU
General Public License version 2 or later.

Bazaar is part of the GNU Project to produce a free operating system.

[Martin Pool (mbp)]

Does diff show that JSprinterGen.java was actually changed between 235 and 236, or not?

[Frits Jalvingh (fjalvingh)]

No, the diff only shows that single file as changed.

[Frits Jalvingh (fjalvingh)]

Hm. Somehow a lot of crud has attached itself to this single commit. I pulled up to revision 235 into a new repo which results in:
jal@odeon:~/test$ bzr branch -r 235 ~/bzr/vp-trunk
Branched 235 revision(s).
jal@odeon:~/test/vp-trunk$ bzr repository-details ..
Commits: 2189
                      Raw % Compressed % Objects
Revisions: 975 KiB 0% 196 KiB 0% 2189
Inventories: 179235 KiB 27% 5285 KiB 3% 15356
Texts: 465721 KiB 72% 141394 KiB 96% 25741
Signatures: 0 KiB 0% 0 KiB 0% 0
Total: 645932 KiB 100% 146876 KiB 100% 43286

Extra Info: count total avg stddev min max
internal node refs 1967 501585 255.0 0.0 255 255
internal p_id refs 542 133548 246.4 44.5 3 255
inv depth 8775 17550 2.0 0.0 2 2
leaf node items 8775 531612 60.6 12.2 28 132
leaf p_id items 1883 93929 49.9 55.8 1 613
p_id depth 1883 4471 2.4 0.8 2 4

So this revision has 25741 text's attached in history. Now pulling that single one-file commit:

jal@odeon:~/test/vp-trunk$ bzr pull -r 236 ~/bzr/vp-trunk
 M viewpoint/src/iRed/Svb/Maintenance/ComplexAggregatie.java
All changes applied successfully.
Now on revision 236.
jal@odeon:~/test/vp-trunk$ bzr repository-details ..
Commits: 2191
                      Raw % Compressed % Objects
Revisions: 976 KiB 0% 197 KiB 0% 2191
Inventories: 187080 KiB 26% 7815 KiB 5% 15808
Texts: 506023 KiB 72% 148053 KiB 94% 29682
Signatures: 0 KiB 0% 0 KiB 0% 0
Total: 694080 KiB 100% 156066 KiB 100% 47681

Extra Info: count total avg stddev min max
internal node refs 1969 502095 255.0 0.0 255 255
internal p_id refs 542 133548 246.4 44.5 3 255
inv depth 9223 18446 2.0 0.0 2 2
leaf node items 9223 561364 60.9 12.1 28 132
leaf p_id items 1883 93929 49.9 55.8 1 613
p_id depth 1883 4471 2.4 0.8 2 4

The #of texts increased to 29682 (+3941) in that single one-file commit.

By itself this looks like the same bug that hit us with content filtering on a --1.14 repo has hit us partially here - which more than bothers me because it cannot be the same bug we had earlier:
- We fully upgraded to bzr 2.x, and have not used older versions since
- I talked with the programmer that did that commit and he was certain he used the correct version
- The repo was fully recreated from scratch as a --2a repo
In addition the number of texts added in this commit is way less as the other bug was causing if I remember correctly - I would expect it to equal the #of text files in the tree if it was the same bug...

Regardless of how this commit came to be (which I am hugely bothered about) log should still show those files as touched by that commit I think..

BTW: does diff also show permission bit (x) changes?