Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
109 lines (80 sloc) 4.71 KB
BBChop Readme
BBChop is like 'git bisect' (or equivalent), but works when your bug
is intermittent. That is, it works in the presence of false negatives
(when a version happens to work this time even though it contains the
bug). It assumes that there are no false positives (in principle, the
same approach would work, but adding it may be non-trivial).
It is currently experimental code. At the time of writing it has not
found any real bugs, nor has anyone other than me checked the
maths. You have been warned. Having said that, the code as written
shouldn't be able to do anything dangerous, and works fine finding
'simulated bugs'.
========== How to try out BBChop
Prerequisites: only python, but mpmath is strongly recommended because
otherwise we have to use python's built in decimal module which is very slow.
Then you need to add the BBChop/source to PYTHONPATH and PATH
environment variables.
The simplest way to try it out to is run it in manual mode:
> bbchop -l 10 -c 0.9
This means, search in a linear history of 10 revisions, numbered 0 to
9, until bbchop thinks it has found the faulty location with
probability at least 0.9. It will start asking questions:
|Most likely location is 0 (probability 0.100000).
|Please test at location 3.
|Target detected at 3? Y/N/S(kip)
Eventually the search will terminate and BBChop will print out its
conclusion, with evidence, eg:
|Search complete. Most likely location is 3 (probability 0.919614).
|Number of tests at 3: 3 of which 1 detected
|Number of tests at parents of 3:
| At 2, 5 of which 0 detected
In serious use, it is best to use always use the following options:
-g logfile # the search cannot be restarted without a logfile
-n N # number of detections seen at the known failing revision
-p N # number of non-detecting observations made at the known failing revision
The -n and -p options give bbchop its initial idea of how frequently
the failure occurs. Without this, it (roughly speaking) assumes a
probability of 0.5. It is always advisable to supply these: If its
initial idea is too low, the search will take longer than
necessary. More seriously, if its initial idea is too high, the search
may terminate (at an incorrect location) before it has learnt
this. With a sufficiently high initial probability, bbchop behaves
like got bistect (at least for linear histories - it would be
interesting to compare with nonlinear ones).
BBChop is based on Bayesian Search Theory
(http://en.wikipedia.org/wiki/Bayesian_search_theory ). Hence the name,
'Bayesian Binary Chop'. You do not, however, have to know, want to
know, or even believe in, bayesian probability theory in order to use
it. When it comes to a conclusion, it always prints out some
supporting evidence, which you can evaluate with your native
intuition. In fact, it is better to ignore the probability values it
prints unless you are confident that you understand the assumptions
behind them (which are nontrivial).
Unlike 'git bisect', bbchop is not integrated with a version control
system. It has hooks to allow it to be integrated with a version
control system:
-i or --identifiers <file> : a file containing a list
of location identifiers, one per line.
-a or --ancestry <file> : file containing description of location
DAG, one location per line.
Each line is of the form:
<id of location>[<whitespace><id of parent]*
-s or --switchscript <script>: : script to call to switch
from one location to another. May be used with
or without testscript. Hook for
integration with version control systems.
There is also a script (interfaces/git/git-bbchop-list) to extract a history DAG from git and
print it in the format required by the -a option.
To see all the options, run bbchop with no arguments.
LIMITATIONS
- Each decision is O(N) in the number of revisions to be
searched. (It was quite hard to get this down from O(N^2)). When
this renders it impractical depends on when it starts to be
comparable to the time to actually do a test. For many purposes it
may be practical at 10,000 revisions or so.
- For non-linear histories, it is currently worse than O(N). However,
it is only above O(N) in the number of nodes which have more than
one parent or child.
- For manual operation, bbchop does not support the same interface as
git bisect; the tests need to be run in a separate shell. However,
it does support reloading its log, so this could be added.