Without a doubt, the most helpful and well thought out paper review I have ever received. This is the review in its entirety, except for the two-sentence summary of the paper which is excluded.
"The proposed technique appears to be novel. The authors have implemented the technique and evaluated on several significant software systems. The results are very promising."
Wednesday, April 30, 2008
Monday, April 14, 2008
QVM Talk
Just gave a QVM talk in Dagstuhl seminar on Scalable Program Analysis.
The slides are available here: http://kathrin.dagstuhl.de/08161/Materials2/
The slides are available here: http://kathrin.dagstuhl.de/08161/Materials2/
Tuesday, April 1, 2008
QVM: From 16.5K to 10K in 1 hour
This is a story that should serve as a warning to all people submitting to conferences. The tragic part is mostly the fact that we have learned that lesson
so many times, and still manage to ignore it... here goes...
When we wrote our QVM submission for OOPSLA, we assumed that the rules of the game are: ACM SIGPLAN template, 20 page limit. These are pretty reasonable rules. To be honest, we were quite pleased with our ability to describe things nicely and in sufficient detail using this format.
A few hours before the deadline (no more than 2), we decided to verify that our
nice submission indeed conforms to the OOPSLA submission directions. Much to our surprise, a strict 10,000 words limit accompanied the 20 page limit. A quick word-count of our submission yielded the fantastic number of 16,500 words... a mere 65% over the limit.
What followed can be only described as a major slash-and-burn frenzy in which whole sections of the paper went out. After a lot of work we reached the more reasonable limit of 12,000 words. Needless to say, the aforementioned "nice and in detail" was no longer a valid description of the paper. However, we believe that we maintained some of the essence of the paper in a reasonable form. (A final word count, without tables and figures, produced a number that was sufficiently close to 10,000 words.)
Some of the things that were described in the slashed 4000 words are:
- formal descriptions of certain components, notably the typestate histories
- some implementation details of typestate history and debug information
- implementation details of parallel checking of heap assertions
- additional experimental evaluation of the heap assertions
- also lost were various discussions of future extensions and optimizations
- references to our online supplement material (see below).
Here are some of the experiment logs recording (some of) our experience with QVM over various benchmarks:
Azureus (here are some bug reports bug1,bug2,bug3,bug4,bug5)
EclipseTrader
GOIM
Feed'NRead
so many times, and still manage to ignore it... here goes...
When we wrote our QVM submission for OOPSLA, we assumed that the rules of the game are: ACM SIGPLAN template, 20 page limit. These are pretty reasonable rules. To be honest, we were quite pleased with our ability to describe things nicely and in sufficient detail using this format.
A few hours before the deadline (no more than 2), we decided to verify that our
nice submission indeed conforms to the OOPSLA submission directions. Much to our surprise, a strict 10,000 words limit accompanied the 20 page limit. A quick word-count of our submission yielded the fantastic number of 16,500 words... a mere 65% over the limit.
What followed can be only described as a major slash-and-burn frenzy in which whole sections of the paper went out. After a lot of work we reached the more reasonable limit of 12,000 words. Needless to say, the aforementioned "nice and in detail" was no longer a valid description of the paper. However, we believe that we maintained some of the essence of the paper in a reasonable form. (A final word count, without tables and figures, produced a number that was sufficiently close to 10,000 words.)
Some of the things that were described in the slashed 4000 words are:
- formal descriptions of certain components, notably the typestate histories
- some implementation details of typestate history and debug information
- implementation details of parallel checking of heap assertions
- additional experimental evaluation of the heap assertions
- also lost were various discussions of future extensions and optimizations
- references to our online supplement material (see below).
Here are some of the experiment logs recording (some of) our experience with QVM over various benchmarks:
Azureus (here are some bug reports bug1,bug2,bug3,bug4,bug5)
EclipseTrader
GOIM
Feed'NRead
Subscribe to:
Posts (Atom)