search

google / thread-weaver

A Java framework for testing multithreaded code.
Apache License 2.0
305 stars 68 forks source link

Limiting number of interleavings #7

Open lamvak opened 7 years ago

lamvak commented 7 years ago

Hi, The code in current shape implies sequential consistency (due to locking after each byte code instruction). This way the analysis is fully reliable only when there are no data races - as defined in JMM https://docs.oracle.com/javase/specs/jls/se8/html/jls-17.html#jls-17.4.5

There is work outlining a different approach to interleavings using only locations where the happens-before relation is forced (i.e. lockin / unlocking the same monitor, acess to a volatile variable, etc.). See https://www.cs.rice.edu/~javaplt/papers/ricken-phd-thesis.pdf

This optimization in number of interleavings, enclosing whole critical blocks, could be considered in Thread Weaver - it seems that with already assumed sequential consistency there would be no loss of reliability. Please, let me know if the above seem a fair statement. I would happily try to tackle this. Marcin

alasdairmackintosh commented 7 years ago

Thanks for the message. I'll take a look at the links you sent, and will try to get back to you in a few days.

Alasdair

On Sun, Oct 9, 2016 at 4:31 PM, lamvak notifications@github.com wrote:

Hi, The code in current shape implies sequential consistency (due to locking after each byte code instruction). This way the analysis is fully reliable only when there are no data races - as defined in JMM https://docs.oracle.com/javase/specs/jls/se8/html/jls-17.html#jls-17.4.5

There is work outlining a different approach to interleavings using only locations where the happens-before relation is forced (i.e. lockin / unlocking the same monitor, acess to a volatile variable, etc.). See https://www.cs.rice.edu/~javaplt/papers/ricken-phd-thesis.pdf

This optimization in number of interleavings, enclosing whole critical blocks, could be considered in Thread Weaver - it seems that with already assumed sequential consistency there would be no loss of reliability. Please, let me know if the above seem a fair statement. I would happily try to tackle this. Marcin

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/google/thread-weaver/issues/7, or mute the thread https://github.com/notifications/unsubscribe-auth/AIn60Hwx4rL9Oqmfnw-a8HvG4Jibhg_eks5qyXlAgaJpZM4KSJdW .