New JCarder release?

[Ulrik Svensson]

Hello everyone!

I think we should make a new release of JCarder soon including all the
recently added functionality.

I have not done any Java programming recently, so I have not used
JCarder with the latest patches. Do you consider the latest code ready
to be released? Have anyone seen any stability issues or bugs?

There are two things I think we should take care of before a release:

1. Add an option for turning gated locks filtering on/off. Any
opinions or arguments for what the default should be, on or off? E.g.
should we prioritize avoiding false positives or avoiding false
negatives?

2. Add a section about gated locks to the manual at http://www.jcarder.org/manual.html
and tell something about pros and cons about filtering gated locks
(false positives vs. false negatives).

Are you thinking about anything more that should be done before a new
release?

/Ulrik

[Todd Lipcon]

On Sun, Feb 21, 2010 at 7:09 AM, Ulrik Svensson <ulr...@gmail.com> wrote:
> Hello everyone!
>
> I think we should make a new release of JCarder soon including all the
> recently added functionality.
>
> I have not done any Java programming recently, so I have not used
> JCarder with the latest patches. Do you consider the latest code ready
> to be released? Have anyone seen any stability issues or bugs?
>

I occasionally see an error in the analysis phase, but I think it
might be user error (eg I accidentally start two jcarder programs with
the same output directory, so they overwrite parts of eachother's
data). We could prevent this particular error by not allowing jcarder
to start instrumentation if the output directory already has jcarder
output in it.

Other than that, it works pretty well for my use cases.

> There are two things I think we should take care of before a release:
>
> 1. Add an option for turning gated locks filtering on/off. Any
> opinions or arguments for what the default should be, on or off? E.g.
> should we prioritize avoiding false positives or avoiding false
> negatives?
>

I agree - this should definitely be an option, and apologies that I
didn't make it one from the start. I can try to carve out some time
this week to add that simple option if no one else volunteers.

> 2. Add a section about gated locks to the manual at http://www.jcarder.org/manual.html
> and tell something about pros and cons about filtering gated locks
> (false positives vs. false negatives).

Is this manual built from anything in the source tree, or manually
maintained? If it's in the source somewhere that I'm missing, I'm
happy to write up some docs.

> Are you thinking about anything more that should be done before a new
> release?
>

I think the "-d <dir>" flag in the analysis step may have regressed.
It didn't seem to work on trunk last I tried.

Thanks
-Todd

[Todd Lipcon]

I've pushed a branch to my github:
http://github.com/toddlipcon/jcarder/tree/cloudera

It includes the changes detailed below:

On Sun, Feb 21, 2010 at 11:20 AM, Todd Lipcon <tli...@gmail.com> wrote:

>> 1. Add an option for turning gated locks filtering on/off. Any
>> opinions or arguments for what the default should be, on or off? E.g.
>> should we prioritize avoiding false positives or avoiding false
>> negatives?
>>
>
> I agree - this should definitely be an option, and apologies that I
> didn't make it one from the start. I can try to carve out some time
> this week to add that simple option if no one else volunteers.
>

I defaulted to removing the gated cycles, and added an option
-include-gated-cycles which restores the old behavior.

>> 2. Add a section about gated locks to the manual at http://www.jcarder.org/manual.html
>> and tell something about pros and cons about filtering gated locks
>> (false positives vs. false negatives).
>
> Is this manual built from anything in the source tree, or manually
> maintained? If it's in the source somewhere that I'm missing, I'm
> happy to write up some docs.
>

I sent Ulrik a patch for the manual off-list so he can take care of
the website side of things.

>> Are you thinking about anything more that should be done before a new
>> release?
>>
>
> I think the "-d <dir>" flag in the analysis step may have regressed.
> It didn't seem to work on trunk last I tried.
>

I discovered that the option parsing code was looking for -i instead
of -d. I fixed the code to look for -i and also modified the output of
dot files to go into the same directory.

Thanks
-Todd