Hi Rick, Xiao-Feng told me that he didn't intend to use TuningFork due to unclear license. So I think that it's better for me to develop a GCSpy client as an Eclipse plug-in while you work in the integration of GCSpy with Harmony. What do you think about it, Xiao-Feng? Any comments are welcome.
Thanks, Patrícia On Sat, Apr 12, 2008 at 7:53 AM, Rick Walker <[EMAIL PROTECTED]> wrote: > Hi Patrícia, > > In my proposal, I elected just to integrate GCSpy with Harmony - I had > a chat with Richard Jones at Kent, who's responsible for GCSpy and has > a similar GSoC project running for JikesRVM, and discussed exactly how > far he thought I could get with it over the GSoC period. The > conclusions we came to from that discussion was that integrating > GCSpy, writing drivers for each GC, and then updating GCSpy itself to > include the latest changes made to the architecture > (http://pubs.doc.ic.ac.uk/GCspy/ has some information on that) was > possible. It's likely that the latter part would benefit from > communication with the JikesRVM project, to make sure that the GCSpy > servers (C++ for Harmony, Java for JikesRVM) are consistent. > > Richard also stressed that it's very important that GCSpy have > negligible performance impact when the visualizer's not connected, and > suggested ways I could benchmark to check that this is the case. > I think that it's realistic to integrate GCSpy and test it properly > over the GSoC period, and I wrote up my proposal based on that, from a > deliverable-oriented perspective. While I'd definitely be interested > in longer-term involvement with Harmony if this summer goes to plan, I > felt that this was a sensible way to approach GSoC. > > I didn't, then, look too deeply at writing Eclipse plugins, though I > did have a brief scan of the Rich Client Platform documentation. The > proposal I looked at from Jikes: > http://jira.codehaus.org/browse/RVM-388 > discusses the possibility of adding GCSpy-style visualization to Tuning > Fork: > http://www.alphaworks.ibm.com/tech/tuningfork > > http://domino.watson.ibm.com/comm/research_projects.nsf/pages/metronome.tenedor.html > (has a link to their research paper on it too). You could extend > tuning fork to support GCSpy-style visualizations as a plugin. The > only problem is that while I think Tuning Fork's going to be released > as OSS, it isn't at the moment. > > Perhaps a split of integrating GCSpy with Harmony vs developing a > better client as an eclipse plug in could work? I'm sure there are > other interesting ways of visualizing GC trace output, so there's > enough depth on that side too. > > Rick > -- * Patrícia Lustosa Ventura Ribeiro * Ciências da Computação - 2005.1 AJaTS - AspectJ Transformation System
