TPTP-PMC-20080910

Logistics

Attending: Oliver, Harm, Eugene, Joanna, Chris

• Chris was late and apologized

4.5.1

Biggest issue is related to [1]

• Joanna has dome some setup script work
• Query to Intel: Can Asaf dedicate more time to this for next few days
  • TPTP goal to try to have a fix in place by Monday
  • We do know that Asaf has been looking at it
• Oliver asks whether sending a note to Asaf's manager would smooth things out for getting Asaf pretty much full time for a few days
  • AI Chris to find out if such a note from Oliver the PMC lead is a good idea

Further 244265 Discussion

It worked in 4.4.1 and started showing up between jan/feb of this year

TPTP did not hit it in our own testing

• On our target systems it was only showing up after 300-400 executions of the test
• On consumer's target systems, it is showing up every 3 or 4 runs

Oliver asks Eugene if there is more than can be done to resolve 24465.

• Eugene notes that this type of defect could be caused by a but in a number of places (latent failure condition). Wider hunt in code may be a good idea.

Oliver asks if this this block 4.5.1 delivery

• 4.5.1 date is set. We want the fix in 4.5.1 release but if it is not done, we cannot delay the 4.5.1 (Fall maintenance release of Eclipse)
• Harm summarizes that we will make date even if fix does not happen

Joanna notes that other than this defect, the only thing left is really testing for 4.5.1

• We intend to do light testing plus more intense profiler tests

Harm would like to see the whole team doing some walkthrus of critical code path looking for more general memory management problems

• Jonathan is looking thru AC codepath
• Eugene has been doing build manipulations for working on the defect
• Harm notes that "the whole" team in this context is literally only ~2 or so people. TPTP resourcing is not what it used to be.
• Query whether the Intel Beijing guys could help walk thru the code too
  • Joanna asked which engineers would have best language skills
  • Chris noted that most of what would be needed should be written details about where to look and IM so both engineers should be able to do it.

We discussed that a malloc/free diagram or allocation sequence diagram would be helpful

The question of klokwork came up, asking if it has been run against the agent code modules

• Team discussed if Eugene should try to run it
• Chris agreed to touch base with Asaf to see if Asaf's team has started to extensively use Klokwork yet.

Oliver noted that his team has some tools that help match malloc/free. He will see if someone can try these out.

Joanna/Eugene noted that the failure happens most frequently on RHEL 5 (plain install with no updates)

• Update: Chris noted that it fails very often on Ubuntu 8.04 as well.

Longer Term thoughts

We have a number of Java developers and project policies designed primarily for Java code. Perhaps we need to have some different policies/practices for C code in the project

Perhaps we need to have more module cross training with the few people left on the project.