Re: meeting 2006-11-07

[Stan Cox <scox at redhat dot com>]

Agenda:

Ongoing topics:

Testing and kernel problems: cmoller, roland, cagney, len
-test system can now run frysk's funit tests
 -it is also possible to run a custom rpm through
 -we need to monitor the results
-FC5 no longer hangs 
 -it was not detaching the main thread because each individual thread
  was not detached
-does Roland's patch change the behavior of ptrace?
 -difference between FC6 and previous kernels
 -have reproducer of existing frysk teardown frysk3381
 -For FC6 tasks are dying and final ptrace detach fails because
  tasks are gone 
 -either a signal is lost or the detach failed
-Need Dogtail tests for source window
 -pros and cons of testing on dogtail versus using junit
 -Rick to add dogtail tests
-Slow tests
-one of the terminating tests is slow because of a timeout
-fstack tests are also slow

action: dogtail tests (rmoseley,nov-20)
action: frysk3381 (roland, cmoller,nov-20)

Breakpoints: timoore, mjw
-there are tests for low level breakpoint functionality
-timoore adding breakpoints to hpd

action: hpd breakpoints (timoore,nov-20)

Command line utilities: npremji, pmuldoon
-timoore added registers Phil needs
-need tests for command utilities
-task observer not terminating

action: command utility tests (npremji, pmuldoon, mjw,nov-20)

Source Code Viewer: rmoseley, mcvet, scox
-slicing may not work
 -Should Location.java be a bytebuffer instead of an arraybytebuffer?
-multiple source files done
-discussion of the way insight does window layout and stepping

action: add dejagnu framework and tests (scox,nov-13)

Stepping: mcvet, mjw
-Testing instruction stepping with hpd versus source window
 -Some functionality is not in hpd
 -Stepping is done via callbacks which is done at gui startup
 -perhaps need to refactor so hpd can share source window code
 -check insight which uses states to maintain consistency

action: investigate how to test instruction stepping (mcvet,nov-13)

Discussion:
-insight had performance problem refreshing variable window
 -after step/continue everything was refreshed
 -time spent redrawing instead of getting information from gdb
 -variable objects were a diff between previous and current state
 -current frysk has lower layer to draw from scratch
 -upper layer can do the optimization
-all developers need to do test driven development