meeting minutes 20070228

meeting minutes 20070228

[Elena Zannoni]

[-- Attachment #1: Type: text/plain, Size: 1 bytes --]



[-- Attachment #2: frysk-20070228 --]
[-- Type: text/plain, Size: 6221 bytes --]


Nurdin, Andrew, Phil, Elena, Sami, Mike, Chris, Rick, Tim, Mark, Stan.


Andrew: rt directory tightening out the objects there: line, symbol,
object frame easy to get those out of the runtime model.  Uses
lilbunwind, to get size of a symbol. Useful for disasesmbling w/o
debug symbols. Still about a week to go.

Chris: no status. RH internal.

Elena & Kris: setting up autobuild systems. 32-bit and 64-bit machines.
FC6 used. Compromise between fc5 kernel and rawhide kernel.

Mark: frysk building for rawhide (gcj with java 1.5). Stepping and bp
safe for MT applications looking at kprobes and uprobes (uprobes
doesn't work yet, not finished, missing pc-relative addresses work
still. Kprobes does this but it works in the same address space as
where it is running) as examples of out of stream stepping. Not sure
what to do for arches where you need pc relative instructions like
x8664. You need to emulate the whole instruction.  Stop the world
model has different type of complexity.

Mike: poking around rt. testcases for rt. command line utilities,
register window stuff on x86-64. source window stuff. --> note Kris is
aiming at making same stuff working. Will talk to coordinate.

Nurdin: libunwind bindings are done. using bindings in
frysk-rt. Redone bindings, actually. Create bindings for cross
platform unwinding. DMT promises it works.

Phil: core file. native corefile/native platform work, cross
architecture is tricky. Sizes of "long" are of course different.
Looking at elfutils layer. Mssing a few edge cases in the ptrace
states (?).  rawhide machine.

Rick: parser work. Filed a couple of bugs, one is to do parsing on
demand instead of parsing the whole file when opening the source
window. Dramatic speedup. Found bug inside parser and fixed.  Better
testcase needed. There is test to create dom, and run parser.  Good
for regression testing. Dwarf headers are used to collect info-> move
that to use frames and rt info.

What happens if there is no source code--> pops up the assembler.  If
there is sourcecode but cannot be parsed-->just pops up the text file
w/o highligthing.

Eclipse parser has changed radically changed, frysk parser is on its
own now, no need to contribute changes back.

Sami: event viewer rearranging. Eliminate side bars, merge
timelines. Intgration with source window and capturing bt at event
time. If user clicks on event it will show sourcewindow and
stack. Storage issues? this can get pretty big very
quickly. Highlighting of timelines needs to be made working.

Bug with stack frames leaking file descriptors: bug --> BZ number?

strace with backtraces -->Phil can you blog this?

Stan: stabilizing expr handling for optimization handling. Elfutils
bugs where it breaks with inline functions. Intention is to stabilize
everything. Added PID handling to the CLI, can attach to given PID.
But never returns control to the cli, working on this bug.  Extend
absract model for expression. Accessing dwarf directly, intention is
to change that. Antlr bugs, there are a lot. Possibly import a new
version of it.  Problems will be when program manipulates the
terminal. Create a pty and have program under the pty, CLI controls
the pty, when thereis output fromthe program, the cli can decide what
to do. Maybe then starting up a differnt term and run the program in
that would be just easier, logic of switching it very complicated.

Tim: pty stuff. Finish jline to not use system call to do term
manipluation.  Didn't have chance to work on that yet. Some code for
pty term was added to local copy of jline, maybe it's useful
reference.  Breakpoint stuff: insert multiple brakpoints for inline
functions. Still problems with the state machine, there is a bug in BZ
about deleting breakpoints. There are panics when deleting bps.  Mark
may have a fix. It happens when in 'detaching' state.  the bp code
doesn't care whether attached or detached, that is dealt with at a
different level. The core deals with it.
There is a testcase, but it's not passing-->check it in and mark is at XFAIL.

Setting bp by line number works, but using function name doesn't yet.
breakpoint by function name instead of line number: tied to Stan's
parser, will be able to parse the function specification.

Next step would be to click on line in source window and set bp
there.

Break foo-> if foo is overloaded, you would use 'tab' to get list of
possible matches, and then you start qualifying the name so it knows
where to set the breakpoint. You need parser to do this.

xen testing: works fine, test results are consistent.

dogtail tests: kept up to date with the ui changes. Problem with
seeing stuff inside the timeline widget. Anticipated that it will be
too complicated for dogtail.

Do we have test coverage numbers? Not recent ones.
Dogtail is in rawhide now.

Java 1.5-> building problems are being sorted out by Mark. Java-gnome
packages: are they building with 1.5? Somebody was doing the builds a
few months back, and fixes were submitted upstream. Shouldn't be a
problem.

Frysk on rawhide may not be buildable on rawhide while the packages
will be coming out.

Uptake of java 1.5 for other ditros? Ubuntu and Debian: very soon will
carry 1.5 backported to gcj. They are going to be using the public gcc
branch used for fedora. The branch is off gcc 4.1, the patches are
backported, instead of waiting for 4.2 to come out.  Fedora compiler
is going to be 4.1 + 1.5 backports. Similar for Ubuntu and Debian.
OpenSuse: not sure if they are going to include java 1.5?

Meeting format: alternate btw UI and roundtable status.

next week UI: Sami demoing the event viewer. Druid also will be shown.

Move funit from frysk core to frysk gui, so it can pick up the other
test runners as well. There is a bug there though, where junit dumps
core.

Meaning of BZ stated: trackers in suspended->initial majow work is
done, coding is done, committed. Waiting for users to report bugs.

Possible work item for Oracle: add more features to the sourcewindow
to see assembler and source code side by side.  Oracle will
consider. RH thought is that Oracle works on isolated items. Oracle
has to think about this, not sure it matters since we are all on irc
together and interact closely.