Minutes 20070411 meeting

[Elena Zannoni <elena dot zannoni at oracle dot com>]

Here we go.

elena


Mike, Phil, Stan, Elena, Sami, Kris, Tim, Nurdin, Mark



test/build system: what do people run on their systems? fc6 and
rawhide mostly.  People are invited to try the autotest and to send
the results for uploading.  Mark: There are warnings on rawhide,
please send the specifics to Kris and he will add the configuration to
the system, if it needs to be different.

Phil: Could the script be made available as an RPM? Sure.  Could the
whole test system be made available as an RPM? It's trickier because
it keeps getting updated often, and doing autodownload it doesn't
need to rely on people to install the latest version of it. Also it
would require that the log parsing, etc logic be kept working with
older versions of the test scripts. But surely a valid concern, and
something to figure out.

it would be interesting to try the tests on an AMD box to see if they
different from Intel box.

Anybody testing on RHEL5? Is that a platform that people care about?

dogtail testing: take a long time to run and generate a lot of
output. Still needs a full gnome environment.  xvnc server up is the
requirement. Try in headless mode. [missed one comment here, please
fill in]

test coverage: Kris is also working on integrate the generation of
coverage info. Almost done. Using Stan's
instructions/scripts. Problems with concept of source tree vs. build
tree. There is still sone cleanup to do. 

Results reporting: web page being worked on, working on how to access
the log files, and how to get more detailed information (navigation,
basically). 

System will mail you back the summary report of your test run when you
upload it. The report is being cleaned up a bit. 

Kris will post more info and links on the mailing list.


Sami: editing a session and switching sessions and some other changes
in the monitor. Some changes in the druid as well: adding a page with
summary of what the druid does. --> possible for next week gui meeting

Stan: enums are working now. Working on itnernals of how expressions
are evaluated. Each type was handled separately, try to follow the C
rules for evaluation. There was lot of duplicated code, with slight
variation of handling different types. Unified now in "scalar" type,
doing arithmetic in widest type and the reconvert for displaying
results.

Mark: watch out for assignment from a smaller to a larger type.  So if
you want to assign an byte to an int and want to treat the byte as
unsigned use int i = b & 0xff;

Question: how to handle unsigned types? Just do as signed, easier to
use C++ to output an integer and an unsigned value and to read it
in. Class ULong that can do unsigned comparisons.

When to use Ulong? Ulong doesn't allocate any objects, just class of
statics for comparison functions. Use Long. There isn't a place where
it's needed to use BigInteger. Corefiles allocate several Longs. Hard
to pack them correctly.

Stan: hope to finish the clean up by end of week.

Tim: kernel interface: setting bp on functions that have no debuginfo,
using just elf information.  --> possible for next week gui mtg

Tim: We seem to allocate a lot of dwarf info all over the place,
should there be a more centralized allocation?  Similar question from
Phil on Elf. If the object consumes system resources, we need to be
more conservative.  Dwarf stuff is mapping things into memory, so no
FD used.

Has anybody some some memory usage and fd usage profiling, kind of
measures on Frysk?


Mark: out of line instruction mapping. Looking at kernel source code.
put a BP on something that is PC relative, how to execute the next
insn, whether to emulate it or not.


Nurdin: libunwind, no real change. 

Mike: mutiprocess source window stuff.