[cdt-dev] Feb 5th CDT conference call minutes

[Sebastien Marineau <sebastien@xxxxxxx>]

Hi all,

here are the minutes of the CDT conf. call from last Wednesday.
I also had an opportunity to talk with John Duimovich on Friday,
and so have added follow-up comments on CPPUnit, testing and 
automatic cvs patch generation.

Present:

Rational:
Michel Paquet, Doug Schaefer, Sean Evoy, John Camelon, 
Andrew Niefer, Hoe Amer, Martin Lescuyer, Sky Matthews

QNX:
Sebastien Marineau, Alain Magloire, Judy Green, Mikhail
Khodjaiants, Dave Inglis, Thomas Fletcher, Peter Graves

Timesys:
Sam Robb, Omar Naseer

Exor Intl:
Mike Charles

Tensilica:
Chris Songer

Redhat:
John Healy, Elena Zannoni, Phil Muldoon, John Banes,
Rick Moseley, Michael


The following was discussed:

1. CDT 1.0.1 status:
The CDT 1.0.1 bugfix release went GA mid-January. All of the features
outlines in the december conf. call were rolled in, with the exception
of disassembly breakpoints.

Alain mentionned that he has merged some critical fixes from the CDT
head branch into the 1.0.1 CVS branch (post 1.0.1 GA release). 
This might be useful for OEM's wanting to ship on CDT 1.0.1.


2. Update from participants:

- Mike Charles from Exor mentionned that his company's interest 
was in potentially using Eclipse and CDT as the basis for an
IDE for their embedded panels.

- Michel Paquet presented the Rational CDT team to the group. 
Rational has geared up quite a significant team to work on CDT.
The slide set has been mailed to the CDT mailing list, and will
be posted on the web site.

- Redhat talked about their Eclipse activities; this includes
building an RPM plugin/spec file editor, a profiler project,
as well as ongoing work to make a compiled version of Eclipse.

- Timesys just release a beta of their Timestorm IDE for Linux.
This is based on Eclipse 2.0.1, along with CDT 1.0.

- QNX is now releasing their latest version of Momentics,
based on Eclipse 2.0.1 and CDT 1.0.1.


3. Technical discussions -- CDT core components
There were two main thrusts to thsi discussion. First, Michel
presented the details of Rational's plans for CDT (see his
slide set). This includes deliverables for the parser, cmodel,
CDOM, indexer, refactoring etc. He also offered help on the
build model, class browser etc. Rational's deliverables are split
between two major milestones, Aug. 2003 and Dec. 2003. This is in line
with internal project milestones. Everyone was very excited about 
Rational's plans and contribution -- this will really help push
the CDT to the next level.

As part of the discussion, Thomas brought up the point that QNX
was likely going to contribute to the class browser and class utilities.

The discussion then veered off to CPPUnit. CPPUnit is a test automation
framework for C/C++ (a port of JUnit to C++), and Rational is planning
an Eclipse plugin for this. A proposal has been made to put this
as part of the CDT project. The discussion centered on whether this
should be part of CDT, a separate project under the Eclipse tools
umbrella, or a separate project hosted somewhere else, but pulled in to
CDT as a convenience (note that JUnit does the latter).
I discussed this with John D., but did not come to a conclusion.
I will discuss this with Martin separately.

Sam Robb then discussed his build model proposal. The proposal is posted
on a web site, and the URL has been sent to the list. His proposal includes
Javadoc API's. Sam's build model has a reference implementation for a GNU
toolchain,
and his prototype has support for the initial "high-runner" toolchain
options.

The one remaining hurdle for Sam to contribute his code is to have his
management approve the donation. Sam was going to discuss this with his VP
and
get back to the group this week. 

Everyone felt that the build model was very important, and the participants
agreed
to review Sam's proposal. Rational also emphasized that they could provide
additional help with the review and implementation if needed.

Afterwards, other topics relating to CDT core were discssed: this included
UI
enhancements and usability-related updates. The proposal was again made that
these
be put up in a "projects" section on the website for people to take on.


4. CDT debug:

This seems to be progressing well. Alain gave a quick overview of the CDT
debug
plan, as posted to the mailing list. A lot of items on the list have already
been
implemented, with the balance being scheduled for completion by mid-march.

There was some discussion of unimplemented MI commands, which Alain will
post as
bugs to the gdb project. There was also a discussion on supported gdb
version;
which still look to be 5.2.1 and 5.3. Elena mentionned that 5.4 was not
branched
yet, and would likely not be out before our next minor release of CDT.

What was not discussed was the debugger roadmap beyond the 2-3 month window.
One
possibility, which was discussed after the call ended, was to tackle C/C++
and Java
debugging integration (JNI/Java integrated debugging).


5. Documentation and testing:

Documentation was discssed. The lack of a CDT user's manual is still
a significant gap, and the contributing companies are all spending time
writing their own user documentation. A lot of us have raw content to 
provide, but the project still lacks the resources to move forward on this.
QNX and Rational have had discussions on this, and Tensilica has also 
offered to help out.

The issue of documentation format was also brough up as a barier, as most 
companies provide documentation in their own format, and having this 
intermixed with Eclipse-style docs is not very seamless.

The testing/verification aspect was heavily discussed, along two main
thrusts: JUnit unit-testing for the API's and functionality, and automated
UI testing.

On the JUnit topic, Rational mentionned that they were going to implement
not
only the features (parser/CDOM/etc), but also the verification suite (JUnit
tests) for those features. This will improve our test coverage. QNX will
also keep providing resources in this area (Peter Graves). Peter recently
finished automating the running of the CDT testsuite from the CDT build, and
will now be focussing on testcases. However, the challenge of retro-fitting
testing to the large existing codebase will remain.

The discussion then moved to automated UI testing. There was concensus that
this was sorely needed, especially given the fact that, as opposed to the
eclipse base platform, the CDT is not really used self-hosted, so it
receives
less implicit testing.

Chris (and others) felt strongly that we needed to start building a
testsuite.
The issue with using commercial tools for CDT UI testing was discussed
again,
namely reconciling this with the open-source nature of CDT. The issue of 
host coverage (Windows, Solaris, Linux, QNX etc) was also brought up; 
unfortunately, Windows has the widest choice of UI testing tools, but also
has the best Eclipse SWT support; having coverage on the other hosts would
be 
highly desirable.

Redhat proposed that idea that some of the contributing companies
provide remote server access to run UI testsuites using commercial tools,
although it was not clear if the licenses for those tools would allow
this scenario.

I discussed this aspect with John D. last Friday, and his feeling was
this it would probably be acceptable to use commercial tools for CDT
testing. We would check into CDT CVS the test scripts for the UI 
test tools, and a few of the participants would sign up to running the
tests. The major conflict-of-interest potential John saw would 
be to use a commercial tools sold by one of the CDT contributors
(for example, Rational) unless those were made available openly to the
CDT community.

The tools that QNX uses internally is called Winrunner (from a 
company called Mercury Interactive). They also sell a product 
called XRunner, which automates UI testing on
Linux and Solaris. We can discuss this further on the mailing list.


6. Release strategy:

In light of Rational's August and December milestones, the CDT release
strategy and timing was revisited. The agreement was to stick to the course
outlined in the december call, and have frequent releases of smaller scope,
to address the features and stability needed by the community. Thus, we are
still
aiming for an April CDT 1.1 release, which would include the debugger 
enhancements, as well as a prototype build model.
As the build model is one of the central points of integration in the CDT,
it was felt that having a prototype out as soon as possible (to gather as
much
feedback as we can) would be preferable to having a more polished build
model later in the year.

Everyone also seemed in agreement with an Aug/Sept release that would
include
the Rational contributions with parsing/CModel/etc. This would still keep
us on a ~4 month release schedule.


7. Varia:

- The CDT conference calls will be scheduled the first Wednesday of 
every month; thus, the next CDT call will be March 5th from 11 to 1.

- I discussed the issue of having the Eclipse CVS server automatically
generate
patches from checkins, using the scripts Tom Tromey had send, with John D.
John will
revisit this issue with the Eclipse infrastructure group. As the Eclipse
source/web
site is hosted on IBM corporate hardware, the CVS server is in a more
controlled
environment than most other projects.


Any comments, additions, corrections etc. welcome!

As always, many thanks to everyone for your ongoing support.

Sebastien