Hi all,

This message tries to cover a proposal for a tactical release "process"
for the OpenOCD community to consider.  If there seems to be agreement
that this will be acceptable, I will write start some the relevant bits
up in a document and add it to the reference manual.


At the moment, we are continuing to apply outstanding work and bug fixes
as we move to close the merge window, which I would like to have occur
by the end of this week.  With that in mind, I would like to plan to
branch 0.2.0 as soon as possible.  That means we need to wrap up
everything that is either done, or choose to postpone it; the last
message about features

Essentially, I want us to cut a release with what we have and plan to
repeat the process again repeatedly in the following weeks.  Right now,
the project has taken too much time since the last release to ensure
that the set of supported targets will be a strict superset of those
from 0.1.0.  I think we must expect some things will be broken, and we
will not find out what until we produce a new release for users to test.

To prevent future regressions, I have started prototyping test result
reporting tools that will allow users to report success or failures with
their specific configurations.  For simplicity, I would like this to be
used only with released versions; this consolidates the testing points.
Reports of problems with the SVN trunk always go to the mailing list.
We should be releasing "often enough" that the successful test coverage
improves with each releases, so we need to start building up clear and
consistent records of our progress over time.

This methodology supports the idea of "alpha" or "beta" release phases,
where the same release processes are used to create those deliverables.
This will allow the release, testing, and reporting processes to be
verified and debugged fully.  The resulting data should allow us to
ensure we are delivering regular fixes and improvements to the user
community, without any known exceptions.

The tactical goal of each release should to be focus on delivering a
product with steadily increasing stability and device support (whether
host, target, or interface).  Once we have machine-generated records of
certain combinations of hardware having worked with a certain version of
OpenOCD, it will be virtually impossible for us to miss regressions with
particular hardware combinations.


Does this proposal sound acceptable?

Cheers,

Zach Welch
Corvallis, OR



_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to