Skip to content

Releases

Jump to bottom Edit New page
Michael Lyle edited this page Feb 13, 2018 · 27 revisions

dRonin releases

Release-20180212 ("Wired")

Vital statistics

wired faulty release (pulled):

wired main release:

Postmortem facts and opinions

  • Screwed up androidGCS despite checklist improvement.
  • Mac OS X artifacts were invalid (?)
  • Showstopper bug with arming resulted from last-minute corner arming change.
  • I think the bug with the arming was highly avoidable--- it was ill-advised to make and pressure-through a risky change at the last minute, and if it was going to happen there should have been more validation of adjacent functionality. The Mac OS X one happening on the release artifacts was less-so, as the intermittent issue wasn't known. Still, artifacts should always get verified.
  • Not having shipped in a long time (rusty), and being desperate to ship because of other commitments, results in badness.
  • USB bug still lingers.
  • The cycle dragged in the middle because core developers had lots of other things to do. Eventually kicked this one out the door so high-risk, high-value work could proceed.

Release-20170717 ("Neat")

Vital statistics

Postmortem facts and opinions

  • As always, putting off complicated work results in feature creep, churn, and release date uncertainty.
  • USB bug still lingers.
  • RE1 continues to be difficult to support.
  • AndroidGCS is annoying to release, tried to improve with checklist change for next time.

Release-20170213 ("Artifice")

Vital statistics

artifice main release:

artifice 20170213.1 hotfix:

Postmortem facts and opinions

  • Chose to run a risk of regression by merging new Qt late; it's likely GCS is more crashy as a result.
  • Work continued to trickle in very late in process until merge button was forced to be pushed by hardware timelines.
  • It's a real pity the USB bug was not fixed before release, even if it is an old issue.
  • RE1 significantly regressed with a LED bug that was not detected in the 2 months before release.

Release-20161004 ("Quixote")

Vital statistics

quixote main release:

Postmortem facts and opinions

  • Seemed like we tried to run this cycle too aggressively. Chose to slip a couple weeks in the middle and it was still super fast.
  • Likewise I think we all got burnt out towards the end.
  • However, the actual PR queue and fixing activity at the end was really rather tame.
  • Took time to redocument aspects of release process.
  • Release process was not followed for 20161004.1 and as a result build was not locked or updated here.

Release-20160720 ("samsara")

Vital statistics

samsara hotfix 1:

samsara main release:

Postmortem facts and opinions

  • Made an unnecessary hotfix, not awesome.
  • As always, big churn at the end and panic.
  • Shipped with regressions to CC3D + PPM.

Release-20160409 ("tanto")

Vital statistics

tanto hotfix 2:

tanto hotfix 1:

tanto main release:

Postmortem facts and opinions

  • A lot of risk was incurred late in the cycle by decisions to add NazeR6 support and the upgrader features. There was a lot of churn and the normal test cycle was disrupted.
  • The Lux target was problematic. There are insufficient users on it and regressions affected it a number of times and were undetected.
  • There was confusion with makefiles and thrash.
  • Two hotfixes were needed. The first one was minor, and added support for inverted SBus and corrected a long-standing problem with bootloader detection on Windows. Unfortunately, the first hotfix regressed PPM on Naze32, and required a second hotfix.

Release-20160120 ("renatus")

Vital statistics

Renatus hotfix 1 (numbered 3 because there were two "bad" set of artifacts first):

Renatus main release:

Postmortem facts and opinions

  • The release operated without a formal schedule because we were unaware of scoping of branding changes and new targets. We have a better baseline to work from for future releases.
  • The event system changes skewed with at-the-time out-of-tree targets-- Lux, BrainFPV, Naze32Pro, etc. Because there were per-target changes, multiple downstreams missed the associated changes. Some thought should be given in the future about how to "forcibly break" downstream trees in a way that is obvious.
  • There were a few PRs that took multiple passes to get right and could have done with more thorough dev-testing and third-party-feedback. Input wizard and switch arming were the main ones, and I (@mlyle) was guilty.
  • A change was merged during the holiday period to alter default output rates that could have been harmful to physical servos (400Hz) and there was no community consensus behind.
  • Various issues lingered "in people's heads" to the end of the cycle. When we merge "incomplete" functionality there needs to be a blocking issue opened. This included the lack of user notification of usage tracking and the final items for crash reporting.
  • A hotfix was necessary because minor issues were missed during the cycle.
    • Throttle reversal did not work in input wizard and was tricky to set up otherwise because of a range validation bug.
    • BrainFPV alarms had been broken for a month and no one noticed :/
    • The .deb package was missing firmware
    • Lumenier LUX UART configuration was incorrect.

ARM

Add a custom sidebar
Clone this wiki locally