Skip to content

Cold run of a 9.24 app: harvest on every run, feedback report, widgetdrop, smokewalk and journeys fixes - #14

Merged
MendixMau merged 11 commits into
mainfrom
fix/cmm-19-21
Nov 17, 2026
Merged

MendixMau merged 11 commits into
mainfrom
fix/cmm-19-21

Conversation

@MendixMau

Copy link
Copy Markdown
Owner

Fixes from the cold run of a Mendix 9.24 app (9.24.2 → 10.24.28 → 11.12.7) on the toolkit. Every fix has a control that fails on the old code and passes on the new.

What changes

  • Harvest is now on for every run. bin/feedback_report.py runs at every pipeline exit, whether the run passes, stops or crashes. It harvests the learnings and writes FEEDBACK-REPORT.md: where the run stopped, the draft learnings, and an issue summary ready to paste (metadata only, checked against the denylist). Harvest skips files the pipeline writes itself.
  • CONTRIBUTING: contributions are open. A bug goes in as an issue built from FEEDBACK-REPORT.md.
  • Journeys: the row verb finds Data grid 2.2 rows, whose cells carry no role.
  • Smokewalk: a widget's own error box on the page fails the step. Before this, "Unable to get filter store" passed.
  • stepupdate: run by hand without a git identity, it commits as upgrade-pipeline.
  • Pipeline:
    • exports UPG_NAME, so runapp stops the pipeline's own containers
    • rung 2 gets the content ids from rung 1's plan for localedits
  • widgetdrop (bin/widgetdrop.py, repair row 33): the update step stops before update-widgets drops settings that a newer widget definition brought, for example a 3.x drop-down filter in association mode under a 2.x package.
  • Playbooks:
    • Mendix 11 removed IContext.setExecutionId
    • Charts AnyChart maps to Custom chart
    • Datagrid 2.30 needs DataWidgets in the same rung
    • backlog 151-154

Checks

  • Full control suite on a frozen copy of da92082: 2716 passed, 0 failed, 34 skipped.
  • Full suite on 1e87d94 (the head of this branch): 2731 passed, 0 failed, 34 skipped.
  • Deny check on every commit's content diff (git show --format= | grep -ciE "$DENY"): 0.

🤖 Generated with Claude Code

MendixMau and others added 11 commits November 17, 2026 18:24
… role

Cold CMM (9.24.2): Data grid 2.2.x renders rows as div.tr role="row" (display:
contents) with div.td cells and no role, and the grid root has no
widget-datagrid class. No ROWS selector matched: every row step threw an
instrument gap, 110 baseline steps FAILed and journey coverage was INVALID.

ROWS gains '.tr[role="row"] > .td'; nthRow takes a row as visible when its
first cell is (a display:contents row has no box) and clicks that cell.

Controls: journeys-verbs-browser-row-dg22-pass (fixture with the gridcell
roles and widget-datagrid class stripped). Old code: the 'dg2.2 cells without
a role' case FAILs (and with it the 7 browser-row controls that reject any
FAIL line); new code: 7 passed, 0 failed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…eline

Cold CMM: repair row 32 says to run stepupdate.sh with a one-line plan. Run by
hand under a HOME with no git identity, the DataWidgets step updated, passed
idcheck, and then STOPped at "git commit failed" (Author identity unknown).
The pipeline exports a fallback identity (git_identity); stepupdate run alone
did not. It now sets the same fallback when the app repo has no user.email.

Control: stepupdate-no-identity-pass (HOME and XDG_CONFIG_HOME empty,
GIT_CONFIG_NOSYSTEM=1, GIT_* unset). Old code: 6 passed, 1 failed (this one);
new code: 7 passed, 0 failed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cold CMM run 10: after Atlas_Web_Content brought Data grid 2.30 while the
DataWidgets 2.x filters stayed, the Lots page showed "The Text filter widget
must be placed inside the header of the Data grid 2.0 or Gallery widget" in a
red box in place of every filter. landedOk only knew the login form and the
"An error occurred" dialog, so the after walk PASSed 23/23.

landedOk now also fails on a visible .alert-danger that mentions a widget;
validation messages (.mx-validation-message) and app alert text without
"widget" still pass. widgetrender.js (journey step "rendered") looks inside
one named widget; this is the walk's own check on every page it opens.

Control: smokewalk-landedok-widget-error-catch, plus chromium cases in
smokewalk_predicates.js. Old code: the widget error box gets null (PASS),
exit 1; new code: all landedOk cases ok.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cold WM1 journeys, run next to cold CMM: the pipeline names its containers
${UPG_NAME:-mig}-<name> but never exported UPG_NAME. runapp.py, seeing it
unset with another run's cmm-oldjp present, REFUSED to stop mig-oldjp (the
FM-9 guard), and newjp then found port 8080 taken ("INFRA port 8080 in use by
mig-oldjp"). The pipeline now exports UPG_NAME="${UPG_NAME:-mig}" once, after
git_identity; the guard still protects a runapp call made by hand.

Control: pipeline-exports-upg-name-pass runs the pipeline's own export line,
then runapp stop oldjp with mig-oldjp and cmm-oldjp present. Old code: 5
passed, 1 failed (REFUSED); new code: 6 passed, 0 failed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
harvest.py (PR #3) only ran when a person remembered steps/report.md 3c, so a run that
stopped taught the toolkit nothing (cold WM1). The exit trap now calls bin/feedback_report.py:
harvest drafts go to $WORK/learnings/, and FEEDBACK-REPORT.md says where the run stopped,
lists the drafts, and explains how to send them back (submit-learnings.sh) or file a toolkit
bug (an issue summary of metadata only, checked against the denylist). Rung 2 reports on the
whole run folder. The exit code never changes. UPG_HARVEST=0 skips the harvest.

Controls: feedback-report-{pass,content-pass,nocode-pass,nocode-says-pass,deny-catch},
pipeline-feedback-on-exit-pass (6 fail on the old code, all pass on the new).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…FEEDBACK-REPORT.md

The page still said the repo was private and outside contributions were not open, which
contradicted the invitation to send PRs. It now invites issues, learnings drafts and PRs
under the same bar (a field run or a control), and points bug reports at the issue
summary every run now writes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…151-153 from cold CMM

widgets.md: Atlas_Web_Content 3.9 swaps in Datagrid 2.30, and DataWidgets 2.x filters then
render in red error boxes until DataWidgets is updated in the same rung; plus the Data grid
2.2 DOM shape. BACKLOG: rung 2 plan pre-fill, localedits content id 0 after rung 1, and
resumable journeys.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Modules that rung 1 updated carry content id 0 in the MPR, so rung 2's localedits printed
UNKNOWN ("give --cid Module=<id>") for each one and the run stopped for a hand-written
localedits.args (cold CMM, 7 modules). Before rung 2 starts, rung2_cids adds a
--cid=Module=ID line for every module in rung 1's plan.txt that the args file does not
already name.

Controls: pipeline-rung2-cids-pass, pipeline-rung2-cids-wired-pass (both fail on the old
code, pass on the new).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first live FEEDBACK-REPORT.md (cold CMM rung 2) listed 37 "scripts the agent wrote", all
pipeline output: the patched journey engine copies in <work>/e2e-*/engine (run_journeys.py
patches them, so no hash matches a toolkit file), editroute's replay/*.mdl and localedits'
saved-edits/. harvest.py now skips those folders. On the same run folder: 37 scripts -> 0.

Control: harvest-toolkit-written-pass, and the fixture's harvest-run-pass now carries the three
kinds (both fail on the old code, pass on the new).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…m chart; backlog 153-154 from cold CMM

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…brought (repair row 33)

Cold CMM: Administration 4.5.0 saved Account_Overview's Roles filter with Data Widgets 3.x in association mode; the
kept DataWidgets 2.2.1 made update-widgets rewrite it to the 2.x definition, and the association settings were gone.
On 3.11.8 the filter came back in attribute mode: "Unable to get filter store" on 10.24 and 11.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MendixMau

Copy link
Copy Markdown
Owner Author

CI: 4 failed (cockpit-decide-cards-pass, remedy-keptdeps-file-catch, smoke-users-bad-entity-catch, smokewalk-landedok-pass). These 4 also fail on main's last run (485728e: 10 failed). They come from the runner: no python bcrypt, no git user.name, a long runner path in a wrapped line. This branch fixes the other 6 (the stepupdate controls, via the upgrade-pipeline git identity). Local full suite on 1e87d94: 2731 passed, 0 failed, 34 skipped.

@MendixMau
MendixMau merged commit c282df7 into main Nov 17, 2026
1 check failed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant