Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions upgrade-toolkit/FEEDBACK.md
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,8 @@ MET = run metrics (branch run-metrics): step timing in `state/timing.tsv` with m

LAY = the layout rung (branch feat/layout-rung): why two layout regressions of the LineApp 10.24.13 to 11.12.6 React upgrade passed the gate with '0 regressions' (a vendor module's Dojo table rules halved the Task List gallery, the Static image to Image conversion mapped width unit Auto to percentage). Source: the root-cause note `~/upgrade-lab/lineapp/WHY-NOT-CAUGHT.md`; golden input `controls/fixtures/layout/` (trimmed geometry of the old app and the pre-fix new app). Rows LAY-1 to LAY-12.

RX1 = a held-modules audit on a Mendix 10.24 to 11.12 run, re-checked against the app's own delivered `.mpk` rather than its archived run folder. Rows RX1-1 and RX1-2.

PLM = the PLM ERP LAB Connect 9.24 to 10.24 run (`~/upgrade-lab/tc-apps/TOOLKIT-FEEDBACK.md`, 31 rows PLM-1 to PLM-31; PLM-RI-n in the text is the release-impact pass). Verdict: the lesson is PLM-29 and PLM-26, two silent losses (a local layout edit and unmapped new module roles) that only the screens and smoke rungs showed.

RT = the remedy trial on the notification-service runs (`~/upgrade-lab/tc-apps/TOOLKIT-FEEDBACK.md` section "RT: remedy trial", 14 rows RT-1 to RT-14; evidence `notif/r1/work/REMEDY-TRIAL.md` and `notif/r2/work/REMEDY-TRIAL.md`, run ids "notif RT r1" and "notif RT r2"). The remedy cards were applied option by option; the lesson is RT-1 and RT-3 (letter order picked a pin that broke the app) and RT-8 (a prove-clean update ignored a kept dependent).
Expand Down Expand Up @@ -743,6 +745,8 @@ Add new items at the end with status open; the next triage round gives each one
| LAY-10 | Combo box widget 2.x (the replacement of the Reference, Reference set and Drop-down selectors): the input container has min-width 135px and the placeholder is in the flow, not position absolute, so theme left/top rules on `.widget-combobox-placeholder-text` misplace it and a narrow old control grows | doc | `playbooks/dojo-successors.tsv` notes of the four selector rows: placeholder position absolute, min-width 0 where the old control was narrower | playbook-combobox-placeholder-catch, playbook-lay-rows-9-columns-pass |
| LAY-11 | Image widget (from Static image): the classes land on the wrapper `div.mx-image-viewer`, not on the img, so theme height and padding rules written for the img leave it overflowing | doc | Static Image row of the playbook: `.mx-image-viewer.mx-image-viewer-responsive > img, .mx-image-viewer > img {max-width:100%; max-height:100%}` (two classes, because the widget's own `.mx-image-viewer-responsive img` rule loads after the theme) | playbook-image-wrapper-catch, playbook-lay-rows-9-columns-pass |
| LAY-12 | The React rule `.mx-scrollcontainer-center .mx-layoutgrid .mx-layoutgrid-fluid {padding:0}` zeroes the padding of grids that were not nested on Dojo, because React wraps the page in an extra scroll container region; HTMLSnippet 4.x also wraps in `.widget-events` with flex 1 1 auto, which takes row space | fixed | theme_clash.py kind `react-wrapper`: a padding 0 rule under `.mx-scrollcontainer-*` that targets `.mx-layoutgrid*`, in own or vendor SCSS (summary gains `, W react-wrapper` only when W > 0); the HTMLSnippet part is a playbook note (doc) | theme-clash-react-wrapper-catch, theme-clash-react-wrapper-mutant-blind, theme-clash-react-wrapper-mutant-differs, theme-clash-react-wrapper-clean-app-pass, playbook-htmlsnippet-events-catch |
| RX1-1 | An audit of a real app's held Marketplace modules read Atlas_Core's own `web/main.scss` as a hand-edited vendor file at FILES-route risk, because two of its lines (`@import "../../../theme/web/exclusion-variables"`, `@import "../../../theme/web/custom-variables"`) reach outside the module into the project's top-level `theme/` folder. The same two lines, in the same position, are present in 46 of the app's 52 themesource modules, including modules the audit itself had confirmed carried no other edits (DataWidgets, ObjectHandling): Mendix's own module-template scaffolding ships them by default. An Atlas_Core update regenerates this exact boilerplate; it is not what an update wipes. The app's real custom styling (`theme/web/custom-variables.scss`, `exclusion-variables.scss`) lives outside every themesource module and no Marketplace update ever touches it | fixed | `bin/scss_boilerplate_check.py`: given a themesource module directory, classifies every outward `@import "<n x ../>theme/web/<name>"` line in its `web/main.scss` as BOILERPLATE (name is `custom-variables` or `exclusion-variables`: stock Mendix scaffolding, no FILES route) or RISK (any other name: an app edit, route it with editroute.py FILES same as before) | scss-boilerplate-pass (only the two stock imports: 0 risk), scss-boilerplate-app-edit-risk-catch (a third outward import of app content: 1 risk, FILES route named) |
| RX1-2 | The same audit worked only from an archived run's logs and decisions (no app model survived in the archive), and concluded HTMLSnippet was still on 3.9.9, not migrated to 4.0.0. The real delivered app (pulled from its shipped `.mpk`) had already migrated every instance to the Events widget + a bridging nanoflow and to native HTMLElement; 0 classic HTMLSnippet widgets remained. An archive of decisions and logs is not proof of the shipped model: a module-version or widget-usage audit needs the delivered `.mpk` (or a model export), not the run folder, as its source of truth | doc (closed by guidance) | none (a process note, not a tool change); flagged in playbooks/widgets.md's React-client readiness checklist as a caution on HTMLSnippet-usage claims made from decision logs alone | none |

Recounted from the table on 14 Oct after the QI and V3D rows (563 rows: the 542 of the night merge, which the earlier count had one closed row short, plus QI-1 to QI-10 and V3D-1 to V3D-10, plus HG-1 to HG-3 from the run history PR, 566 in all; counted by the first word of the status column).

Expand Down Expand Up @@ -872,3 +876,16 @@ Recounted from the table on 4 Nov after LAY-1 to LAY-12 (686 rows, counted by th
| backlog | 166 |
| open | 0 |
| total | 686 |

Recounted from the table on 17 Nov after RX1-1 and RX1-2 (688 rows: the 686 above plus two; counted by the first word of the status column: 398 fixed, 95 doc, 14 closed, 7 won't fix, 8 unproved, 166 backlog, 0 open; RX1-1 is fixed, RX1-2 is doc, no other row changed). RX1 = a held-modules audit re-checked against the app's delivered .mpk.

| status | count |
|---|---|
| fixed | 398 |
| doc (closed by guidance) | 95 |
| closed | 14 |
| won't fix | 7 |
| unproved | 8 |
| backlog | 166 |
| open | 0 |
| total | 688 |
80 changes: 80 additions & 0 deletions upgrade-toolkit/bin/scss_boilerplate_check.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,80 @@
#!/usr/bin/env python3
"""scss_boilerplate_check.py: is an @import that reaches outside a themesource module Mendix's own stock
boilerplate, or a real app-specific edit a FILES route must carry forward?

Every Mendix-generated themesource/<module>/web/main.scss ships two lines by default, reaching three levels up
into the project's own top-level theme/ folder:

@import "../../../theme/web/exclusion-variables";
@import "../../../theme/web/custom-variables";

(WM1-style audits mistook these for a hand-edited vendor file at FILES-route risk on Atlas_Core: 46 of a real
app's themesource modules carried the identical line, including modules with zero other edits. An Atlas_Core
update regenerates this same boilerplate every time; it is not what an Atlas update wipes.)

A line that reaches outside the module by the SAME depth but imports something OTHER than those two stock
names is not boilerplate: it is an app edit, and the FILES route applies (move the imported content into the
app's own design module, so the single import-path line is the only thing replayed on every update).

scss_boilerplate_check.py <module-dir> [<module-dir> ...]
<module-dir> is a themesource/<module> directory containing web/main.scss.

Output: one line per outward-reaching import found:
"BOILERPLATE <module> <line>: stock Mendix theme-folder import, not a FILES-route risk"
"RISK <module> <line>: app-specific outward import, needs a FILES route (editroute.py FILES)"
then a summary: "scss_boilerplate_check: N boilerplate, M risk across K module(s)".
Exit 0 if no RISK lines found (including no outward imports at all), 1 if any RISK line was found, 2 on bad input
(a module dir with no web/main.scss).
"""
import os
import re
import sys

STOCK_NAMES = {"custom-variables", "exclusion-variables"}
# Mendix's generated main.scss reaches out with a "../"-only relative path (any depth) into theme/web/.
OUTWARD_RE = re.compile(r'@import\s+[\'"](\.\./[\./]*theme/web/([A-Za-z0-9_-]+))[\'"]\s*;?')


def check_module(module_dir):
main_scss = os.path.join(module_dir, "web", "main.scss")
if not os.path.isfile(main_scss):
print(f"usage: no web/main.scss in {module_dir}", file=sys.stderr)
return None
module = os.path.basename(os.path.normpath(module_dir))
boiler = risk = 0
with open(main_scss, encoding="utf-8") as f:
for line in f:
m = OUTWARD_RE.search(line)
if not m:
continue
name = m.group(2)
stripped = line.strip()
if name in STOCK_NAMES:
boiler += 1
print(f"BOILERPLATE {module} {stripped}: stock Mendix theme-folder import, not a FILES-route risk")
else:
risk += 1
print(f"RISK {module} {stripped}: app-specific outward import, needs a FILES route (editroute.py FILES)")
return boiler, risk


def main(argv):
if not argv:
print("usage: scss_boilerplate_check.py <module-dir> [<module-dir> ...]", file=sys.stderr)
return 2
total_boiler = total_risk = 0
modules = 0
for module_dir in argv:
result = check_module(module_dir)
if result is None:
return 2
b, r = result
total_boiler += b
total_risk += r
modules += 1
print(f"scss_boilerplate_check: {total_boiler} boilerplate, {total_risk} risk across {modules} module(s)")
return 1 if total_risk else 0


if __name__ == "__main__":
sys.exit(main(sys.argv[1:]))
7 changes: 7 additions & 0 deletions upgrade-toolkit/controls/run-controls.sh
Original file line number Diff line number Diff line change
Expand Up @@ -646,6 +646,13 @@ ctl datamove-moved-pass 0 '^MOVED MendixSSO.AuthRequest.Nonce -> MendixSSO_Ext.A
ctl datamove-lost-catch 1 '^LOST .*: 5 before, 3 after$' -- $DM lost
ctl datamove-count-fail-catch 1 '^UNCOUNTED .*count failed after$' -- $DM fail
ctl datamove-bad-plan-catch 2 "not 'Src.Entity.Attr -> Dst.Entity.Attr'" -- $DM bad
# RX1: an audit on a real app read Atlas_Core's own "../../../theme/web/custom-variables" import as a hand-edited
# vendor file at FILES-route risk; the identical line is Mendix's own module-template boilerplate (46 of the app's
# own themesource modules carried it). Only an outward import of something OTHER than those two stock names is a
# real app edit needing a FILES route.
SB="bash $B/../controls/scss_boilerplate_control.sh"
ctl scss-boilerplate-pass 0 '^BOILERPLATE mod .*exclusion-variables' '^BOILERPLATE mod .*custom-variables' '!^RISK' '^scss_boilerplate_check: 2 boilerplate, 0 risk across 1 module' -- $SB pass
ctl scss-boilerplate-app-edit-risk-catch 1 '^RISK mod .*company-brand-overrides.*needs a FILES route' '^scss_boilerplate_check: 2 boilerplate, 1 risk across 1 module' -- $SB catch
# WM1 rerun 7: baseline ran microflow steps, the after walk skipped them: 4 false REGRESSIONs
ctl pipeline-smoke-mf-follows-baseline-pass 0 '^--microflows$' -- bash -c "WORK=\$(mktemp -d); echo '{\"landed\": \"microflow ran\"}' > \$WORK/smoke-before.json; eval \"\$(awk '/^smoke_mf\(\)/{f=1} f{print} f&&/^}|; }\$/{exit}' \${PIPE:-$B/cloud/upgrade-pipeline.sh})\"; smoke_mf"
ctl pipeline-smoke-mf-off-without-baseline-pass 0 '!microflows' -- bash -c "WORK=\$(mktemp -d); eval \"\$(awk '/^smoke_mf\(\)/{f=1} f{print} f&&/^}|; }\$/{exit}' \${PIPE:-$B/cloud/upgrade-pipeline.sh})\"; smoke_mf; echo done"
Expand Down
25 changes: 25 additions & 0 deletions upgrade-toolkit/controls/scss_boilerplate_control.sh
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
#!/usr/bin/env bash
# scss_boilerplate_control.sh CASE: bin/scss_boilerplate_check.py on a fake themesource module.
# pass only the two stock Mendix theme-folder imports (no app edit)
# catch the two stock imports plus one app-specific outward import (a real FILES-route edit)
set -euo pipefail
B=$(cd "$(dirname "$0")/../bin" && pwd)
T=$(mktemp -d); trap 'rm -rf "$T"' EXIT
mkdir -p "$T/mod/web"
case "$1" in
catch)
cat > "$T/mod/web/main.scss" <<'S'
@import "../../../theme/web/exclusion-variables";
@import "../../../theme/web/company-brand-overrides";
@import "../../../theme/web/custom-variables";
S
;;
*)
cat > "$T/mod/web/main.scss" <<'S'
@import "../../../theme/web/exclusion-variables";
@import "variables";
@import "../../../theme/web/custom-variables";
S
;;
esac
python3 "$B/scss_boilerplate_check.py" "$T/mod"
6 changes: 5 additions & 1 deletion upgrade-toolkit/playbooks/widgets.md
Original file line number Diff line number Diff line change
Expand Up @@ -133,7 +133,11 @@ error 4 to 6 steps later. The "Cumulative" checkbox caption lost on the Ticket t
3. Replace, do not patch: Microflow Timer -> **Events** widget (component 224259, min Mendix 9.24, pluggable);
HTMLSnippet -> HTML/JavaScript Snippet 4.x (its 4.0.0 notes: the content property must be re-entered) or HTML
element; classic containers holding a pluggable widget that loads data: keep, but update the pluggable widget to
its newest version for the target.
its newest version for the target. An Events widget + a bridging nanoflow is another real migration shape for
HTMLSnippet, seen on a React-client app: 140 of 146 Events-widget instances wired to one bridge nanoflow, the rest
converted to native HTML element. Either way, judge "is HTMLSnippet still on an old version" from the shipped
`.mpk` or a live `mxcli` catalog query on it (RX1-2), not from an archived run's decision log: a decision log can
predate the fix, or the fix can land in a later rerun the log never mentions.
4. Accept as PASS* only what is proved harmless: the page content equal (shotdiff and text), no failed request, the
error only on leaving the page. A teardown error is still a FAIL if the walk shows a missing value or a stuck
progress bar.
Expand Down
Loading