Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
36 commits
Select commit Hold shift + click to select a range
83cb9ff
feat(brain): milestones live on the slices they group
claude Nov 16, 2026
943aa11
fix(report): the traceability hint names commands that exist
claude Nov 16, 2026
03736b4
feat(report): the software map shows the app by feature
claude Nov 16, 2026
4231d1f
Merge remote-tracking branch 'origin/main' into claude/gifted-fermi-2…
claude Nov 16, 2026
ecdd9b7
Merge pull request #2332 from ako/claude/gifted-fermi-2udffj
ako Nov 16, 2026
c14176b
Keep a show message's other translations on an in-place edit (#2327)
claude Nov 16, 2026
d9f491d
Pair a rebuilt flow's objects across Studio Pro's order (#2329)
claude Nov 16, 2026
f8c6af5
Warn when a grant's where clause narrows nothing (#2320)
claude Nov 16, 2026
7a05662
docs(proposal): hedge the 'no cheaper mx check' claim in the diagnost…
claude Nov 16, 2026
00d03d3
Keep a retyped association's own id in the transplant (#2322)
claude Nov 16, 2026
3dbc601
Write nothing when a move entity's target is refused (#2323)
claude Nov 16, 2026
5b87647
Say what a git merge plan's whole-unit choice drops (#2325)
claude Nov 16, 2026
65f80c4
Merge remote-tracking branch 'origin/main' into claude/issue-triage-p…
claude Nov 16, 2026
5d7fd9b
docs(proposal): mxcli verify staged gate with baseline comparison (#2…
claude Nov 16, 2026
4fbdd33
feat(check-gap): parity harness and spike results for mendix check as…
claude Nov 16, 2026
6851588
fix(check-gap): add mdl 1; header to fixtures for the conformance gate
claude Nov 16, 2026
687be3a
Merge pull request #2336 from ako/claude/issue-triage-prioritization-…
ako Nov 16, 2026
140e3f8
docs(proposal): verify gate — address review (finding identity, run-a…
claude Nov 16, 2026
acbef62
Store `Visible: roles (…) and <condition>` as roles plus condition (#…
claude Nov 16, 2026
12a6510
fix(exprcheck): replace() is not a Mendix expression function
claude Nov 16, 2026
c579a15
Make alter navigation/menu insert re-runnable (#2312)
claude Nov 16, 2026
71a6828
Draw a create-or-modify insert's connectors as the script states (#2311)
claude Nov 16, 2026
1aa998c
feat(check): type-check DECLARE, SET and RETURN values against the de…
claude Nov 16, 2026
fda2585
Read the roles-in-expression refusal's keys by literal (#2319)
claude Nov 16, 2026
68767c2
Merge pull request #2335 from ako/claude/inspiring-planck-mp8ak5
ako Nov 16, 2026
b31aa82
Merge pull request #2337 from ako/claude/issue-triage-prioritization-…
ako Nov 16, 2026
46fc8f3
Merge pull request #2338 from ako/fix/replace-not-an-expression-function
ako Nov 17, 2026
1f78bde
Merge pull request #2339 from ako/fix/declare-set-return-type-check
ako Nov 17, 2026
6af5d93
Merge pull request #2334 from ako/claude/ecstatic-franklin-smuukm
ako Nov 17, 2026
231f42b
Import Core in a generated Java action that uses it (#2310)
claude Nov 17, 2026
87724d8
Name the literal secrets run restart cannot replay (#2313)
claude Nov 17, 2026
a87b33c
Serialise mxcli processes writing one project (#2326)
claude Nov 17, 2026
fb7954e
Run fix widgets with the project's exact mx version (#2321)
claude Nov 17, 2026
c49b51c
Merge remote-tracking branch 'origin/main' into claude/issue-triage-p…
claude Nov 17, 2026
d234c0b
Keep the run-local skill within its line bound (#2313)
claude Nov 17, 2026
69183c8
Merge pull request #2343 from ako/claude/issue-triage-prioritization-…
ako Nov 17, 2026
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
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "cmd/mxcli", "date": "2026-11-16", "refs": ["ako/mxcli#2325"], "ce": [], "symptom": "`mxcli git merge p4 --plan` with \"take\": \"theirs\" on the conflicted Shop domain model committed a merge without Shop.OursOnly, an entity only ours had added and nobody conflicted on. The review listed it under Yours; nothing said the choice would drop it.", "file": "cmd/mxcli/cmd_git_merge.go wholeUnitDropReport / unitChanges / onlyIn; called in mergeRun's conflict stage and in applyPlan before the units are taken. Test: cmd/mxcli/gitmerge_whole_unit_drop_2325_test.go", "cause": "The plan resolves per unit by checking the chosen side's .mxunit out whole, so every change of the other side to that unit goes; the property report is per element and did not say the resolution is not.", "insight": "**The two mx diff files already hold the answer: a side's top-level change descriptions in a unit that the other side does not also list are exactly what taking the other side drops without a conflict.** Same description on both sides = the conflict itself (the choice), so it is not reported. The review prints both directions per unit before the plan is written; --plan prints only the chosen direction, before the commit. Element-level resolution (take per element) would need writing a merged unit, which the plan deliberately avoids (it checks a side's bytes out whole); warning is the honest scope. Live on 11.12.6 with real mx merge + mx diff: the review names Entity 'Shop.OursOnly' for theirs and 'Shop.TheirsOnly' for ours, and the theirs plan warns about OursOnly before 'Model merge committed'. Control: two sides changing only the same element report nothing."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "cmd/mxcli", "date": "2026-11-16", "symptom": "`mxcli report traceability` on a brain with no plan tells you to run `mxcli brain promote --all`, which fails with 'unknown flag: --all'", "cause": "The hint was written from memory of a flag promote never had; nothing checked the commands an error message tells you to copy", "file": "`cmd/mxcli/cmd_report_kinds.go` (noRequirementsError)", "insight": "A copy-paste hint is code the user runs: test it against the cobra command tree (Find + Flags().Lookup) so a renamed or imagined flag fails the build, not the user. Found by following the hint during an end-to-end run", "refs": [], "test": "TestNoRequirementsHintUsesRealFlags"}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "cmd/mxcli", "date": "2026-11-17", "refs": ["ako/mxcli#2321"], "ce": [], "symptom": "`mxcli fix widgets` on macOS with an 11.12.2 project failed: mx update-widgets 'Project is in different Mendix version - 11.12.2' (exit 4). The same project and command worked in a Linux container with --mx pointing at MxBuild's mx.", "file": "cmd/mxcli/cmd_fix.go runFixTool / resolveFixMx; cmd/mxcli/docker/check.go LocalMxForVersion. Test: cmd/mxcli/fix_mx_resolve_2321_test.go", "cause": "runFixTool resolved mx with docker.ResolveMxForVersion, which treats the project's version as a preference: an mx on PATH comes first, then the newest Studio Pro install or any cached mx. update-widgets and rename-design-properties only work on their own version, so any fallback is a guaranteed failure.", "insight": "**For a tool that edits the project in its own version's format, the version is a requirement, like `mxcli new` (ResolveMxForNewProject), never a preference.** resolveFixMx takes an explicit --mx as given, else an exact-version local mx, else (Linux) downloads that version, else (macOS/Windows, where the CDN's Linux mx cannot run) refuses naming the version and the --mx route. Not measured on macOS (none here): the fix is the resolution order, tested by injected lookups; on Linux an 11.12.6 project still resolves the cached 11.12.6 mx and fix widgets runs. The issue's other half — `replace <pluggable widget> with` writing a stale definition (CE0463) — did not reproduce with stock pluggable widgets (combo box, pop-up menu replaced inside a container on 11.12.6: no CE0463); it needs the tester's dropdown-menu widget."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "cmd/mxcli", "date": "2026-11-17", "refs": ["ako/mxcli#2326"], "ce": [], "symptom": "Four `mxcli exec` processes creating one entity each on one project in parallel all failed ('disk I/O error (5898)', 'replaced on disk while unwritten changes were held', a temp-file rename error) and rolled back. Measured here (11.12.6 blank project, 8 parallel scripts of 2 entities + 1 association each, 3 rounds): every process exited 0 and the project held 10, 14 and 14 of the 16 entities — updates lost with success reported.", "file": "internal/projectlock/lock.go Acquire / BusyError; cmd/mxcli/project_write_lock.go programWritesModel / acquireProjectWriteLock (exit 75); wired in cmd/mxcli/cmd_exec.go and cmd/mxcli/oneliner.go. Test: internal/projectlock/lock_test.go", "cause": "A write run is many unit files plus the .mpr index, read at connect and written later, with a rollback that restores backups on failure; nothing serialised two processes, so each wrote over a project the other had changed, or rolled back over it.", "insight": "**The reported failure was the lucky case: the measured one is a silent lost update.** Four writers in this environment never collided; eight with an association each did, every round, and every process said it succeeded — so the control has to count what landed, not exit codes. The fix is a per-project O_EXCL lock file (.mxcli/write.lock, portable to Windows) held for the whole exec / -c run — not per write, since a failed run's rollback is itself a write — with a 2s heartbeat; a lock whose same-host pid is dead, or whose heartbeat stopped 30s ago, is taken over, so os.Exit paths that skip the deferred release cannot strand the project. The second writer waits up to 60s, then exits 75 (EX_TEMPFAIL) naming the holder. Programs of only show/describe/select/... take no lock. With the lock: 16/16 entities and 8/8 associations in all three rounds, mx check 0 errors. Not locked: the REPL session, and other writing commands (fix widgets, marketplace)."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "cmd/mxcli", "date": "2026-11-17", "refs": ["ako/mxcli#2313"], "ce": [], "symptom": "A run started with `--constant 'Encryption.EncryptionKey=@env:HELPDESK_ENC_KEY' --constant 'Helpdesk.OpenRouterKey=@env:OPENROUTER_KEY'` (and other arguments elided as '...') could not be restarted: `run restart` printed the generic '--db-password @env:' hint, which read as 'constant references are not replayed'.", "file": "cmd/mxcli/run_redact.go redactedArgNames; cmd/mxcli/cmd_run_lifecycle.go restartSecretRefusal. Test: cmd/mxcli/run_restart_redacted_names_2313_test.go", "cause": "The constant references are kept by redactArgs (since 21a9e0676) and replayed; with exactly the reported --constant arguments nothing is redacted and restart proceeds (measured through detachArgs + redactArgs). What restart refuses is a value stored redacted, and its message named none, so one literal flag among several references was indistinguishable from the references failing.", "insight": "**Reproduce the reported arguments through the same code path before believing the reported cause.** detachArgs + redactArgs on the issue's command line keep both @env: constants and redact nothing, so the refusal must have come from an elided literal (or a state file from an older binary). The fix is the message: it now lists each redacted flag by name ('--constant Helpdesk.OpenRouterKey', '--db-password', '--runtime-setting DatabasePassword') and says the reference values are replayed. Not booted live: the claim rests on the state-file arguments, which are what restart replays."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/backend", "date": "2026-11-16", "refs": ["ako/mxcli#2312"], "ce": [], "symptom": "`alter navigation Responsive { insert { menu item 'Helpdesk chat' ( OnClick: show page Helpdesk.Chat ) } };` run twice left two 'Helpdesk chat' items, and the second run reported 'Altered navigation Responsive' instead of Unchanged.", "file": "mdl/backend/menumutator/mutator.go InsertMenuItems / withoutItemsAlreadyThere / sameMenuItem. Test: mdl/backend/menumutator/insert_rerun_2312_test.go", "cause": "InsertMenuItems spliced the built items in unconditionally; insert is the one menu operation whose second application is not a no-op.", "insight": "**An insert is re-runnable when 'already there' means 'already there as this statement builds it', compared without element ids and with the stored captions kept to the languages the statement writes.** Same caption and same content in the target list: dropped, so the patch write is elided and the run reports Unchanged. Same caption, different content (action, sub-items): refused naming `replace`, because inserting leaves two items of one caption and skipping drops the change. Comparing by caption alone would have silently ignored an edited action on re-run. Live on 11.12.6: run 1 Altered, run 2 Unchanged, one item; the edited OnClick is refused before anything is written; mx check 0 errors."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/backend", "date": "2026-11-16", "refs": ["ako/mxcli#2323"], "ce": [], "symptom": "`exec --continue-on-error` of a MOVE ENTITY whose target write was refused reported the statement failed, and the entity was in neither module: removed from the source, not in the target. 12 chained moves on a generated project lost 9 entities and mx check then reported 53 'no longer exists' errors.", "file": "mdl/backend/modelsdk/association_move_write.go MoveEntity / refusedContents. Test: mdl/backend/modelsdk/issue2323_move_atomic_test.go", "cause": "MoveEntity persisted the source domain model, then built and persisted the target; a refusal of the target came after the source was on disk. The default exec rolls the whole script back, so only --continue-on-error kept the half-done move.", "insight": "**A statement that writes two units must decide both before writing either.** Both domain models are now built and encoded first and checked against the refusals the write path makes on content alone (canon.DuplicateElementIDError, canon.BareAttributeRefError); a target write that fails anyway restores the source's stored bytes with UpdateRawUnitPatch (no transplant onto the restore). The test provokes the refusal the way #2322 did — an element in the target with the moved entity's $ID — and without the fix it reports exactly the symptom ('entity Account is gone from Administration'). With #2322 fixed the reported chain no longer reaches the refusal, which is why the test plants one."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/backend", "date": "2026-11-16", "refs": ["ako/mxcli#2327", "mendixlabs/mxcli#1870"], "ce": [], "symptom": "`create or modify microflow` that changed one show message's text reported `spliced: 1 edited in place`, and the message kept only en_US; its nl_NL translation was deleted. Measured on Administration.ChangePassword of the 11.12.6 blank project.", "file": "mdl/backend/mfmutator/splice_edit.go swapKeeping / carryTranslations. Test: mdl/backend/mfmutator/splice_edit_translations_2327_test.go", "cause": "The in-place edit serializes the activity from the statement (one language per text) and saves through UpdateRawUnitPatch, which passes ContentsOwnTranslations, so canon's translation carry never runs on it. Other unstatable properties of an edited activity already had their own carry (carryUnstated, #2169); translations had none.", "insight": "**A patch write opts out of every canon carry, so each thing a splice rebuilds from MDL needs its own.** The splice writes as a patch on purpose (a carry would undo a deliberate removal), but its in-place edit is a rebuild of one activity; it gets canon.CarryTranslations applied to that activity alone (stored activity vs replacement, paired by path), then unmarkRemoved on the result so the carried stored translation elements are not refused as removed. Control: an edit of the only language lands unchanged; without the carry the nl_NL assertion fails."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/executor", "date": "2026-11-16", "refs": ["ako/mxcli#2311"], "ce": [], "symptom": "A `create or modify microflow` that inserted statements succeeded ('spliced: 1 inserted'), and the second run of the same script was refused: 'the connectors of the DeleteObject at (1125, 200) are redrawn (its @anchor, @curve or @merge)' (MDL-V1-REBUILD). The refusal's own advice, drop + create in one script, rewrites the flow on every run. Measured on Administration.ChangePassword, 11.12.6 blank project.", "file": "mdl/executor/cmd_flow_modify.go patchDiff.before / statedCurves; mdl/executor/cmd_alter_flow.go flowVectors; mdl/backend/microflow_mutation.go MicroflowFragment.InCurve/OutCurve, FlowVectors; mdl/backend/mfmutator/splice.go spliceOnFlow; mdl/executor/flow_verdict.go deliberateRebuildAdvice. Tests: mdl/backend/mfmutator/splice_stated_curves_2311_test.go, mdl/executor/issue2311_insert_stated_curves_test.go", "cause": "The splice drew the connector into the inserted run and the one out of it with its own control vectors (sideVector, +/-15), where a create of the script draws them as its @curve states and straight ('0;0', which describe omits) where none is stated. The next diff compared the stated curve with the stored one and called the connector redrawn.", "insight": "**A splice that a re-run cannot recognise is not done: after any create-or-modify write, run the same script again and expect Unchanged.** That is how this was found: the first run of the reproduction succeeded, the second refused. The fix keeps the splice's own vectors for `alter` (which states no curves) and, for create or modify, passes the curve on the declared statement before the run (connector in) and on the run's last statement (connector out), only where both are plain statements. Live: run 1 spliced, runs 2 and 3 Unchanged, mx check 0 errors. The reporter's second refusal ('the stored statement \"Join\" has no activity to address') did not reproduce on flows mxcli wrote — describe prints their merges as @merge, and adding a handler before one spliced and re-ran Unchanged — so it needs their flow; the deliberate-rebuild advice now says to run drop + create once and keep the resulting create or modify, which is the reporter's own workaround."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/executor", "date": "2026-11-16", "refs": ["ako/mxcli#2320"], "ce": [], "symptom": "`grant read * on entity Shop.Customer to Shop.Clerk where [Name != '']` on a role that already had an unconstrained rule reported 'Granted access' and added a second rule; the role still read every row. exec, check and lint said nothing.", "file": "mdl/executor/cmd_security_write.go grantOverlapWarning; mdl/executor/validate_access_rule_overlap.go (ValidateAccessRuleOverlap no-project, CheckAccessRuleOverlap --references); wired in validate_program.go and cmd/mxcli/cmd_check.go. Test: mdl/executor/issue2320_access_rule_overlap_test.go", "cause": "A rule is keyed by role set plus constraint (#936), so a grant with a WHERE is a new rule, not a narrowing; access rules add up at runtime. That is correct model behaviour, but nothing said that the constraint limited nothing.", "insight": "**Report each overlap once, by where its other half comes from.** The check tracks per entity and role whether an unconstrained and a constrained rule are held and whether each came from the project or the script; the no-project pass reports script-vs-script overlaps and the --references pass only those involving a stored rule, so `check -p --references` does not print the same warning twice. `revoke all` clears the role (the documented way to narrow); a partial revoke changes rights, not which rules exist, so it clears nothing. exec prints the warning as a notice, so it still shows on a re-run that writes nothing (the overlap is still there). Live on 11.12.6: check --references and exec both warn; revoke all + grant where leaves one constrained rule and mx check reports 0 errors. Control: a constrained grant to a role with no unconstrained rule warns nothing."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/other", "date": "2026-11-16", "refs": ["ako/mxcli#1258"], "ce": ["CE0117", "CE0844"], "symptom": "`declare $X Integer = 'abc';`, `set $X = 'abc';` and `return $Decimal` from a microflow that returns String all passed `mxcli check -p` and `exec`, then MxBuild reported CE0117 \"Error(s) in expression\" (Mendix CLI 2: CE0844 \"The expression is of type String but should be of type Integer/Long\").", "file": "mdl/exprcheck/target_kind.go checkTargetKind; Context.Target in mdl/exprcheck/interfaces.go; mdl/exprcheck/adapters/check.go checkExprAs / primitiveTarget / setTargetKind. Tests: mdl/exprcheck/target_kind_test.go TestTargetKindMatrix, mdl/exprcheck/adapters/target_kind_test.go", "cause": "The slot table declared DeclareStmt.InitialValue (DeclareType), MfSetStmt.Value (TargetVariable) and ReturnStmt.Value (MicroflowReturn) with a ResolveBy, and checkSlotKind skips every slot with a ResolveBy, so nothing compared the value with the declared type. Same gap #2096 closed for a call argument.", "insight": "**A ResolveBy entry in the slot table that nothing resolves is a rule that looks present and checks nothing; grep the ResolveBy values against what the adapter actually qualifies.** Measured before writing the rule: 216 microflows (3 positions x 6 primitive types x 12 value shapes) in one project, one `mx check -j` on 11.16.1, results keyed by the JSON document-name so one run answers the whole matrix. All three positions behave identically, every rejection is CE0117, and the accepted set equals callArgAccepted, so the predicate is reused and the matrix is the test. `empty` into a Boolean is also CE0117 and is deliberately left silent, as for a call argument. Overlap: MDL041 already reports a COMPUTED Decimal into Integer/Long by DECLARE/SET, so those are skipped here (a bare Decimal literal or variable, and any RETURN, are not MDL041's). False-positive sweep: a throwaway test parsed every .mdl in the repo and ran the adapter with no project, 1649 flows in 1267 scripts, 0 hits after the MDL041 skip (1 before: the intentional negative test bug-tests/f2-decimal-func-into-integer.fail.mdl). `check -p` is blind to this on a script that also has an unresolved reference, so a sweep through the CLI against a stand-in project reports zero for the wrong reason; run the adapter directly. Control: stubbing checkTargetKind to return nil fails TestTargetKindMatrix, the wiring test and the nanoflow test."}
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
{"area": "mdl/other", "date": "2026-11-16", "refs": ["mendixlabs/mxcli#2308", "ako/mxcli#1258"], "ce": ["CE0117", "CE7811"], "symptom": "`return replace($S, 'a', 'b');` passed `mxcli check -p` and `exec`, then MxBuild reported CE0117 \"Error(s) in expression\" (Mendix CLI 2: CE7811 \"Unknown function 'replace'\"). `mxcli syntax microflow.expressions.functions` listed `replace(String, String, String)`.", "file": "mdl/exprcheck/func_checker.go funcTable (the `replace` row, removed); mdl/exprcheck/unknown_funcs.go nearestFunc (sorted candidates). Test: mdl/exprcheck/unknown_funcs_test.go TestUnknownFunctionCalls; fixture mdl-examples/check-gap/fixtures/f09_badfunc.mdl on the check-gap branch", "cause": "funcTable is documented as every Mendix built-in and UnknownFunctionCalls (MDL044) trusts it, so one row for a function Mendix does not have makes the call valid to mxcli and the generated syntax docs. The string replacers are replaceAll and replaceFirst.", "insight": "**A wrong row in funcTable is a false negative in MDL044 and a false claim in the generated docs at once, and the fastest way to find one is to run a function name through an independent checker, not to read the table.** Found by the check-gap parity run (one broken script per miss, mxcli check -p vs mx check vs mendix check): only 1 of the 5 misses was a missing rule, this one was a wrong data row. Fixing it exposed a second defect: nearestFunc iterated the map, so replace (one edit from both replaceAll and replaceFirst) suggested one or the other at random per run; it now walks sorted names. Control: replaceAll and replaceFirst still pass; revert the row removal and TestUnknownFunctionCalls fails with the reported symptom (expected unknown func, got none)."}
Loading
Loading