Skip to content

Sync ako/mxcli: 5 upstream fixes (REST binary body, view entity columns, CE0111 in fragments, grid selection, conversion records); business event messages; verify baseline guard - #1919

Merged
ako merged 25 commits into
mendixlabs:mainfrom
ako:main
Nov 22, 2026

Conversation

@ako

@ako ako commented Nov 22, 2026

Copy link
Copy Markdown
Collaborator

Syncs ako/mxcli:main into upstream: 13 commits since #1915 — 5 upstream issues fixed, business event messages that build and deploy, Studio Pro's conversion records kept, and a guard on accepting new errors into a verify baseline.

check: catch what the build refuses

Pages

Studio Pro's conversion records are kept

mx drops a one-time conversion record it has no provider for when it re-saves Projects$ProjectConversion. After Studio Pro 11.16.1 recorded one, fix design-properties — and the repair after marketplace install, which runs it — carried the re-saved unit back without it, and Studio Pro asked to convert the project again. The harvest now puts back a record the stored list has and the tool's lacks. Fixes #1917

Business events

  • A message published and consumed by one service (an in-app bus, a request/reply loop) is declared once, with its operations as children, as a consumed REST service's operations are written: message X (...) { publish (Entity: M.P); subscribe (Entity: M.C, Microflow: M.H); }. Declaring it twice stored two messages of one name: mx check passed and the deployment failed ("An item with the same key has already been added", measured with mxbuild 11.12.6 --target=deploy and BusinessEvents 3.12.0). The old spelling is deprecated (MDL-DEPR075/076), with a migration page.
  • A message's String lengths are stored (String(36), String(unlimited)); they were dropped and the build applied 200, so mx check reported every attribute's length differing from its entity's. describe prints each operation once, in stored order.

verify and tooling

  • Re-recording a baseline refuses to accept new errors without --accept-errors: --update-baseline is the one command that turns a failing gate green without fixing anything. It now exits 2, naming the errors and leaving the baseline untouched, when the run has errors the existing baseline doesn't hold (lint and mx check).
  • mxcli test asks the Docker daemon first: without --local and with no daemon, it failed only at docker compose up, after generating a .docker/ stack the project never had.
  • Report stacked-bar labels wrap at phone width instead of being cut

claude and others added 25 commits November 21, 2026 13:14
Moves the build of one entity type's attributes and key from
createExternalEntities into contractEntityAttributes, unchanged, so the
single create external entity can map its attributes the same way
(#2497). No behaviour change: the OData executor tests and a live bulk
import (mx check 0 errors) are as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
#2497)

create external entity stored every attribute as a local value, wrote
no key and left $top/$skip off: mx check 11.12.6 reported CE6612 on each
attribute, CE6010 and CE6630, on a contract the bulk import builds at 0
errors. Each stated attribute is now the attribute the bulk import makes
of the property of that name, the key and capabilities follow the
contract, and what cannot build (an unknown name, a type the service
does not publish, a missing key, a contradicting Countable/Creatable/
Deletable) is refused with nothing written. No attribute list imports
them all; a create or modify adding an attribute maps it too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
… mapping (#2497)

An entity an older import made names a property differently (TestApp's
Odata.Devices keys on `_Id`, which today's import calls `DeviceId`), and
describe prints the stored name, so the round trip refused its own
output. A stated attribute of a stored entity is now matched by the
property it maps as well as by name, keeps its name, and so does the key
part. Caught by integration (roundtrip) TestTestAppRoundTrip, which
passes locally with this.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
#2497)

TestDocumentation_SurvivesRewrite/external_entity created its entity
from a client whose MetadataUrl is unreachable, so no contract was
cached. A new external entity's attributes now come from the contract,
and without one the statement is refused. The client now reads a local
file:// contract with the Things set; what the case tests (the doc
comment surviving a rewrite) is unchanged and passes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…24rqgv

odata: map a hand-written external entity's attributes to the contract (#2497)
`[Car_Assignment.ReturnedOn = empty]` was read as one association name:
the reference check called it unknown and MDL047 suggested a not(...)
that kept the dotted name. With a project the access-rule and retrieve
checks now recognise an association followed by an attribute of its far
end, say the entity step is missing, and give the path; without one,
MDL047's suggestion offers that reading beside the absence form.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
A statement appended after an activity with a custom error handler was
spliced onto the activity's normal outgoing flow, between it and the
merge the handler rejoins: the error path skipped it and the next exec
of the unchanged script was refused (MDL-V1-REBUILD). Such an activity
is no longer an insert anchor; the statement goes before the stored
statement after the gap, which is after the merge. Live on 11.12.6: the
re-run is Unchanged and mx check reports 0 errors.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
)

IMPORT ... LINK through a self-association stored as a table named both
junction columns after the one entity, so the insert failed with column
"myfirstmodule$personid" does not exist. Measured on 11.12.6 against
PostgreSQL: Mendix stores them as myfirstmodule$personid1 (parent) and
personid2 (child), and mendixsystem$association records both. The names
now come from that row (parent_column_name is read as well), and the
convention used without a row numbers a self-association's two the same
way. Live: the unfixed build fails, the fixed one links Carol to Alice.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…er's resources (#2449)

create or modify published rest service naming only a new operation
deleted the stored ones and exec said only "Modified". check and exec
now warn with the operations it removes (MDL-REST06), keyed as describe
prints them, so a describe -> exec round trip stays silent (Unchanged).
check also refuses alter ... add resource with a stored resource's name
and drop resource of a missing one, which exec refused after check
passed them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
alter published rest service ... add operation in resource '<name>' {
<operation>; } and drop operation <method> ['<path>'] in resource
'<name>'. The operation is create's own line, built by the same visitor
code; the resource's other operations, with their stored $IDs and what
MDL cannot spell, are written back as stored. A duplicate operation or a
missing resource is refused by check and exec. Live on 11.12.6: mx check
0 errors, a re-run is refused as a duplicate, and the describe output
re-executes Unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…dlQuoted (#2449)

The add/drop operation refusals wrote the resource name and path into
copyable MDL as '%s', which the emitter quoting gate
(TestDescribers_HaveNoHandRolledStringLiterals) refuses: a name with a
quote in it would come out as broken MDL.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…24rqgv

Batch K: XPath missing entity step, splice after handler rejoin, self-association import, published REST operations
…es in MPR017 (#2501, #2502)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82tETa46YQmMS58pQjL
…nics

report all called each kind's collector with that kind's own cobra
command, which cobra never executed, so cmd.Context() was nil. The
dependencies report passed it to the Marketplace and panicked ("cannot
create context from nil parent") with a token and Marketplace modules;
its vulnerability check failed as "net/http: nil Context". And since
files were written only after every kind was collected, the panic lost
the four reports already computed. report readiness -p had the same
call.

- collectKind gives the kind's command the caller's context, and turns
  a panic in a collector (and in the scored report) into that kind's
  error: one failed page, the rest written.
- Each kind's JSON is written as soon as it is collected.

Fixes #2500

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
The lint baseline comparison counts findings the baseline held that the change
removed, and verify printed that line, but its lint stage summary and
--format json left resolved at 0 while the mx check stage filled it. Read
the baseline's fixed count into the stage outcome.

Fixes #2506

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82stfDEB1TGfJLVwUxc
fix(report all): give each kind a context, and survive a kind that panics
lint: skip external entities in MPR017/MPR019 and count mappings as writes in MPR017 (#2501, #2502)
The `queue` ref runs from a task queue to the flow it runs, and module
dependencies counted it as the queue's module using that flow's module.
With one queue in Rooster running flows of Ruilen, Int_Planning and
Verlof, report architecture (and GRAPH_MODULE_CYCLES, behind the
module_cycles() lint builtin) showed a false 9-module cycle. The use
runs the other way: the caller places the call in the queue.

- refs gains `enqueue`: the microflow that places a call in a task queue
  -> the queue, so the caller's module is seen to use the queue's.
- `queue` and `by_name` (a constant naming a flow: configuration) are
  left out of module dependencies and module cycles
  (catalog.RefKindsNotModuleUse, shared by the catalog and the report);
  they still keep their flows reachable.
- Catalog schema 50.

Fixes #2503

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
fix(verify): the lint stage reports fixed baseline findings as resolved
…L-JSON03)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011oK82tETa46YQmMS58pQjL
fix(catalog): task queues and constants are not module dependencies
check: reject a JSON structure custom name Mendix refuses (CE9524)
@ako
ako merged commit 6176c66 into mendixlabs:main Nov 22, 2026
15 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment