[MendixMau]

Closes #1788

Why

A single-flow logic edit currently costs a full mxcli test cycle (regenerate test MDL, mxbuild, boot/reload the runtime, run the suite) even though the actual flow-plus-assertion cost, once the app is already running, is a few milliseconds. This adds a second, opt-in route for that narrower case: checking one or a few flows against a live app without touching test MDL files at all.

What changed

New command mxcli test call <cases.json|-> [--repeat N], requires an app already booted with mxcli run --local --test-endpoint:

• cmd/mxcli/cmd_test_call.go — CLI wiring.
• cmd/mxcli/testrunner/call.go — case parsing (ParseCallCases, single object or array), stale-model check (checkModelNotStale).
• cmd/mxcli/testrunner/endpoint.go / endpoint_call.go — new call route on the existing test endpoint.

Case shape: {flow, params, setup, expect, throws, rollback}.

• params/setup build objects via $new (create), $find (lookup), $ref (reuse a $named object from setup), nested associations inline.
• expect: [{path, op, value}], evaluated before rollback. path supports $result, $<param>, association hops (e.g. /Member/Member), and count:Entity[xpath] for row counts. op is = != < <= > >= contains isnull notnull.
• throws asserts the flow raised.

This reuses the existing --test-endpoint transaction-and-rollback machinery (Core.microflowCall) and the same known.contains(flow) microflow allowlist the pre-existing run route already enforces — it is not a new security surface, just a second entry point into the same gated mechanism.

The all-or-nothing diff is purely additive: 5 files, 744 insertions, 0 deletions — no existing code path is touched.

How tested

Unit:

go test ./cmd/mxcli/testrunner/...

→ ok github.com/mendixlabs/mxcli/cmd/mxcli/testrunner 13.442s, all green, including 2 new tests (TestParseCallCases..., TestCheckModelNotStale...) plus the full pre-existing suite (watch-loop, stale-flow, mtime tests) unaffected.

Real app (~/mta-lab/w/rt1, E1 procurement domain, 8 pre-existing cases — 5x invoice pay/approve/dispute, 2x price/qty matching, 1x ERP posting), each step's 1-minute load average from uptime noted:

step	result	load
clean baseline	8/8 PASS	26.3
wrong @expect (mutated expected value on case 120-1)	1 FAIL (the mutated case), 4 PASS — stays red	21.8
seed pay-guard bug (ACT_Invoice_Pay spliced to skip a status check) + mxcli run restart	case 120-5 flips to FAIL: actual: PAID where APPROVED was expected	21.3-23.7
revert bug + restart	5/5 PASS again	16.6-20.8
rollback-leaves-0-rows: call A creates+counts an Invoice (count=1, pre-rollback); independent call B counts the same row	count=0 in call B — confirms no cross-call leakage	26.8
stale model: edit model without rebuilding, then call	refused: Error: the model changed after the running app was built (...) a call now would test the OLD model and could report a stale green	25.4
broken microflow (references a dropped entity) + restart	build fails [CE1613]; run restart reports failed: initial build failed; call never reachable — never a false green	23.2
cleanup + restart	8/8 PASS restored	21.3
warm timing, --repeat 20, 8 cases/round	round 1: 19.2 ms/case → round 14-20: 7.3-8.4 ms/case	21.6-22.8
re-confirmation under load < 15 (separate --repeat 20 run)	round 1: 10.3 ms/case → round 13-20: 3.5-4.9 ms/case — same convergence shape, even faster	9.5-11.1

Repro commands:

mxcli run --local --test-endpoint --detach --db-type hsqldb --app-port 8301 --admin-port 8311 --serve-port 7901 -p App.mpr
mxcli test call ~/mta-lab/speedlab/cases/e1-all.json -p App.mpr
mxcli test call ~/mta-lab/speedlab/cases/e1-all.json -p App.mpr --repeat 20

Regression evidence

Diff touches only the 5 new/extended files listed above — no existing run/test code path is modified, confirmed via git diff --stat origin/main...HEAD. Full cmd/mxcli/testrunner suite green before and after on this branch.

Known limitation

Additive and opt-in: does not replace file-based mxcli test (no test discovery, no watch mode of its own — see mxcli test --watch, #1786, for that loop). Requires an app already booted with --test-endpoint.

Checklist

• Issue filed first (feat: mxcli test call — JSON runtime executor against an already-running app #1788)
• go test ./cmd/mxcli/testrunner/... green
• Real-app A/B with negative controls, load noted per run
• No existing code path touched (purely additive diff)

🤖 Generated with Claude Code