[ako]

Fixes the two class:urgent items on the triage board (#1357). There is one commit per issue, so the PR can be split if needed. The session could push to only one branch.

1. #2168: describe puts the step after an if/else join inside while true into the THEN branch

Upstream: mendixlabs#1829.

Reported symptom (reproduced verbatim on Mendix 11.14.0):

    if $i = 2 then
      set $created = $created + 1;
      set $i = $i + 1;          -- moved into the then branch
    else

If you rebuild from that text, the ELSE branch skips the increment and the loop never ends.

Cause. Inside while true, every node is on a cycle. Three reachability walks that pair a split with its join run past the split, round the loop, and into the split's own branches:

• findMergeForSplit's bypass filter rejected the real join.
• resolveNestedMergeID swapped the loop header in as the join.

Fix. All three walks for a split now stop at that split (collectReachableDistances, splitJoinCandidateDoesNotHaveDownstreamBypass, resolveNestedMergeID).

Evidence.

• TestDescribeWhileTrue_StepAfterIfJoinStaysAfterEndIf failed with the reported text before the fix. It also checks the description rebuilds the stored graph (assertDescriptionRebuildsGraph).
• Controls, each barrier removed on its own, each fail one test:
  • no distance barrier: ExitInElseStaysInElse fails
  • no bypass barrier: StepAfterIfJoin fails
  • no resolveNestedMergeID barrier: StepAfterIfJoin fails
• I diffed every microflow and nanoflow description over TestApp (261) and PedApp (29), baseline binary against fixed. A first version that fixed only two of the walks changed WorkflowCommons.SUB_UserTaskView_Migrate: the loop exit came out of its ELSE arm. That is why the third barrier exists, and ExitInElseStaysInElse pins it. Final diff: 0 lines on both corpora.
• Real project: describe → change the log text → drop + create → describe gives the same text. The before binary's rebuild, described by the fixed binary, shows the ELSE arm genuinely missing the increment.

2. #2169: an in-place edit resets a REST call's proxy (and the activity's documentation)

Upstream: mendixlabs#1830.

Reported symptom: after changing Timeout: 300 to 301, exec prints "spliced: 1 edited in place", and the dump changes RequestProxyType from Override to DefaultProxy and drops ProxyConfiguration.

Cause. swapKeeping carries the stored $IDs, but the content is serialized from the statement. MDL has no spelling for the proxy or for an activity's documentation, so the writer's defaults win.

Fix. mfmutator.unstated is a table, keyed by $Type, of properties MDL cannot state:

• ActionActivity.Documentation
• RestCallAction.RequestProxyType and ProxyConfiguration

These are copied from the stored element on an in-place edit. The carried sub-element's $IDs are taken off the removed set, so the integrity check accepts them.

Evidence.

• The TestSplice_EditKeeps* tests (Override, NoProxy, documentation) failed with exactly the reported values before the fix.
• Real run on Mendix 11.14.0: I created a REST call, patched its stored proxy to Override, and confirmed with mx check (0 errors) that the patched baseline is valid. Then I spliced a timeout-only edit with each binary:
  • before: ProxyConfiguration: null, RequestProxyType: "DefaultProxy"
  • after: proxy kept, timeout 301, mx check 0 errors
• describe → exec afterwards reports Unchanged.

Out of scope: the reset that drop + create causes is #2178, filed separately.

Checks

• make build && make test && make lint && make check-mdl pass locally.
• Bug-test scripts: mdl-examples/bug-tests/2168-…, 2169-…
• One finding file per fix; make check-findings passes.
• Docs: alter-microflow.md and the write-microflows skill now say what an in-place edit keeps; the describe-round-trip-gaps pattern page is synced.

---

Generated by Claude Code