Repository navigation
Sync ako/mxcli: mxcli verify; production-readiness reports; Studio Pro key order; 4 upstream fixes; MPR018/MPR019 - #1915
Merged
Merged
Conversation
A project that Mendix's consistency check rejects cannot be deployed, but `mxcli report` scored only the lint rules, so it could score 99/100 and not build. Measured on testdata/pedapp: 99/100 with 1055 consistency errors. The report now runs `mx check` on a temporary copy of the project, using the same path as `docker check` (docker.CheckFindings), and puts the verdict above the score: - yes; - no, with the error count; - not checked: no mx for the project's version, or --no-mx-check. It says why and how to set it up, and never means clean; - unknown: the parsed findings disagree with mx's own count. The findings are listed under "Consistency Check", grouped by code with a count, the message and a few locations: a stale theme gives hundreds of one code (924 CE6083 there). CE0463 gets a hint about `mxcli fix widgets`. JSON carries every finding and a `builds` field that is absent when the check did not run. The findings are not scored: no category weight can express "does not build", and a few lint fixes would hide it. The check covers the whole project, also under --modules, and the page says so. New flags: --no-mx-check, --mx-check-results <file> (saved mx check or docker check output) and --mxbuild-path. Docs and skills updated; the CI example now also fails on `builds != true`. Closes #2423 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
…new findings only Step 2 of PROPOSAL_verify_gate.md (#2333). Stages run in cost order and every selected stage runs (--fail-fast stops at the first failure). Findings are compared with per-stage baselines in --baseline-dir; mxcheck reads mx check --json and applies MxBuild's staging rule (unmasked / hidden). Exit codes: 0, 1 usage, 2 could not run, 16+bits per failed stage. Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82stfDEB1TGfJLVwUxc
…d of failing Closes #2413. On a clone without submodules, `make ci-local` could not pass: --- FAIL: TestIssue1640_RegisteredListMarkersAreWhatStudioProStores lstat ../../../testdata/testapp/TestApp/mprcontents: no such file or directory 125 other packages passed. testdata/testapp is a git SUBMODULE (ako/TestApp, a gitlink pinned at 11a8fcac) that .github/workflows/push-test.yml checks out with `submodules: true`, so CI always has it and `git clone` without --recurse-submodules does not. studioProUnits walked both roots unconditionally and fed WalkDir's lstat error into t.Fatal. That matters more than one red test. .claude/skills/steward/SKILL.md makes `make ci-local` the gate to run before every push, so on a fresh clone the gate could never pass, and a pre-existing failure was indistinguishable from one the current change had just caused -- which cost a session real time today. mdl/roundtrip already skipped this same fixture with a remedy message (fixture.skipIfMissing); the two now agree, and reuse its wording. The skip is NOT unconditional. Under CI the test FAILS instead, because CI does check the submodule out, so an absence there is a workflow regression rather than a developer's clone. #808 shipped green because an integration test had only ever skipped, and this census is the only thing holding canon's list-marker registrations against Studio Pro-authored documents: a skip that could fire everywhere would retire that check silently. The units floor had to move too, or skipping would have tripped it instead of the lstat: it was one total of 1000 over both roots, and the submodule holds 900 against pedapp's 369. It is now a floor per root (800 and 300), so the committed fixture alone still proves a real corpus was read. Measured in all three states - submodule present: runs and passes (4.1s). - absent, local: skips that root with the `git submodule update --init testdata/testapp` remedy, still censuses pedapp's 369 units, passes. - absent, CI=true: fails, naming the root and that CI checks submodules out. Co-Authored-By: Claude Opus 5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82sY9fcHtMuZcMNRm2B
main switched docker.CheckFindings from parsing printed `mx check` text to the JSON report `mx check -j` writes (8f4bc30), so the report's check no longer built. It now: - asks for warnings and deprecations too (CheckKinds -w -d), as #2423 asked; - keeps each finding's module, document and element. The per-code "Where" column lists distinct documents instead of widget names. Measured on testdata/pedapp: most of its 1055 errors are in the Atlas modules' page templates; - reads --mx-check-results as that JSON file, and refuses printed text with a message naming the file it takes. Tests use the recorded real outputs in docker/testdata/mxcheck. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
A model can be valid, pass mx check and build, and still drop a table on the next deploy. The runtime keys the database on each element's GUID, not its name, and nothing showed what a release does to the data already stored. `mxcli report release -p app.mpr --against <git revision | .mpr>` reads both models' persistent entities, attributes and associations from their stored form. It pairs them by GUID, the way the runtime does, and lists four effects: - destroys: removed, no longer persistent, now calculated, or a new GUID under the same name (table created empty; column recreated with the model default, #1119). A re-mint is one destroy, not a destroy plus an add. An association whose end's table goes is not listed again; - verify: attribute type, shorter string, association type, storage or ends, generalization. What the runtime does with these is not measured, and the page says so; - migrates: renamed or moved under the same GUID, rows kept; - adds: new tables, columns and associations. Persistence follows the generalization chain, into System entities (meta.SystemEntityPersistable). External entities and calculated attributes have no storage and are left out. A git revision is exported with `git archive` (only the .mpr and mprcontents) into a temporary directory; the project may sit in a subdirectory of the repository. An intended drop is annotated release:<kind>:<name> with status accepted, and the verdict counts it apart. Docs: a project-reports section and a report-release skill; assess-quality links it. Closes #2424 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
feat(report): the scored report says whether the project builds
feat(verify): mxcli verify — one gate over check, lint and mx check, new findings only
fix(test): the list-marker census skips its optional submodule instead of failing
…otstrap and docs verify replaces the separate lint and docker check gates (it runs both against baselines). bootstrap-app records the baselines on the blank app. The default baseline directory is verify-baseline/ next to the project, not .mxcli/, which mxcli init git-ignores. Adds the mxcli verify docs page. Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82stfDEB1TGfJLVwUxc
feat(report): report release says what a deploy does to production data
feat(verify): make verify the default gate in generated CLAUDE.md, bootstrap and docs
Closes #2405. A flow that passes 'Module.SomeFlow' as text leaves no reference for any checker to follow, so renaming or moving that element passes Studio Pro, mxbuild and `mxcli check --references`, and the app fails at runtime. From the benchmark that raised the issue: a Java action received the text "Simulation.SIM_Tick", and renaming the microflow stopped the simulation with an ERROR in the server log every second. mxcli has a particular stake in it. It makes renaming cheap and scriptable (`rename microflow`, `rename entity`), and `check --references` is sold as the thing that catches a stale reference; a string literal is the one case where that promise silently does not hold. Two verdicts: a literal that RESOLVES to a real element is a warning, the rename trap; a literal whose module exists but whose name resolves to nothing is info, because a string may legitimately look like a qualified name. Two narrowings carry the precision, each the inverse of the obvious choice: - String LITERALS, not whole expressions. An expression's own references are dotted too -- an association path $Var/Mod.Assoc/Attr, an enumeration value Mod.Enum.Value -- and those ARE real references Studio Pro resolves and renames. Scanning expressions would report every association path in the project. - Only the fields that hold a value: Expression, Argument, Value, matched by NAME through reflection, so a new action type is covered the day it is added. Reflection over every string field is comprehensive and wrong, because JavaActionCallAction.JavaAction, MicroflowCall.Microflow and JavaActionParameterMapping.Parameter all hold qualified names and are all genuine references: that version flags every call in the project. Missing an expression field costs a finding; missing a reference field invents one. One further test removes nearly all remaining noise: the dotted prefix must be a module the project actually has. That excludes prose, file paths, URLs, CSS selectors and Java packages (com.mendix.core.Core) without enumerating any. Measured - 7 tests, including that a Java action call's reference fields are NOT scanned while its argument IS, member values inside loops, and the doubled-quote escape ('it''s here'). - Positive control, run before polishing: a planted literal naming a real microflow is reported; ordinary prose, a Java class name and a 4-segment name are not; and after `rename microflow` the warning becomes the dangling info. - The PREMISE measured, not asserted: mxbuild reported 28 errors before the rename and 28 after, naming the renamed microflow in neither. (`mxcli check <script> --references` is not that control: it checks a script against a project, so checking the rename script reports errors about the script.) - 0 findings on testdata/pedapp, testdata/testapp-views, and the 900-unit Studio Pro-authored testdata/testapp -- the last being the real sample, since ctx.Microflows() is platform-filtered and a Marketplace-heavy project mostly measures nothing. - Fast catalog mode suffices (no RequiredCatalogMode): the positive control fired on a freshly built catalog. Co-Authored-By: Claude Opus 5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82sY9fcHtMuZcMNRm2B
…, widgets, vulnerable jars Before production you want to know whether the app stands on current, safe dependencies. mxcli had pieces of this (marketplace versions and diff), but no report brought them together, and nothing looked at the Java libraries' known vulnerabilities. `mxcli report dependencies -p app.mpr` lists: - the Mendix version and its release line: LTS, MTS, monthly or unknown, from the version registry's new ReleaseLine. A major whose LTS/MTS lines the registry does not record is unknown, not monthly; - Marketplace modules, identified by the version UUID each records (AppStoreGuid), never by name or number. Each gets its newest and newest compatible version. A version the Marketplace no longer publishes is reported as withdrawn, naming the content that has the name; - pluggable widgets from widgets/*.mpk (package.xml), matched by exact name first, then a unique prefix; - jars in userlib/ and vendorlib/ with their Maven coordinates (from the jar's own pom.properties) and known vulnerabilities from OSV. Details are fetched in parallel, worst first, with the version to upgrade to. What could not be checked (no Marketplace token, --offline, an unreachable database, a jar without coordinates) is said per item and never shown as up to date or clean. High and critical advisories are blocking, unless annotated vuln:<id> with status accepted. Docs: a project-reports section and a report-dependencies skill; assess-quality links it. Closes #2425 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
feat(report): report dependencies — release line, Marketplace modules, widgets, vulnerable jars
feat(lint): MPR018 reports a model element named inside a string literal
…he accept hint On a project whose baseline predates most of its model, verify listed every new info finding (55 lines for one stale baseline), burying the one that matters. Text output now lists the findings that fail at --fail-on and counts the rest by severity, with an --update-baseline hint scoped to the stages that have them, since --update-baseline records every selected stage. JSON is unchanged. Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82stfDEB1TGfJLVwUxc
…nd work A step's server time, database time and queries were every span that started inside its window. Queued tasks, scheduled events and the task queue's polling therefore counted as the step's work. On ako/PizzaV1 the simulator's queued clock tick (a 1.5 s sleep) made "place order and pay" read 4,884 ms of server time; its own requests took 280 ms. Spans are now attributed by trace. A step owns the traces rooted in an HTTP request to the app that start inside its window, whole, also where a child outlasts the window. The app port is the most common server.port among HTTP roots, so the runtime's admin port is not counted. Other roots in the window are background: their union is reported per step (BackgroundMs) and listed on the page, not hidden. Spans without HTTP request attributes (a collector that does not record them) still use the window rule, and the page says so. On PizzaV1's recorded run, database time now matches the project's own analysis on all 12 steps, and queries are within 1 to 3. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
…t included ako/PizzaV1 rebuilt every mxcli report by hand in a "house style". Most of what made those pages read better is presentation that applies to every kind, so it is now the default: - a header line naming project, report and date, then the verdict as a lede (Doc.Lede; the kind's description when empty); - tiles with a status edge and an explaining sentence (Stat.Status); - inline bars in table cells (Table.Bars), stacked bars with a legend (Stacked) and finding cards (Card); - numeric columns right-aligned in a fixed-width face, with their heading aligned too; prose at a reading width; a footer naming the mxcli build; - colour tokens for light and dark, a fixed eight-slot series palette, and an explicit data-theme override; - links between report pages (Doc.Nav), for the overview to fill. The scored `mxcli report --format html` now renders through the same renderer (projectreport.QualityDoc). Its old HTML had a style of its own and score bars with a width but no fill rule, so every bar was an empty grey track. Markdown and JSON stay the linter's. The bar class is .meter, because the software map uses .bar for its control rows. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
…overview `mxcli report all -p app.mpr -o docs/reports` runs every kind, the scored report included, into one folder. Each kind writes <kind>.html and <kind>.json, the scored report as quality.html and quality.json. index.html is an overview with a card per report (its verdict and headline tiles), and every page links the others. - A <kind>-annotations.yaml in the folder is applied to that kind, the convention ako/PizzaV1 already uses. - --against, --journeys and --spans are passed to release, performance and tests. - A kind that cannot run is a card that says why. A <kind>.json an earlier run left in the folder is linked instead, so a performance report made separately shows up. - --offline asks no online source; --skip leaves kinds out. Each kind is recorded with its command as it is defined (reportKinds), so all runs it with its own flag defaults. The scored report's build step is extracted (buildScoredReport) for report and all to share. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
The scored report's HTML now says why the score is what it is. Every figure is computed from the findings; none is written by hand. - Categories show each one's weight, the points it costs the overall score, and its penalty against the floor. Past the floor, fixes do not move the score. The overall tile names the plain mean beside the weighted one. - A rule table lists every rule that found something, with how many of its findings brain decisions accept, the points fixing the rest would gain (scored again without them), and a verdict. A rule whose every finding is accepted still gets a row. `report --annotations` gives rule:<ID> verdicts. - Recommendations are the five rules whose fixes gain the most, rules with a verdict left out; the rules' own suggestions follow. - A tile names the most frequent rule, and says so when fixing all of it gains nothing. On GitPizza this reproduces the hand-written page's analysis: Quality's 0 costs 20.0 points and Security's 50 costs 12.5; Quality is 140 points past the floor; MPR006 × 94 (38% of findings) gains nothing. It adds that ARCH003 is worth +7.2. linter exports the penalty per severity, CategoryScore.Penalty and CategoryWeight, and Report gains AcceptedByRule. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
…lance view Every report's headline tiles now say whether their number is good, a warning or bad, so the overview and each page read at a glance: - security: open high and medium findings, flows to triage, level; - architecture: cycles, unused documents to verify; - tests and journeys: passed of run; - traceability: built, not built or not traceable, untested; - release: destroys and verify; - dependencies: the release line, modules behind, blocking vulnerabilities (warn when not checked). Traceability opens with "At a glance": a stacked bar per milestone and per epic (plan slice). Each shows built and proven, built but untested, partial or changed, and not built, as the hand-written GitPizza page did. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
…low argument A nanoflow passed to a Nanoflow-typed JavaScript action parameter was written as a basic expression value (CE0115/CE0117 at build), and describe left Nanoflows$NanoflowParameterValue mappings out, so an edit of a Studio Pro call dropped them. The argument is now written as NanoflowParameterValue (blank for empty), an expression is refused, and describe prints it. Closes #2450 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…ng entity-type argument A member's value was never compared with its attribute's type, call results were untyped in the flow scope, and a code action call was not checked for a missing argument, so the build's CE0117 and CE0115 were the first report. Members use the measured call-argument rule, call results are typed from the callee, and a missing entity-type argument (the only omission the build rejects) is reported. Closes #2464 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…and timeout The REST call statement was missing from checkStmtExprFunctions, so an unknown function in any of its expressions passed check and exec and failed the build with CE0117. Closes #2408 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
The layout header refused folder: as an unknown property after check had passed it. A layout is now filed in the named folder, a rewrite with another folder moves it, and describe prints the folder. Closes #2460 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82tETa46YQmMS58pQjL
…24rqgv Batch H: progress message, error-handler rejoin, JS nanoflow arguments, member/argument checks, REST MDL044, layout folder
… mappings An empty Security$ModuleSecurity.ModuleRoles list was written with marker 1 and a JavaScript action call's ParameterMappings with marker 3. Studio Pro writes 3 and 2 (8/8 and 25/25 in a Studio Pro-made project), so it rewrote those units on first open. Both are registered per owner/key as well, so canon.NormalizeListMarkers corrects units an older mxcli already stored on their next write. Refs #2467 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
…the mxcli version #2438 made verify the gate in every generated CLAUDE.md. The session-start bootstrap runs init --sync-skills every session and regenerates that section, so an upgraded mxcli moved every bootstrapped project onto verify without baselines, where it reports everything the project already had as new. The generator now gets the project directory (ToolFile.ContentIn) and publishes the verify list only when verify-baseline/lint.json exists; otherwise the classic lint + docker check list and its warning-count paragraph. Recording the baselines moves a project onto verify at the next sync, removing them moves it back, and an upgrade alone changes nothing. The bootstrap skill regenerates CLAUDE.md right after recording the baselines; the skill, docs and proposal describe both lists. Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82stfDEB1TGfJLVwUxc
feat(lint): MPR019 reports an XPath-filtered attribute with no leading index
catalog: resolve an attribute used through a document's own parameter (#2468)
verify: say the scripts were applied when a later stage fails (#2468)
…rioritization-24rqgv # Conflicts: # CHANGELOG.md
…24rqgv Write Studio Pro's list markers for ModuleSecurity.ModuleRoles and JS action call mappings
fix(init): the gate list follows the project's verify baselines, not the mxcli version
Two items left of #2443, after ako/PizzaV1's performance page: - each median shows the min–max of its warm runs, before and after: a difference inside the spread is run-to-run noise, not the fix; - findings: `finding:<ID>` (name, status, note) and `finding: <ID>` on each `step:<journey / step>` it targets. The page shows one card per finding with its steps' before → after, and names the finding on each step's row. Note gains a general `finding` field for this. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
The page had a table per milestone and the at-a-glance bars, but no
single place to scan every requirement. A compact table after the
glance lists each one with its milestone, epic, status and what proves
it ("2 tests, 1 journey" or "nothing"); the per-milestone sections keep
the full detail.
Refs #2443
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
Studio Pro writes $ID, $Type, then the rest of an element's keys sorted case-insensitively (no exception in 10.24, 11.12.6 and 11.14 projects; byte order is not it). mxcli wrote metamodel order, so the first Studio Pro open rewrote every unit mxcli had touched and the next mxcli edit flipped it back. The write path (canon.CompleteForStorage with listMarkers) now sorts. The comparison path does not, and elision is order-insensitive, so a stored unit is reordered only when it really changes; a delete+insert that changed nothing lands the stored bytes rather than sorted ones. Measured: the same script leaves 29 types out of order and 9 units that mx convert rewrites on main, and 0 and 0 with this change. Refs #2467 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
'63 pre-existing' reads the same whether they are infos or errors a baseline
accepted, so a baselined error or a pile of unread warnings disappeared into a count.
The summary now breaks it down worst first ('8 pre-existing (2 errors, 1 warning,
5 info)'): lint reports its already-known findings by severity in the baseline
object of its JSON, Classify counts them for mxcheck, and the JSON carries
preExistingBySeverity.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_011oK82stfDEB1TGfJLVwUxc
…rioritization-24rqgv
Performance: - finding cards show DB time as well as server time, and mark a change inside the run-to-run spread (or under the table's threshold) as such; - `scope: background` on a finding renders as background work, not as a finding no step names; - `accepted`, `intended` and `by design` statuses render as decisions; - a step's row names its finding by name, and the ID is no longer written twice when the note starts with it; - the JIT tile says "no plans read" instead of "0 of 0". Traceability: - the every-requirement table groups rows under milestone/epic header rows (Table.Groups) instead of repeating them in columns, so status and proof stay on screen at phone width; - a built requirement no microflow implements says "nothing to reach"; - stacked bars put the bar on its own line below 600 px. Refs #2443 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qTgUsRMETmzvjS27w
…24rqgv Write every unit's keys in Studio Pro's order
The template mx create-project ships can be a one-time conversion behind the Studio Pro of the same version: Studio Pro 11.16.1 appends UseMeToSetupNodeConversion and adds a Projects$SetupSettings unit that mx create-project and mx convert 11.16.1 never write. Copying that into the project would mean shadowing Studio Pro's template per version, so mxcli new and the quickstart say it instead: expect one conversion on first open, and commit it on its own. Closes #2467 Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_011oK82qrr1WnXxq1eRn7qRJ
feat(report): performance ranges and findings, one table of every requirement
feat(verify): say what the pre-existing findings are
…rioritization-24rqgv
…24rqgv mxcli new: say the first Studio Pro open may convert the project
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.