Improving Bean Shell Support in Playground#4780
Merged
shai-almog merged 116 commits intomasterfrom Apr 22, 2026
Merged
Conversation
shai-almog
changed the title
Contributor
Cloudflare Preview
|
Contributor
✅ Continuous Quality ReportTest & Coverage
Static Analysis
Generated automatically by the PR CI workflow. |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR modernizes the Codename One Playground BeanShell fork to support more Java-like syntax and class constructs, and adds CI automation to prevent regressions in Playground language support.
Changes:
- Replace the previous ASM/bytecode-generation approach for script-declared types with a namespace-backed
ScriptedClass/ScriptedInstanceruntime model and integrate it into name resolution, field access, method dispatch, and allocation. - Extend parser/runtime rewriting to handle additional Java syntax patterns (e.g.,
instanceofbindings, improved try-with-resources parsing, and additional lambda rewriting paths). - Add/expand smoke + syntax-matrix CI coverage and update tooling/docs to run these checks reliably in GitHub Actions.
Reviewed changes
Copilot reviewed 41 out of 41 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| scripts/cn1playground/tools/run-playground-smoke-tests.sh | Switches registry checks to grep, updates Maven build/run steps, and runs both smoke + syntax matrix harnesses. |
| scripts/cn1playground/tools/generate-cn1-access-registry.sh | Adds conditional tools.jar support to compile/run the registry generator across JDK setups. |
| scripts/cn1playground/common/src/test/java/com/codenameone/playground/PlaygroundSmokeHarness.java | Forces process exit after successful run to avoid CI hangs due to non-daemon threads. |
| scripts/cn1playground/common/src/main/java/com/codenameone/playground/PlaygroundRunner.java | Introduces a script “plan” that evals type declarations separately, plus new rewrite passes (inline AutoCloseable helpers, top-level lambdas). |
| scripts/cn1playground/common/src/main/java/bsh/cn1/CN1LambdaSupport.java | Makes LambdaValue implement Runnable to improve SAM assignment compatibility. |
| scripts/cn1playground/common/src/main/java/bsh/ScriptedInstance.java | Adds new runtime object representing instances of script-declared classes (namespace-backed fields + dispatch). |
| scripts/cn1playground/common/src/main/java/bsh/ScriptedClass.java | Adds new runtime descriptor for script-declared classes (methods/ctors/fields, enum constants, inheritance among scripted classes). |
| scripts/cn1playground/common/src/main/java/bsh/Parser.java | Adds instanceof pattern binding parsing and loosens try-with-resources grammar (incl. trailing semicolon). |
| scripts/cn1playground/common/src/main/java/bsh/Operators.java | Fixes shift operator semantics so int-sized shifts don’t get incorrectly promoted to long. |
| scripts/cn1playground/common/src/main/java/bsh/Name.java | Adds scripted-class field access and method dispatch paths (instance + static + enum built-ins). |
| scripts/cn1playground/common/src/main/java/bsh/BSHType.java | Allows typed declarations referencing scripted classes to resolve as Object (enables assignment of ScriptedInstance). |
| scripts/cn1playground/common/src/main/java/bsh/BSHPrimarySuffix.java | Implements scripted instance/static field access and scripted method/static method dispatch for primary suffixes. |
| scripts/cn1playground/common/src/main/java/bsh/BSHEnhancedForStatement.java | Adds explicit null-iteratee error for enhanced for-loops. |
| scripts/cn1playground/common/src/main/java/bsh/BSHClassDeclaration.java | Evaluates class declarations into ScriptedClass objects rather than throwing “unsupported” errors; adds basic interface/enum marking. |
| scripts/cn1playground/common/src/main/java/bsh/BSHBinaryExpression.java | Supports instanceof binding by capturing identifier at parse time and binding on successful match. |
| scripts/cn1playground/common/src/main/java/bsh/BSHAllocationExpression.java | Enables new against scripted classes (via namespace lookup), returning ScriptedInstance. |
| scripts/cn1playground/README.md | Updates test instructions and documents a language feature rollout process tied to the smoke/syntax harness. |
| .github/workflows/cn1playground-language.yml | Adds a dedicated workflow to run Playground language smoke tests (with Xvfb) on PRs and pushes. |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/TypeReference.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/TypePath.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Type.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Symbol.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Opcodes.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/MethodVisitor.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Label.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/LICENSE.txt | Removes vendored ASM license file (corresponding to ASM source removal). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Handler.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Handle.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/FieldWriter.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/FieldVisitor.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Edge.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/CurrentFrame.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Constants.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/ClassWriter.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/ClassVisitor.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/ByteVector.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
| scripts/cn1playground/common/src/main/java/bsh/org/objectweb/asm/Attribute.java | Removes vendored ASM source (part of eliminating ASM-based class generation). |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
|
||
| echo "Verifying Component is present in generated registry..." | ||
| if ! rg -q 'index.put\("com\.codename1\.ui\.Component"' "$ROOT/common/src/main/java/bsh/cn1/GeneratedCN1Access.java"; then | ||
| if ! grep -q 'index.put("com.codename1.ui.Component"' "$ROOT/common/src/main/java/bsh/cn1/GeneratedCN1Access.java"; then |
There was a problem hiding this comment.
grep patterns here are interpreted as regex, so the unescaped . characters in com.codename1... will match any character. This weakens the smoke test and could allow false positives. Use fixed-string matching (e.g., grep -Fq) or escape dots to ensure the exact class names are verified.
Comment on lines
+162
to
+166
| String importSection = normalized.substring(packageEnd, importEnd); | ||
| List<String> declarationScripts = new ArrayList<String>(); | ||
| StringBuilder remainingBody = new StringBuilder(); | ||
| int cursor = importEnd; | ||
| for (int i = 0; i < declarations.size(); i++) { |
There was a problem hiding this comment.
importSection is built from normalized.substring(packageEnd, importEnd), which drops any leading package ...; declaration. When you later eval each top-level type declaration separately, it won't include the package statement, which can change semantics for scripts that declare a package. Consider including the full prefix up to importEnd (package + imports) when building declarationScripts.
Comment on lines
+62
to
+64
| * <p>Interfaces and enums are still rejected — they require richer | ||
| * runtime support (SAM binding, singleton instances) that is layered | ||
| * on top of the basic scripted-class core. |
There was a problem hiding this comment.
This Javadoc says interfaces/enums are “still rejected”, but the implementation now builds a ScriptedClass, marks isInterface/isEnum, and (for enums) populates constants. Update the comment to reflect the current behavior (e.g., interfaces are declarable but not instantiable; enums have limited support).
Suggested change
| * <p>Interfaces and enums are still rejected — they require richer | |
| * runtime support (SAM binding, singleton instances) that is layered | |
| * on top of the basic scripted-class core. | |
| * <p>Interfaces and enums also use the {@link ScriptedClass} machinery. | |
| * Interfaces may be declared and can expose static and default methods, | |
| * but they are not directly instantiable. Enums have limited runtime | |
| * support: the declaration is represented as a scripted class marked as | |
| * an enum, and its constants are populated during evaluation. |
Updates the BeanShell-Interpreter-Tradeoffs section to reflect: * sealed/permits is now runtime-enforced, not just stripped * records support compact constructors * pattern-switch statements with type bindings are supported * Collection.stream() routes through CN1StreamBridge with the full filter/map/reduce surface (no Optional, BinaryOperator-keyed reduce) * "Did you mean" suggestions are emitted for missing static fields/methods so typos surface fixable hints Non-static inner classes remain a documented gap. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Switch the three lookup tables (CLASS_INDEX, METHOD_INDEX, FIELD_INDEX) to the initialization-on-demand holder idiom. Before, a single <clinit> on GeneratedCN1Access eagerly built all three — paying the METHOD_INDEX and FIELD_INDEX cost on the first class touch even though those tables are only used by the "did you mean" diagnostic path. After, the FIELD_INDEX and METHOD_INDEX classes stay cold until a suggestion is requested. findClass still forces the CLASS_INDEX holder, but the diagnostic-only indexes are deferred. Benchmark (cold JVM): first findClass ~120ms (same), first getFieldNames now ~6ms (previously amortised into findClass). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mark nested non-static class declarations as inner on their ScriptedClass (setEnclosingClass pointing at the outer). A ScriptedInstance now also carries an optional enclosingInstance reference so inner instances can walk out to the outer's instance namespace. ScriptedClass.newInstance gains an overload that accepts an enclosing instance. The new instance's namespace is parented off the enclosing namespace rather than off the outer's static namespace, so outer instance fields resolve inside inner methods. BSHAllocationExpression walks the callstack at `new Inner()` to find a `this` whose ScriptedClass matches (or transitively encloses) the required outer. Constructing a non-static inner class when no enclosing `this` is in scope fails fast with a clear diagnostic. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…rors
A concrete (non-abstract) class that declares `implements Iface` for
a Java interface must now provide every method declared on that
interface. The check runs at class-declaration time and lists the
missing method names in the diagnostic. Catches the
`class Other implements ActionListener {}` gotcha where the user
meant to override actionPerformed but forgot.
Missing-method errors on scripted instances now surface a
"Did you mean: X?" hint drawn from the class's declared instance
methods, or list a few known method names when no close match is
found. Same Levenshtein-close / prefix match heuristics used by the
static-field suggestions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The earlier enforcement pass only covered Java interfaces — a
concrete class implementing a scripted interface with an abstract
method slipped through because the interface's methods are merged
into the class's instance-method table during build, so a naive
name-presence check saw them as "provided". Real concrete classes
should still have to supply a body.
MethodTemplate now exposes isAbstract() based on whether the
declaration has a blockNode. ScriptedClass gains a
getConcreteInstanceMethodNames() view that filters out bodyless
merged interface stubs, and getAbstractInstanceMethodNames() so
the check can enumerate what must be implemented.
BSHClassDeclaration.enforceInterfaceMethodImplementations now
walks both Java and scripted interfaces: missing names fail with
"class X is not abstract and does not implement all methods from
scripted interface 'Iface'. Missing: ...". Default methods still
satisfy the requirement; explicit empty bodies (void foo() {})
count as concrete.
Two new matrix cases: unimplemented scripted-iface method is
rejected; scripted-iface default method is inherited cleanly.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
BSHAllocationExpression.constructFromEnclosingInstance now handles scripted inner-class allocation: when the outer is a ScriptedInstance, look up the named inner class in its static namespace, verify it's non-static, verify the outer's type is assignable to the inner's required enclosing class, then newInstance(args, outer). Static nested classes and Java outer instances are rejected with clear diagnostics. Root findClass no longer builds a giant Map<String, Class<?>> eagerly. Instead it derives the package name from the lookup and dispatches to the matching per-package GeneratedAccess_* helper (which itself uses a lazy if-chain over class literals). Nested class lookups walk back through dotted prefixes until a known package matches. Only the package whose class is asked for gets its helper loaded. Cold-start impact (median of 3 fresh JVMs): before: first findClass 120 ms, first ui getStaticField 18 ms after: first findClass 27 ms, first ui getStaticField 8 ms PlaygroundColdStartHarness is checked in so regressions are visible — run it via `java -cp ... com.codenameone.playground. PlaygroundColdStartHarness` from a fresh JVM. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
liannacasper
force-pushed
the
feat/bsh-java-parity-matrix
branch
from
a84ece3 to
7b207c1
Compare
…start Reflect the features that actually ship on CN1's runtime surface: * Non-static inner classes (including outer.new Inner()). * Abstract-method enforcement at declaration time for both Java and scripted interfaces. * Instance-method "did you mean" hints in addition to the existing static-field ones. * PlaygroundColdStartHarness as a regression baseline. Also tightens the non-goals list: JDK surface that isn't in CN1's runtime (Optional, List.of, Map.of, Set.of, Stream.of, IntStream.range, extended Collectors, etc.) is explicitly out of scope — the playground mirrors CN1's actual API surface so scripts that compile here also run on device. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Parse errors now carry three things the old format lacked: * precise location pulled from the ParseException's currentToken (line + column, not just line) * an "unexpected <token>" tag so the user sees exactly what broke * the offending source line with a caret marker under the bad column — easier to eye-find than a line number in a long script Token-aware hints replace the generic "check for mismatched braces" line: mismatched parens, braces, commas, semicolons, and stray keywords like `else`/`catch`/`finally` each get a tailored hint pointing at the likely cause. Matrix grows by ~20 cases: * 4 parse_diagnostics cases pin the new formatter's output so regressions in location / hint / EOF-handling surface in CI. * extended/ category covers CN1-supported surface that wasn't previously exercised: String methods, StringBuilder chaining, Integer/Math APIs, Arrays.asList iteration, HashMap entrySet walk, Integer.toHexString, button-listener chaining, nested try/catch/finally with rethrow, enhanced-for over int[], ternary chains, bitwise ops, indexed charAt iteration, ArrayList sort with Comparator, three-level class hierarchy with super chain, enum with abstract per-constant method. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Enhanced-for over int[], long[], double[], float[], short[], byte[],
char[], boolean[] and String[] now works via dedicated
primitive-array iterators in CollectionManager. The previous
implementation only iterated Object[], so a bare
for (int x : new int[]{1,2,3}) ...
used to throw "Only Object[] arrays are supported".
BshArray.getIndex / setIndex / arrayLength pick up the same
dispatch: indexed reads return the boxed wrapper, indexed writes
unwrap Numbers/Characters/Booleans back into the primitive slot.
Callers that used to cast `(Object[]) obj` in BSHPrimarySuffix
(`.length` field access and subscript length-derivation) now go
through BshArray.arrayLength so primitive arrays report their
actual length.
CN1's reduced reflection surface forbids Float.TYPE / Short.TYPE /
Character.TYPE / Boolean.TYPE, so primitive-array reads now return
raw boxed wrappers (Integer.valueOf(x), Character.valueOf(c), ...)
instead of going through Primitive.wrap with a primitive-type
class literal.
Matrix grows by 26 cases:
* 7 primitive-array enhanced-for variants covering all eight
primitive types plus String[].
* extended/ category now covers CheckBox/RadioButton/Slider/
ComboBox, TextField/TextArea, layout variations (BorderLayout
regions, GridLayout, FlowLayout), style API, Calendar field
setters, Math.sqrt+floor, HashMap put/remove, generic holder
class, class field holding an int[], iterator.remove, nested
enhanced-for, recursive method, short-circuit boolean eval.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reports of the preview keeping pixel artifacts from the previous run's rendering even after the new component is added. The fix: call previewRoot.repaint() after revalidate() to force a full redraw of the preview region. revalidate() normally triggers a repaint, but on CN1 7.0.234 the combination of an embedded Form-as-component with a subsequent Container swap occasionally leaves the outgoing component's pixels on screen until the user triggers another paint (typing in the editor, scrolling). The explicit repaint() closes that gap. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This reverts commit 8c7748a.
Two related changes that address the refresh regression reported by users: * Side-menu sample + history buttons now close the side menu FIRST and then call setScript. Previously the order was reversed: setScript ran, then closeSideMenu started its slide-out animation while the 1ms UITimer scheduled by runScript was still pending. The animation preempted the scheduled work, so replacePreview only visibly landed when the user took another action (typing, scrolling) that forced another paint. * runScript now enqueues executeRunScript via CN.callSerially instead of UITimer.timer(1, ...). callSerially respects the EDT event queue — it runs after the current animation frame completes, so the preview swap happens AFTER the side-menu close animation settles, not during it. Together these resolve the symptom where selecting a sample from the side menu left the previous sample's rendering on screen until the user typed into the editor. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This reverts commit 78991d8.
The repeating UITimer(250ms, true, form, syncOpenSideMenuTheme) was stomping on CN1's side-menu close animation: mid-slide, the animation fades the panel's transparency down from 255, and every 250ms the timer would force bgTransparency back to 255. The panel never actually faded out — it remained rendered, opaque, and stuck on the form underneath. The stuck panel produced two visible symptoms: * "Two menus": opening the side menu showed a fresh slide-in panel on top of the stuck previous panel. * "Overlay on top of the new sample": after selecting a sample from the side menu, the stuck panel's pixels covered the preview area. Typing into the editor eventually triggered enough EDT work that the stuck panel got cleared by CN1's normal paint cycle, revealing the new sample underneath. Fix: only force the theme styling when Toolbar.isSideMenuShowing() returns true. While the menu is closed (or mid-close), the timer ticks become a no-op and CN1's own animation / paint cycle finishes the slide-out cleanly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This reverts commit a355dc7.
This is aimed at the JS-port-specific rendering bug where the previous sample's pixels remain visible as a non-interactive background after selecting a new sample (until the user types into the editor to force another paint cycle). On the JavaScript port, plain Container.revalidate() on a scrollable viewport can leave the previous component's pixels cached in the canvas offscreen buffer when the replacement has different bounds. forceRevalidate() is the CN1-documented "more powerful" variant that recursively re-lays out the full hierarchy before repainting. Chaining that with a parent revalidate walks the invalidation one level up the tree so the SplitPane / Tabs panel containing previewRoot also refreshes its backing surface. The simulator build isn't affected (its Swing paint path invalidates more aggressively), which is consistent with the user's observation that the bug is only visible in the JavaScript port build. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This reverts commit 2fc5b83.
Experiment to isolate a long-standing JS-port-only rendering bug: after a sample switch the previous sample's UI remains visible in the background as a non-clickable layer, and opening the side menu shows a second menu underneath the fresh slide-in. The simulator build (which uses the TextArea fallback, not a BrowserComponent) is unaffected, which points at the iframe peer. The flag was originally added so that showing a Dialog (which deinitialises the enclosing Form) wouldn't destroy the editor iframe while the Dialog was up. The side effect is that the HTML5Peer stays pinned in the DOM across any detach/reattach, which is suspected to produce orphaned iframes stacked over their supposed-to-be-current positions when CN1's JS-port peer moves the iframe around. Removing the flag restores CN1's default behaviour: iframes go away on deinit and come back when the component is reattached. Users can verify the fix by flipping to a sample that shows a Dialog and checking whether the expected editor-reappears-after- close behaviour still works. If that regresses and the original symptom is gone, we have a pinpointed test case: the pin-on- deinit flag is the root cause of the doubling, and we can replace it with a more targeted guard. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This reverts commit 2092bda.
Adds three commands to the toolbar's right bar, one for each of the first three samples (Welcome, build(ctx), Lifecycle Demo). Each button calls exactly the same setScript(sample.script, true) path the matching side-menu entry uses — no side-menu open/close animation, no toolbar.closeSideMenu() call. Purely a diagnostic so we can test whether the "previous demo visible as background" symptom reproduces without any side-menu involvement. Verdict on this run will pinpoint whether the bug is inherent to the script-swap path itself or specific to the side-menu open/close cycle. To be reverted (or made conditional) once the cause is identified. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Suspected root cause of the JS-port "previous demo visible as background" + "two side menus on open" symptoms. Verified: a title-bar button that calls setScript(...) without going through the side menu doesn't reproduce the bug; side-menu buttons do. The 250ms UITimer walked the current form's tree every tick force-setting bgTransparency=255 / bgColor on any component with a SideNavigation* UIID. On the JS port, running this repeatedly while CN1 was in the middle of close-side-menu animation / DOM teardown appears to have accumulated stale side-menu DOM elements (the "replicated" view the user reported). Simulator (Swing) paint path rebuilds the frame on every repaint, so the stale DOM doesn't manifest there. Also drops the now-orphaned syncOpenSideMenuTheme and applySideMenuContainerTheme helpers. If side-menu dark styling regresses visually, the right fix is to apply it via CN1's theme props (addThemeProps in applySideMenuPalette) rather than polling. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Following the diagnostic that confirmed the side-menu path is the trigger for the JS-port "previous demo visible as background" + "two side menus" symptoms, pull out the remaining overrides so CN1's own side-menu theming can run unimpeded: * Delete applySideMenuPalette + its two callers. The aggressive addThemeProps of SideNavigationPanel / RightSideNavigationPanel / StatusBarSideMenu / SideCommand bg/border props was layered on top of CN1's own theme on every applyCurrentCss / applyDarkMode. * Remove the applyWebsiteTheme walk on every side-menu component and the sideMenuComponents tracking list used to re-theme them on dark-mode toggle. * Drop the custom PlaygroundSideCommand / PlaygroundSideCommandLine* / PlaygroundMenuEmpty UIID assignments on side-menu children so CN1's theming picks up the default SideCommand UIID chain. Visual appearance of the side menu in dark mode will regress (probably translucent / default CN1 side-menu look) but the correctness regression we were chasing takes priority. Restoring the dark-mode appearance without re-introducing the orphan-DOM bug is a separate follow-up. Matrix still 325/325; UI paint 312/312. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The diagnostic confirmed the bug is specific to the side-menu path, and persists after removing all the dark-mode theming hacks. That leaves the click-plumbing itself as the difference: the title-bar buttons that don't reproduce the bug use CN1 Commands, while the side-menu entries were hand-rolled Buttons added via addComponentToSideMenu + a manual closeSideMenu in the action listener. Swap sample + share entries to toolbar.addCommandToSideMenu(Command) so CN1 owns the tap -> close-animation -> actionPerformed sequence. That's the same machinery the working title-bar buttons use. If the ghost-preview / duplicate-side-menu symptoms clear up, it was the manual-close order interacting badly with CN1's own close handling on the JS port. History entries are still MultiButton-based (they have two lines) and use the old path — the sample path alone should be enough to confirm whether the switch fixes the bug. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This reverts commit 06e5424.
This reverts commit a09321e.
This reverts commit 5818f2a.
This reverts commit 7e62f6b.
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.