Back to Super Productivity

Android edge-to-edge + soft keyboard (IME)

docs/android-edge-to-edge-keyboard.md

18.13.120.6 KB
Original Source

Android edge-to-edge + soft keyboard (IME)

How the global add-task bar is positioned over the keyboard, and the full #8508 saga. Read this before touching anything keyboard/IME-related on Android — this area has regressed repeatedly (#8295, then #8508).

Update (2026-06-22): migrated off @capawesome/...edge-to-edge-support to Capacitor's built-in SystemBars (insetsHandling: 'css'). Edge-to-edge insets + IME padding are now handled by SystemBars on WebView ≥ 140 (or API ≥ 35); the WebView < 140 / API < 35 tail is covered by env() + a native keyboard shim (adjustWebViewHeightForKeyboardBelowApi30, now gated to WebView < 140 so it never fights SystemBars). Bar backgrounds are no longer painted by a plugin (SystemBars has no color API) — the bars are transparent and the theme color shows through via NavigationBarPlugin.setWebViewBackgroundColor (window decor + WebView surface). The #8508 sections below describe the former @capawesome mechanics and are kept as history. Full rationale + device matrix: docs/plans/2026-06-22-android-systembars-migration-corrected.md.

⚠️ Do NOT inset the WebView for the IME based on an assumption that the system "doesn't resize on Android 15/16." Real devices (incl. a Pixel-class Android 16 phone) still resize the window for the keyboard. Insetting on top of that double-counts and squashes the WebView. See #8508 below. Any future inset must detect whether the window already resized.

How the bar is positioned

The global add-task bar is position: fixed and lifted off the bottom by a CSS variable only:

scss
// add-task-bar.component.scss
:host-context(.isTouchOnly).global {
  bottom: calc(var(--keyboard-height) + var(--s2));
}

--keyboard-height defaults to 0px. On Android/web it is set by GlobalThemeService._initVisualViewportKeyboardTracking() (src/app/core/theme/global-theme.service.ts) from obscured = window.innerHeight - visualViewport.height, with a 100px floor (KEYBOARD_THRESHOLD_PXobscured <= 100 is treated as 0). On iOS the Capacitor Keyboard plugin sets it.

So the bar floats above the keyboard if either the window/WebView shrinks (then bottom: 0 is already above the IME) or the visual viewport shrinks (then --keyboard-height lifts the bar). On the devices we have tested, the window does shrink (the system resizes for the IME), so --keyboard-height stays 0 and the bar sits correctly at bottom: var(--s2).

#8508 — reversed / invisible characters (the actual root cause)

Symptom. On Android, the add-task bar (and search) showed reversed or invisible characters; some users reported "I can't see what I'm writing and Enter does nothing." Reported on v18.11.0 only (Pixel 10/Android 17, Galaxy S23 Ultra, Pixel 8a, Tab S5e/Android 15).

Root cause. v18.11.0 shipped a patch-package patch (commit 5497212b9) to @capawesome/capacitor-android-edge-to-edge-support that always inset the WebView by the IME height (bottomMargin = max(imeInsets.bottom, …) on every OnApplyWindowInsetsListener callback), to fix the bar sitting behind the keyboard under assumed enforced edge-to-edge.

On real devices the assumption is false: the system still resizes the window for the IME even on Android 16. Measured on an Android 16 phone with the keyboard up: window.innerHeight went 732 → 141 (and --keyboard-height stayed 0). The patch then added another ~909px inset on top of the already shrunk window → the WebView was squashed to a ~141px sliver with a huge blank gap above the keyboard. That squashed layout is almost certainly the "can't see what I'm writing" report.

Fix (this change). The patch was removed entirely. The plugin's stock behavior — bottomMargin = keyboardVisible ? 0 : max(imeInsets.bottom, …), i.e. no inset while the keyboard is up — lets the system handle the keyboard. Verified on an Android 16 phone: gap gone, WebView fills the resized window, bar sits just above the keyboard (no behind-keyboard regression).

Theories that were RULED OUT (don't re-chase)

  • "Angular ngModel writeValue resets the caret during composition." REFUTED. NgModel's isPropertyUpdated guard skips writeValue while the model equals the just-typed value, and the add-task bar never touches value/setSelectionRange/focus mid-composition. Proven with an e2e CDP IME probe (since removed) and the unit specs.
  • "Per-keystroke DOM churn (signal updates) during composition." Not the cause. On-device logging showed the WebView does not relayout during steady typing.
  • An SDK-version gate (inset only on API 36+) and an inset "latch." Both tried and reverted. The gate is wrong because the Android 16 phone resizes (so it still double-counted there); the latch held a stale keyboard height and produced its own gap.

Open items — if this is NOT fixed for the reporters

  1. Confirm the "reversed characters" symptom on the reporters' devices. The squashed-WebView / gap is verified fixed on the maintainer's Android 16 phone. It is not yet confirmed that the reversal is gone for all reporters (Pixel 10/A17, S23, Pixel 8a, Tab S5e). Ask them to test the next build.
  2. Residual: the system itself resizes on suggestion-strip changes. Even with the patch gone, the logs show the IME inset oscillating (imeBottom 909↔996) as the suggestion strip toggles — and the system resizes the window each time. Typing during that system resize could still disrupt composition. This is Android's own adjustResize, not our code. If reports persist, this is the next lead (e.g. a content-stable layout, or debouncing).
  3. The other v18.11.0 change. If the reversal persists with the patch gone, re-examine the @angular/* 21.2.11 → 21.2.17 bump (commit f51954f80) — the only other IME-adjacent change in the release.
  4. Long-term proper fix. Removing the patch only puts the bar behind the keyboard on a device that enforces edge-to-edge and whose visual viewport also fails to shrink for the IME — otherwise --keyboard-height still lifts the bar. That is cosmetic and likely rare, vs. the squashed layout on every real device tested. The correct inset would be resize-detecting: only inset when the window did not already shrink for the IME. Web-side detection already exists — GlobalThemeService._isVisualViewportResizedForKeyboard() — so a future native inset can reuse that logic rather than re-derive it. Validate on the device matrix below.
  5. Re-enable diagnostics. Add android.util.Log.d("SP8508", …) in EdgeToEdge.applyInsetsInternal logging kbVisible / imeBottom / bottomMargin / whether a relayout fired, then adb -d logcat -s SP8508. On the web side, chrome://inspect{innerH: innerHeight, vvH: visualViewport.height, kb: getComputedStyle(document.documentElement).getPropertyValue('--keyboard-height')}.

#8508 follow-up — SDK 28 (Android 9): add-task bar sits BEHIND the keyboard

Status: fix implemented (CapacitorMainActivity.adjustWebViewForKeyboardBelowApi30), PENDING ON-DEVICE VALIDATION across the matrix below. After 18.12.0 (patch removed) a user on Android 9 / API 28 reports the global add-task bar sits below / behind the soft keyboard. This is the realization of open item #4 above, and the device class it predicted.

Why API 28 specifically. The bar is positioned only from --keyboard-height, which GlobalThemeService._initVisualViewportKeyboardTracking() derives from obscured = window.innerHeight - visualViewport.height. It is correct iff either the window resized for the IME or the VisualViewport shrank. On API 28 neither does:

  1. targetSdk 36 + the @capawesome edge-to-edge plugin call setDecorFitsSystemWindows(window, false) on all API levels → the window goes edge-to-edge → the system stops resizing for the IME.
  2. The plugin does detect the IME on this device (WindowInsetsCompat.Type.ime() reports visible) and sets WebView bottomMargin = 0 while the keyboard is up — EdgeToEdge.applyInsetsInternal: "the system already resizes the window for the keyboard". But it does not resize (point 1), so the WebView keeps its full height and the bar stays put. (An on-device logcat confirmed keyboardVisible == true here; the earlier guess that Type.ime() is simply unreliable < 30 was wrong for this device.)
  3. The WebView's VisualViewport doesn't shrink either → obscured ≈ 0--keyboard-height = 0 → the position: fixed bar sits behind the keyboard.

Do NOT "fix" this on the web side. It is tempting to feed --keyboard-height from a native height fallback (the activity already measures the IME on every layout pass — CapacitorMainActivity OnGlobalLayoutListener: keypadHeight = screenHeight - rect.bottom, reliable on every API level). The trap: obscured is ≈0 in both the working case (window resized 732→141) and this broken case (nothing resized), so the web side cannot tell them apart without tracking a baseline innerHeight and computing max(obscured, nativeKbHeight - layoutShrink) — which is precisely the reverted #8295 formula in "What NOT to do" below. On a device that does resize, that double-counts and floats the bar mid-screen. The web layer lacks the signal to disambiguate; native has it unambiguously.

Implemented fix (native, explicit WebView height while the IME is up, scoped to API < 30) — CapacitorMainActivity.adjustWebViewHeightForKeyboardBelowApi30. Driven from the existing keyboard OnGlobalLayoutListener:

  • while the keyboard is up: set an explicit WebView layout height to the keyboard top, height = rect.bottom − webViewTopOnScreen (getWindowVisibleDisplayFrame, reliable on API 28). Shrinking the view shrinks the web layout viewport, so the existing CSS resolves the bar above the keyboard with no web-side keyboard-height math.
  • while the keyboard is down: restore the resting height (webViewLayoutHeightDefault, captured at startup, e.g. MATCH_PARENT), so the plugin's normal margin-based layout applies unchanged.
  • gated Build.VERSION.SDK_INT < 30, so on API >= 30 it is a strict no-op and the behavior verified in 18.12.0 is untouched.

Why height, not bottomMargin and not the plugin's listener. The plugin owns webView.bottomMargin and rewrites it to 0 on every inset dispatch while the IME is visible (EdgeToEdge.applyInsetsInternal, because it expects the system to resize — which enforced edge-to-edge prevents on API < 30). Correcting the margin from a second writer made the bar flicker constantly (on-device logcat showed the margin alternating 0 ↔ lift every frame); WebView bottom padding doesn't move the web layout viewport; and fully replacing the plugin's listener fixed the flicker but stopped the plugin re-sizing its status/nav color overlays, so the navbar showed a white gap. Setting an explicit layout_height is the way out: it is a different property than the margin the plugin manages, and for an explicit-height view the bottom margin does not change the view's size — so the two never fight, and the plugin keeps doing everything else (insets + color overlays, no white gap). The target is read from the visible frame and does not depend on the WebView's own height, so it is stable pass-to-pass (no feedback loop).

Upstream status (why a local workaround at all). This is a known, repeatedly regressed area in @capawesome/capacitor-android-edge-to-edge-support (pinned 8.0.8): see capawesome-team/capacitor-plugins #845/#490/#596/#725/#819 (closed) and #847 (open). The buggy keyboardVisible ? 0 : max(ime, navbar) ternary in EdgeToEdge.applyInsetsInternal is acknowledged — the maintainer redirects to Capacitor core ionic-team/capacitor#8466 (fixed for the built-in SystemBars by core PR #8481, merged), and plugin PR #848 ("correct WebView margin calculation") would fix the ternary but is still open/unreleased. So there is no shipped fix on the plugin path we use; this native workaround is independent of that timeline. Longer term, migrating to Capacitor 8's built-in SystemBars (insetsHandling) + dropping the plugin is the maintainer's implied direction.

Why not the web side: obscured cannot distinguish "window resized" from "nothing resized", so a web --keyboard-height fallback is the reverted #8295 formula. Native has the unambiguous geometry.

Still REQUIRED before release: validate across the device matrix below — this area has silently regressed at #8295 and twice at #8508. Confirm on a real API < 30 device that the bar lands flush on the keyboard top (no white gap, no flicker) and that the status/nav-bar layout is unchanged with the keyboard down, and on an API >= 30 device that nothing changed at all. A debug-only Log.d("SUPKeyboard", "webView height …") reports each height write — in steady state expect one per show/hide, not a stream. Remove that log before merge.

#8508 follow-up — fullscreen markdown / notes editor squashed

Status: CSS fix implemented, PENDING ON-DEVICE VALIDATION. Reported on #8508: editing a project (or task) note on Android with the keyboard up, the DialogFullscreenMarkdownComponent toolbar + textarea + Close/Save controls were squashed into the top of the screen with a large blank gap down to the keyboard.

Why. The bar is not the only position: fixed surface that must clear the keyboard — this dialog is position: fixed; height: 100% too. Its keyboard rule subtracted --keyboard-overlay-offset, which is set only on iOS, so on Android it was a no-op. With the keyboard up the dialog therefore kept whatever height 100% resolved to: full (content behind the keyboard) on a non-resizing device, or the squashed sliver on the buggy v18.11.0 WebView.

Fix (dialog-fullscreen-markdown.component.scss). Use the same resize-detecting --keyboard-height the add-task bar uses for the Android / mobile-web case; keep the iOS --keyboard-overlay-offset path in a separate rule. iOS carries both isNativeMobile and isIOS (and sets --keyboard-height non-zero), so the Android rule excludes iOS with :not(.isIOS) — the two rules are mutually exclusive and order-independent (rather than relying on equal-specificity source order):

scss
:host-context(body.isNativeMobile:not(.isIOS).isKeyboardVisible) {
  height: calc(100% - var(--keyboard-height, 0px));
}
:host-context(body.isIOS.isKeyboardVisible) {
  height: calc(100% - var(--keyboard-overlay-offset, 0px) - var(--safe-area-top));
}

This is not the reverted-#8295 trap above: it reads the pure VisualViewport --keyboard-height, never augments it with native data. Coverage across the device classes this doc tracks:

  • API < 30 — the SDK 28 native fix shrinks the WebView layout height, so 100% is already above the keyboard and --keyboard-height == 0; the rule is 100% - 0. Works.
  • API >= 30, window resizes (verified 18.12.0) — --keyboard-height == 0, so 100% - 0. Works.
  • API >= 30, no resize but VisualViewport shrinks (open item #4) — --keyboard-height > 0 lifts the dialog above the keyboard, on par with the add-task bar.

Do NOT also subtract --safe-area-top here. An earlier version of this fix did (100% - --keyboard-height - --safe-area-top). That is a double-count: :host is border-box (global * { box-sizing: border-box }) and already has padding-top: var(--safe-area-top), so the top inset is inside height: 100%. Subtracting it again left a --safe-area-top-sized gap between the Close/Save controls and the keyboard. It was invisible while --safe-area-top was 0 on API < 30, then surfaced the moment the status-bar fix above made it non-zero (also latent on API >= 30, where env() already gave a non-zero --safe-area-top). The iOS rule keeps its - --safe-area-top term for now — its keyboard runtime differs (the WebView does not resize) and it is unverified on an iOS device; if an iOS bottom gap appears, drop the term there too.

#8508 follow-up — SDK 28 (Android 9): header draws BEHIND the status bar

Status: fix implemented (CapacitorMainActivity.pushStatusBarOverlapBelowApi30), PENDING ON-DEVICE VALIDATION. Separate from the keyboard — on API 28 the web header overlaps the status bar (no top gap), reported on #8508.

Root cause. Post the SystemBars migration, Android no longer writes --safe-area-inset-* from JS; --safe-area-top resolves via the SCSS fallback var(--safe-area-inset-top, env(safe-area-inset-top, 0px)) (_css-variables.scss). On API >= 35 SystemBars injects --safe-area-inset-top, and on WebView >= 140 the WebView's own env(safe-area-inset-top) is correct — but on the WebView < 140 tail under enforced edge-to-edge the WebView extends under the status bar while env(safe-area-inset-top) resolves to 0 (old WebViews map only display cutouts into safe-area insets, not the status bar). So --safe-area-top == 0 and content draws under the status bar (Android 9 / API 28).

Why not a pure web-side fallback. The web side cannot tell "WebView is edge-to-edge under the status bar" from "WebView is already inset below it" — env() is 0 in both, and adding the status-bar height blindly would double-count in the inset case. Native has the geometry.

Fix (native overlap → SCSS fallback) — pushStatusBarOverlapBelowApi30. From the existing keyboard OnGlobalLayoutListener, measure the overlap max(0, rect.top − webViewTopOnScreen)rect.top is the visible-frame top (= status-bar height, reliable on API 28; the same frame the keyboard path reads) and getLocationOnScreen is the WebView's top (0 edge-to-edge, == status-bar height once inset). Publish it (physical px → CSS px, deduped) as the --android-status-bar-overlap CSS var, gated SDK < 30 AND WebView < 140 (mirrors the keyboard shim, never fights SystemBars). The var is folded into the SCSS fallback (_css-variables.scss) — NOT written from JS, so it never races SystemBars on --safe-area-inset-*:

scss
--safe-area-top: var(
  --safe-area-inset-top,
  max(env(safe-area-inset-top, 0px), var(--android-status-bar-overlap, 0px))
);
  • max(), not a sum, so it never double-counts: WebView < 140 edge-to-edge → env 0, overlap = status bar → status bar; once inset → env 0, overlap 0 → 0.
  • On API >= 35 / WebView >= 140 --safe-area-inset-top is set (SystemBars) or env() is correct, so var() precedence / max() ignore the overlap entirely — verified behavior untouched.
  • JS readers (_patchCdkViewportForSafeArea) still parse the var(max(...)) token to 0, so overlay positioning is unchanged — preserving #8283 scoping (only the header padding is affected).
  • Known small gap: an API 30–34 device on an old WebView < 140 also has env()==0 but is excluded by the SDK < 30 gate; rare (WebView auto-updates above API 30) — broaden the gate to WebView-only if it ever surfaces.
  • The var lives only as an inline style on the document, so a web-side reload (window.location.reload() — language change, PWA update, sync-conflict recovery) wipes it. The native dedupe (lastStatusBarOverlapCssPx) is reset in flushPendingShareIntent() (runs on every frontend (re)load) so the next layout pass re-publishes it; without the reset the unchanged value would be skipped and the overlap would regress after a reload.

What NOT to do

Do not stack a second/third keyboard-height source on top of the VisualViewport signal (native physical-px height + a baseInnerHeight-tracking path combined as max(obscured, nativeKeyboardHeight - layoutShrink)). That was #8295; the sources race on separate async events, the baseline gets reset to the shrunk innerHeight mid-animation, the double-count guard collapses, and the bar is mispositioned. It was reverted. Fix the inset at the source, and only after detecting whether the system already resized.

Device test matrix (required before merging IME changes)

Behavior differs across devices — test the add-task bar opening the keyboard, and typing a word fast right after tapping +, on:

  • Android 10 (API 29) — pre-edge-to-edge; Type.ime() insets are unreliable here
  • Android 14 (API 34) — edge-to-edge opt-out still possible
  • Android 15 (API 35) — we opt out via windowOptOutEdgeToEdgeEnforcement
  • Android 16 (API 36) — our target; the system was observed to still resize for the IME on a real device

Both gesture-nav and 3-button-nav, light and dark. Confirm: no blank gap above the keyboard, bar visible just above the keyboard, and typed characters appear in order (not reversed).