marfrit-packages/upstream-submissions/qt6-base-fourier/qt-bug-report.md

bugreports.qt.io issue draft

Project: QtBase Component: OpenGL Affects Version/s: 6.5, 6.6, 6.7, 6.8, 6.9, 6.10, 6.11, 6.12-dev Severity: Significant — affects every Qt 6 GLES3 client on Mali-class hardware Type: Bug

---

Summary

QOpenGLTextureGlyphCache, QRhiGles2::toGlTextureFormat (RED_OR_ALPHA8 case), and QOpenGLTextureUploader (Format_Alpha8 and Format_Grayscale8 cases) all hard-code GL_ALPHA (and GL_LUMINANCE) as the glTexImage2D internalFormat when qtbase is configured with QT_CONFIG(opengles2) (true on every aarch64 Linux distribution shipping Qt 6). GL_ALPHA was deprecated as a valid glTexImage2D internalFormat in OpenGL ES 3.0 (ES 3.0 spec, section 3.8.3, table 3.13 — only sized internal formats remain).

On Mali-class GPUs (mesa panfrost / panthor) under Wayland, KWin requests an OpenGL ES 3.2 context. Qt's text-glyph cache upload path then runs the #else branch with GL_ALPHA, mesa returns GL_INVALID_VALUE, the texture allocation never succeeds, and every subsequent glTexSubImage2D to that texture errors at level 0. KWin's glDebugMessageCallback fills the journal with:

kwin_wayland: 0x4: GL_INVALID_VALUE in glTexImage2D(internalFormat=GL_ALPHA)
kwin_wayland: 0x4: GL_INVALID_OPERATION in glTexSubImage2D(invalid texture level 0) × N

…on every text-glyph cache invalidation, every frame, indefinitely.

Reproduce

• Hardware: any Mali-G31 / G52 / G57 / G610 with mesa panfrost or panthor on Linux (RK3399 / RK3566 / RK3588 board, Pinebook Pro, PineTab2, CoolPi 4 etc.).
• Software stack: KDE Plasma 6.x Wayland + Qt 6.x built against this hardware (any Arch Linux ARM, Debian arm64, Ubuntu arm64 with Plasma).
• Trigger: open any Qt 6 application that renders text and watch journalctl --user -u plasma-kwin_wayland.service (or equivalent) for the glTexImage2D(internalFormat=GL_ALPHA) line. Examples that trigger reliably:
  • kate opening any text file
  • dolphin listing a directory
  • any system tray icon update
  • any window decoration title-bar redraw

Root cause

Three sites in qtbase 6.x:

1. src/opengl/qopengltextureglyphcache.cpp:111-117 — createTextureData() hard-codes internalFormat = GL_ALPHA in the QT_CONFIG(opengles2) branch.
2. src/opengl/qopengltextureglyphcache.cpp:213-217 — load_glyph_image_to_texture() matching path on glyph reupload.
3. src/gui/rhi/qrhigles2.cpp:1373-1378 — QRhiGles2::toGlTextureFormat RED_OR_ALPHA8 case is gated only on caps.coreProfile, missing the ES≥3 disjunct.
4. src/opengl/qopengltextureuploader.cpp:253-275 — Format_Alpha8 and Format_Grayscale8 short-circuit on context->isOpenGLES() before reaching the existing TextureSwizzle fallback.

Fix

Three patches (attached as a series), one per file. Each adds a runtime predicate (variations on caps.gles && caps.ctxMajor >= 3 or ctx->format().majorVersion() >= 3) so:

• desktop GL Core profile → GL_R8 + GL_RED (existing behavior)
• desktop GL Compat profile → GL_ALPHA (existing behavior)
• OpenGL ES 3.x → GL_R8 + GL_RED (new behavior; spec-clean)
• OpenGL ES 2.x → GL_ALPHA (preserved)

Memory layout, swizzle, and shader behavior identical to the existing core-profile path. No new dependencies. ~9 net lines added across three files.

Validation

Tested on PineTab2 (RK3566 / Mali-G52 / mainline kernel 6.19.10 / panfrost mesa 26.0.5 / KDE Plasma 6.6.4 Wayland / Qt 6.11.0 patched). Before the patches: constant journal stream of GL_INVALID_VALUE during normal desktop interaction. After the patches: zero GL_INVALID_VALUE in a fresh-session journal, including under heavy text-rendering load.

Patch series and end-to-end validation context maintained at https://github.com/marfrit/fourier under qt6-base-fourier/.

Patch series

(Three patches will be uploaded in Gerrit format as a single series with a common Change-Id thread.)

1. [PATCH 1/3] QOpenGLTextureGlyphCache: pick GL_R8 on OpenGL ES 3.x
2. [PATCH 2/3] QRhiGles2: use GL_R8 for RED_OR_ALPHA8 on OpenGL ES 3.x
3. [PATCH 3/3] QOpenGLTextureUploader: route ES 3.x through TextureSwizzle path

---

Gerrit submission notes

(Per Qt's contribution workflow — https://wiki.qt.io/Qt_Contribution_Guidelines)

Account setup (one-time)

1. Create account at https://account.qt.io
2. Sign Qt's CLA (electronic, takes 5 minutes)
3. Configure SSH key at https://codereview.qt-project.org
4. Install git-review: pacman -S git-review (Arch) or pip install git-review

Per-patch commit-message trailers

Each patch needs Qt-style trailers below the Signed-off-by:

Bug: QTBUG-XXXXX
Pick-to: 6.10 6.9 6.8 6.7 6.6 6.5
Change-Id: <generated by git-review>

Bug: is the issue number from bugreports.qt.io (the issue above, once filed). Pick-to: requests cherry-picks to LTS branches — this fix should backport cleanly to 6.5 (the oldest currently- supported Qt 6 LTS).

Change-Id: is auto-generated by git-review's commit-msg hook on first push.

Push to Gerrit

cd qtbase
# Apply our 3 patches with `git am`, then:
git review master       # pushes branch to refs/for/dev

Reviewers to add

• Eskil Abrahamsen Blomfeldt — primary glyph-cache reviewer
• Laszlo Agocs — RHI reviewer
• Volker Hilsheimer — QtGui maintainer