chore(ffmpeg): vendor windows-x86_64 prebuilt tree + strip from-source CI

Captured from CI run 26067455150 build-windows job (commit f31bb80 —
the one-shot upload-artifact step). Together with the already-bundled
macos-arm64 + macos-x86_64 trees, all three macOS+Windows ffmpeg
trees are now in the repo. Linux remains the only platform building
from source on CI (next iteration after the JUCE_LINUX camera-code
conditional lands).

Bundled tree contents (140 files, 7.3 MB):
- include/{libavcodec,libavformat,libavutil,libswscale,wels}/ — headers
- lib/ — ffmpeg 7.1.2 + openh264 2.1.1 Windows DLLs alongside the MSVC
  .lib import libraries that scripts/build_ffmpeg_lgpl.sh's post-process
  generates via gendef + dlltool. Consumers (cmake/ffmpeg.cmake WIN32
  path) link against .lib at build time and find the .dll at runtime
  via JamWide.exe-adjacent path (the workflow's Stage artifacts step
  copies the DLLs next to the executable).

.gitignore: vendored .lib files were excluded by the existing `*.lib`
general rule at line 41 (originally meant for build-output static
libraries that shouldn't be committed). Added a `!libs/ffmpeg/**/*.lib`
exception AFTER the general rule (gitignore later-rule-wins semantics)
to whitelist the vendored MSVC import libraries.

CI workflow (Windows job) simplification:
- Stripped the "Vendor LGPL ffmpeg" step (15-25 min from-source ffmpeg
  + openh264 build via MSYS2 + autotools + gendef + dlltool). Now a
  no-op git checkout.
- Stripped the "Upload vendored windows-x86_64 ffmpeg tree" one-shot
  capture step — served its purpose, tree now bundled.
- Minimized MSYS2 install from `base-devel mingw-w64-x86_64-gcc
  mingw-w64-x86_64-pkg-config mingw-w64-x86_64-nasm
  mingw-w64-x86_64-binutils mingw-w64-x86_64-tools-git autoconf
  automake make git` (full build toolchain) down to `base-devel`
  (just bash + strings for the LGPL discipline gate).
- KEPT the Verify LGPL discipline step — iterates the bundled tree to
  assert no libx264 contamination on every push.

Net Windows CI time: ~25-30 min → ~6-8 min per run (saves ~20 min,
matching the mac speedup from the earlier from-source removal). And
the entire MSYS2 toolchain-rot surface (which took 3 separate fix
commits — gendef PATH, .lib generation, pkgconfig) goes away.

Net repo size: +7.3 MB. Negligible compared to JUCE/clap/wdl submodules.

scripts/build_ffmpeg_lgpl.sh stays in the repo unchanged — useful for
bumping ffmpeg or openh264 version (re-run, commit refreshed tree) or
reproducibility verification.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: mkschulze

.github/workflows/juce-build.yml

@@ -184,52 +184,36 @@ jobs:
       - name: Setup CMake
         uses: lukka/get-cmake@latest
 
-      - name: Setup MSYS2 (for ffmpeg ./configure + make)
-        # Phase 14.3-01: ffmpeg's autotools build needs a unix-like environment
-        # on Windows. JamWide itself still builds via MSVC; only the vendored
-        # ffmpeg+openh264 step uses MinGW64-gcc to produce DLLs (DLL ABI is
-        # interop-stable with MSVC consumers).
+      - name: Setup MSYS2 (for LGPL verify only)
+        # The full from-source ffmpeg build packages (gcc, nasm, binutils,
+        # tools-git, autoconf, automake, make) were dropped once
+        # libs/ffmpeg/windows-x86_64/ landed bundled in the repo. The verify
+        # step still wants bash + strings — base-devel covers both.
+        # If you need to regenerate the vendored Windows tree (ffmpeg or
+        # openh264 version bump), restore the full install list and re-add
+        # the Vendor LGPL ffmpeg step temporarily.
         uses: msys2/setup-msys2@v2
         with:
           msystem: MINGW64
           update: true
           install: >-
             base-devel
-            mingw-w64-x86_64-gcc
-            mingw-w64-x86_64-pkg-config
-            mingw-w64-x86_64-nasm
-            mingw-w64-x86_64-binutils
-            mingw-w64-x86_64-tools-git
-            autoconf
-            automake
-            make
-            git
 
-      - name: Vendor LGPL ffmpeg
-        # Phase 14.3-01: produce libs/ffmpeg/windows-x86_64/ on the CI runner.
-        # Invoked under MSYS2 bash so uname returns MINGW64_NT-* and the script
-        # picks the windows-x86_64 leg (Cisco openh264 win64 prebuilt + ffmpeg
-        # LGPL from source). All inputs are static; no untrusted github event
-        # data flows into shell.
-        shell: msys2 {0}
-        run: bash scripts/build_ffmpeg_lgpl.sh
-
-      - name: Verify LGPL discipline
+      - name: Verify LGPL discipline (vendored trees in libs/ffmpeg/)
+        # Iterates every populated libs/ffmpeg/<platform>/ tree (Windows
+        # vendored as of this commit) and asserts no libx264 strings + clean
+        # PE-import dependency check. Same gate that the mac CI runs against
+        # the bundled trees.
+        # The "Vendor LGPL ffmpeg" rebuild-from-source step that used to run
+        # here was removed once libs/ffmpeg/windows-x86_64/ landed bundled
+        # in the repo — saves 15-25 min/CI run and eliminates the gendef/
+        # dlltool/lib generation toolchain-rot surface that took multiple
+        # iterations to get green. Run scripts/build_ffmpeg_lgpl.sh manually
+        # under MSYS2 when bumping ffmpeg or openh264 version, then commit
+        # the refreshed tree.
         shell: msys2 {0}
         run: bash scripts/verify_ffmpeg_lgpl.sh
 
-      - name: Upload vendored windows-x86_64 ffmpeg tree (one-shot for bundling)
-        # Mirror of the mac-arm64 one-shot capture. Once libs/ffmpeg/windows-x86_64/
-        # lands in the repo, the MSYS2 setup + Vendor LGPL ffmpeg + Verify
-        # steps can be deleted from the Windows job entirely (saves 15-25 min
-        # per CI run and eliminates the gendef/dlltool/lib generation
-        # toolchain-rot surface that took multiple iterations to get green).
-        uses: actions/upload-artifact@v4
-        with:
-          name: libs-ffmpeg-windows-x86_64
-          path: libs/ffmpeg/windows-x86_64/
-          retention-days: 7
-
       - name: Configure CMake
         run: |
           cmake -B build `

.gitignore

@@ -28,6 +28,8 @@
 !libs/ffmpeg/**/*.dylib
 !libs/ffmpeg/**/*.so
 !libs/ffmpeg/**/*.dll
+# .lib exception lives further down — after the `*.lib` general exclude
+# at line ~41 (otherwise the later general rule overrides this one).
 
 # Fortran module files
 *.mod
@@ -38,6 +40,10 @@
 *.la
 *.a
 *.lib
+# EXCEPTION: vendored Windows ffmpeg MSVC import libraries — generated by
+# scripts/build_ffmpeg_lgpl.sh post-process and bundled at libs/ffmpeg/
+# windows-x86_64/lib/. Must come AFTER the *.lib rule above to override it.
+!libs/ffmpeg/**/*.lib
 
 # Executables
 *.exe

libs/ffmpeg/windows-x86_64/include/libavcodec/ac3_parser.h

@@ -0,0 +1,36 @@
+/*
+ * AC-3 parser prototypes
+ * Copyright (c) 2003 Fabrice Bellard
+ * Copyright (c) 2003 Michael Niedermayer
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_AC3_PARSER_H
+#define AVCODEC_AC3_PARSER_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+/**
+ * Extract the bitstream ID and the frame size from AC-3 data.
+ */
+int av_ac3_parse_header(const uint8_t *buf, size_t size,
+                        uint8_t *bitstream_id, uint16_t *frame_size);
+
+
+#endif /* AVCODEC_AC3_PARSER_H */

libs/ffmpeg/windows-x86_64/include/libavcodec/adts_parser.h

@@ -0,0 +1,37 @@
+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_ADTS_PARSER_H
+#define AVCODEC_ADTS_PARSER_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+#define AV_AAC_ADTS_HEADER_SIZE 7
+
+/**
+ * Extract the number of samples and frames from AAC data.
+ * @param[in]  buf     pointer to AAC data buffer
+ * @param[out] samples Pointer to where number of samples is written
+ * @param[out] frames  Pointer to where number of frames is written
+ * @return Returns 0 on success, error code on failure.
+ */
+int av_adts_header_parse(const uint8_t *buf, uint32_t *samples,
+                         uint8_t *frames);
+
+#endif /* AVCODEC_ADTS_PARSER_H */