feat: answer mimetype without libmagic, and deprecate the option

.github/workflows/apple.yml

@@ -200,8 +200,8 @@ jobs:
 
   # The iOS *device* slice is only ever link checked; nothing runs it. The
   # simulator run is the analogue of android's instrumented job and the only one
-  # that sees what a device sees — that `+load` fired, that `NSBundle` found
-  # `magic.mgc`, that a temp directory is writable inside an app container.
+  # that sees what a device sees — that `+load` fired, that `NSBundle` found the
+  # renderer assets, that a temp directory is writable inside an app container.
   test:
     needs: xcframework
     runs-on: macos-15

.github/workflows/build_test.yml

@@ -125,7 +125,6 @@ jobs:
           -DCMAKE_CXX_FLAGS="-Werror"
           -DCMAKE_INSTALL_PREFIX=install
           -DODR_TEST=ON
-          -DODR_WITH_LIBMAGIC=ON
           -DODR_BUNDLE_ASSETS=ON
           ${{ matrix.bindings && '-DODR_JNI=ON -DODR_PYTHON=ON' || '' }}
 
@@ -138,7 +137,6 @@ jobs:
           -DCMAKE_BUILD_TYPE=Release
           -DCMAKE_INSTALL_PREFIX=install
           -DODR_TEST=ON
-          -DODR_WITH_LIBMAGIC=OFF
 
       - name: build
         run: cmake --build build --config Release

.github/workflows/python.yml

@@ -109,9 +109,8 @@ jobs:
           restore-keys: |
             ccache-${{ env.CACHE_FLAVOR }}-${{ matrix.host_profile }}-${{ env.CCACHE_KEY_SUFFIX }}-
 
-      # Match the dependency set that pyproject.toml enables. libmagic stays on
-      # and `bundle_assets` bridges its database path into the build so
-      # `magic.mgc` lands in `pyodr/data`.
+      # Match the dependency set that pyproject.toml enables. `bundle_assets`
+      # is what stages the renderer's css/js so they land in `pyodr/data`.
       - name: conan install
         run: >
           conan install .

AGENTS.md

@@ -18,9 +18,10 @@ bytes ─▶ magic/open_strategy ─▶ DecodedFile ─▶ Document ─▶ Eleme
          DecoderEngine)                          format)    elements)          walks public API)
 ```
 
-1. **Detect** — `internal/magic.cpp` (+ `internal/libmagic`) sniffs the file;
+1. **Detect** — `internal/magic.cpp` sniffs the head of the file;
    `internal/open_strategy.cpp` picks a `FileType` + `DecoderEngine` and builds
-   the matching `abstract::DecodedFile`.
+   the matching `abstract::DecodedFile`. `odr::mimetype` composes the two, so a
+   zip is named by what is inside it.
 2. **Decode** — a document file yields an `abstract::Document`.
 3. **Element tree** — a `Document` exposes a root `ElementIdentifier` plus an
    `abstract::ElementAdapter`. Public value-semantics handles (`Element`, `Slide`,
@@ -88,9 +89,9 @@ cmake --build cmake-build-relwithdebinfo --target translate  # CLI: file → HTM
   when really necessary.
 - **Run the test binary from the build dir** so output stays out of the repo tree.
 - **For debugging, prefer the `translate` CLI** on a single file over the suite.
-- CMake options (`CMakeLists.txt`): `ODR_TEST`, `ODR_CLI`, `ODR_WITH_LIBMAGIC`,
-  `ODR_PYTHON`, `ODR_JNI`, `ODR_APPLE`, `ODR_CLANG_TIDY`. A new `.cpp` must be
-  added to `ODR_SOURCE_FILES`.
+- CMake options (`CMakeLists.txt`): `ODR_TEST`, `ODR_CLI`, `ODR_PYTHON`,
+  `ODR_JNI`, `ODR_APPLE`, `ODR_CLANG_TIDY`. A new `.cpp` must be added to
+  `ODR_SOURCE_FILES`.
 - **Test data is fetched, not vendored**, and opt in: `-DODR_TEST_FETCH_DATA=ON`
   makes `cmake/test_data.cmake` clone the repositories pinned in
   `test/data.cmake` into `test/data/` (they were submodules until then; read

CMakeLists.txt

@@ -16,7 +16,11 @@ option(ODR_TEST_FETCH_DATA "Fetch the test data pinned in test/data.cmake" OFF)
 option(ODR_CLI "enable command line interface" ON)
 option(ODR_WITH_HTTP_SERVER "Build the HTTP server (requires cpp-httplib)" ON)
 option(ODR_CLANG_TIDY "Run clang-tidy static analysis" OFF)
-option(ODR_WITH_LIBMAGIC "Build with libmagic" ON)
+# Removed, and accepted only so a consumer still passing it keeps configuring.
+# `odr::mimetype` is our own detection now: it reads inside zip and cfb
+# containers, which libmagic cannot, and it costs neither a dependency nor the
+# 8 MB `magic.mgc` database every consumer had to ship and point us at.
+option(ODR_WITH_LIBMAGIC "Removed, does nothing (deprecated)" OFF)
 option(ODR_BUNDLE_ASSETS "Bundle assets during build and install" OFF)
 option(ODR_PYTHON "Build Python bindings" OFF)
 option(ODR_JNI "Build JNI bindings" OFF)
@@ -274,30 +278,6 @@ set(ODR_BUILD_ODR_DATA_PATH "${CMAKE_CURRENT_BINARY_DIR}/data")
 file(COPY "${odr.js_SOURCE_DIR}/" DESTINATION "${ODR_BUILD_ODR_DATA_PATH}")
 set(ODR_INSTALL_ODR_DATA_PATH "${CMAKE_INSTALL_DATADIR}")
 
-# Validate an injected data-file path before it is used to bundle assets.
-#
-# These paths (LIBMAGIC_DATABASE_PATH, ...) are consumed but
-# never discovered by CMake — they must be injected from outside via `-D...=`;
-# our conanfile.py bridges them from the dependency runenv. A consumer that
-# builds odrcore without that bridge left the variable empty, and
-# `file(COPY "${empty}/" ...)` degraded to `file(COPY "/" ...)`, i.e. an attempt
-# to recursively copy the whole root filesystem (see issue #599). This guard
-# fails fast with an actionable message instead of copying "/".
-#
-#   path  value of the externally-injected path variable (may be empty)
-#   name  name of that variable, for the diagnostic
-function(odr_require_data_path path name)
-    if (NOT path)
-        message(FATAL_ERROR
-                "ODR_BUNDLE_ASSETS is ON but ${name} is not set. Provide it "
-                "explicitly (e.g. -D${name}=<path>) or build through the conan "
-                "recipe, or disable bundling with -DODR_BUNDLE_ASSETS=OFF.")
-    endif ()
-    if (NOT EXISTS "${path}")
-        message(FATAL_ERROR "${name} points to a path that does not exist: '${path}'.")
-    endif ()
-endfunction()
-
 if (ODR_WITH_HTTP_SERVER)
     find_package(httplib REQUIRED)
     target_sources(odr
@@ -314,33 +294,10 @@ if (ODR_WITH_HTTP_SERVER)
     )
 endif ()
 if (ODR_WITH_LIBMAGIC)
-    find_package(libmagic REQUIRED)
-    target_sources(odr
-            PRIVATE
-            "src/odr/internal/libmagic/libmagic.cpp"
-    )
-    target_link_libraries(odr
-            PRIVATE
-            libmagic::libmagic
-    )
-    target_compile_definitions(odr
-            PRIVATE
-            ODR_WITH_LIBMAGIC
-    )
-
-    if (ODR_BUNDLE_ASSETS)
-        odr_require_data_path("${LIBMAGIC_DATABASE_PATH}" LIBMAGIC_DATABASE_PATH)
-
-        set(ODR_BUILD_LIBMAGIC_DATABASE_PATH "${ODR_BUILD_ODR_DATA_PATH}/magic.mgc")
-
-        file(COPY_FILE "${LIBMAGIC_DATABASE_PATH}" "${ODR_BUILD_LIBMAGIC_DATABASE_PATH}")
-
-        set(ODR_INSTALL_LIBMAGIC_DATABASE_PATH "${ODR_INSTALL_ODR_DATA_PATH}/magic.mgc")
-    else ()
-        set(ODR_BUILD_LIBMAGIC_DATABASE_PATH "${LIBMAGIC_DATABASE_PATH}")
-
-        set(ODR_INSTALL_LIBMAGIC_DATABASE_PATH "${LIBMAGIC_DATABASE_PATH}")
-    endif ()
+    message(DEPRECATION
+            "ODR_WITH_LIBMAGIC no longer does anything: libmagic is gone and "
+            "`odr::mimetype` is our own detection. Drop the option; "
+            "`GlobalParams::libmagic_database_path` is inert too.")
 endif ()
 
 configure_file("src/odr/internal/project_info.cpp.in" "src/odr/internal/project_info.cpp")

android/AGENTS.md

@@ -66,6 +66,5 @@ calls it.
   finds the incompatibilities first.
 - The manifest stays permission-free; permissions the server needs are the
   app's call, and documented in `README.md` instead.
-- Assets keep the `core/odrcore` + `core/libmagic` layout of the droid conan
-  deployer, so an app can move between the two packagings without touching its
-  extraction code.
+- Assets keep the `core/odrcore` layout of the droid conan deployer, so an app
+  can move between the two packagings without touching its extraction code.

android/README.md

@@ -9,7 +9,6 @@ odr-core-android.aar
 ├── jni/<abi>/libodr_jni.so            the bindings with the core linked in
 ├── jni/<abi>/libc++_shared.so         the c++ runtime they were built against
 ├── assets/core/odrcore/*              css/js of the html renderer
-├── assets/core/libmagic/magic.mgc     libmagic database
 └── proguard.txt                       keeps the classes JNI resolves by name
 ```
 

android/build.gradle.kts

@@ -104,10 +104,10 @@ android {
         targetCompatibility = JavaVersion.VERSION_17
     }
 
-    // the instrumented apk carries the bindings and the 8 MB libmagic database,
-    // and pushing that onto a cold emulator outlasts ddmlib's default timeout,
-    // which surfaces as a ShellCommandUnresponsiveException rather than as a
-    // failing test
+    // the instrumented apk carries the bindings and the renderer assets, and
+    // pushing that onto a cold emulator outlasts ddmlib's default timeout, which
+    // surfaces as a ShellCommandUnresponsiveException rather than as a failing
+    // test
     installation { timeOutInMs = 10 * 60 * 1000 }
 
     sourceSets {

android/build_native.py

@@ -12,7 +12,6 @@
     prebuilt/jniLibs/<abi>/libodr_jni.so     the bindings, core linked in
     prebuilt/jniLibs/<abi>/libc++_shared.so  from the NDK, see below
     prebuilt/assets/core/odrcore/*           css/js of the html renderer
-    prebuilt/assets/core/libmagic/magic.mgc  libmagic database
 
 `libc++_shared.so` has to be shipped because the android profiles build against
 the shared c++ runtime and nothing else in a consuming app pulls it in; the
@@ -99,7 +98,6 @@ def build(architecture: str, conan: str, build_profile: str, output: Path) -> No
          "-DODR_JNI=ON",
          "-DODR_CLI=OFF",
          "-DODR_TEST=OFF",
-         "-DODR_WITH_LIBMAGIC=ON",
          "-DODR_WITH_HTTP_SERVER=ON",
          "-DODR_BUNDLE_ASSETS=ON"])
     run(["cmake", "--build", cmake_dir, "--target", "odr_jni",
@@ -117,9 +115,7 @@ def build(architecture: str, conan: str, build_profile: str, output: Path) -> No
     data = cmake_dir / "data"
     assets = output / "assets" / "core"
     shutil.rmtree(assets, ignore_errors=True)
-    shutil.copytree(data, assets / "odrcore", ignore=shutil.ignore_patterns("magic.mgc"))
-    (assets / "libmagic").mkdir(parents=True, exist_ok=True)
-    shutil.copy2(data / "magic.mgc", assets / "libmagic" / "magic.mgc")
+    shutil.copytree(data, assets / "odrcore")
 
 
 def main() -> int:

android/src/androidTest/java/app/opendocument/core/OdrAndroidTest.kt

@@ -33,9 +33,6 @@ class OdrAndroidTest {
         assertTrue("$data is not a directory", data.isDirectory)
         assertTrue(File(data, "document.css").isFile)
         assertTrue(File(data, "document.js").isFile)
-
-        val magic = File(GlobalParams.libmagicDatabasePath())
-        assertTrue("$magic is not a file", magic.isFile)
     }
 
     @Test
@@ -46,12 +43,12 @@ class OdrAndroidTest {
     }
 
     @Test
-    fun detectsTypeWithTheBundledMagicDatabase() {
+    fun detectsTypeInsideTheContainer() {
         val directory = TestSupport.tempDir("magic")
         val odt = TestFiles.odtFile(directory)
 
-        // goes through libmagic, so it only works when the database extracted above
-        // is the one the native library actually opened
+        // naming the odt rather than the zip holding it means detection opened the
+        // container
         assertEquals("application/vnd.oasis.opendocument.text", Odr.mimetype(odt.toString()))
     }
 }

android/src/main/java/app/opendocument/core/android/OdrAndroid.kt

@@ -9,8 +9,8 @@ import java.io.FileOutputStream
 import java.io.IOException
 
 /**
- * Makes the library usable on android: the renderer reads its CSS/JS and the libmagic database as
- * plain files, and an APK holds them as assets, which are not files.
+ * Makes the library usable on android: the renderer reads its CSS/JS as plain files, and an APK
+ * holds them as assets, which are not files.
  *
  * Call [init] once before anything else touches the library:
  * ```
@@ -22,7 +22,7 @@ object OdrAndroid {
     /** Asset directory this AAR ships its runtime data under. */
     private const val ASSETS = "core"
 
-    /** The libmagic database alone is 8 MB, so the default 8 KB would be a lot of syscalls. */
+    /** The default 8 KB would be a lot of syscalls for assets this size. */
     private const val BUFFER_SIZE = 64 * 1024
 
     private var initialized = false
@@ -56,7 +56,6 @@ object OdrAndroid {
         }
 
         GlobalParams.setOdrCoreDataPath(File(root, "odrcore").absolutePath)
-        GlobalParams.setLibmagicDatabasePath(File(root, "libmagic/magic.mgc").absolutePath)
         initialized = true
     }
 

apple/AGENTS.md

@@ -34,17 +34,17 @@ it — a simulator binary mistagged `IOS` is the classic
 | `macos-arm64_x86_64` | `apple-macos-armv8` + `apple-macos-x86_64` |
 
 `assemble` also fails the build if a framework is missing its headers, module
-map, `magic.mgc` or the plist's platform keys — the analogue of
+map, renderer resources or the plist's platform keys — the analogue of
 `android/build.gradle.kts`'s `checkNative`, and for the same reason: those all
 publish happily and then fail at the consumer.
 
 ## Why a dynamic framework
 
-`+load` in `src/OdrCoreBootstrap.mm` is what points odrcore at the css/js and
-`magic.mgc` in this bundle, so an app never has to. In a **static** framework
-nothing references that translation unit, the linker drops it, and the
-bootstrap never runs — and a SwiftPM binary target gives the consumer no way to
-pass `-ObjC`/`-force_load` to get it back. Two lesser reasons: `.binaryTarget`
+`+load` in `src/OdrCoreBootstrap.mm` is what points odrcore at the css/js in
+this bundle, so an app never has to. In a **static** framework nothing
+references that translation unit, the linker drops it, and the bootstrap never
+runs — and a SwiftPM binary target gives the consumer no way to pass
+`-ObjC`/`-force_load` to get it back. Two lesser reasons: `.binaryTarget`
 has no `resources:`, so the assets have to live in the bundle; and an undefined
 symbol becomes a link error here instead of a crash at the consumer.
 
@@ -130,8 +130,8 @@ which would leave the tag serving the previous version's binary.
 The iOS *device* slice is only ever link-checked — nothing runs it. The
 simulator suite is the analogue of android's instrumented job and the only
 place that sees what a device sees: that `+load` fired, that `NSBundle` found
-`magic.mgc`, that `temp_directory_path()` is writable inside an app container.
-A new binding is only covered once something in `tests/` calls it.
+the renderer assets, that `temp_directory_path()` is writable inside an app
+container. A new binding is only covered once something in `tests/` calls it.
 
 `tests/Fixtures/mixed-layout.odt` is 9 KB of `odt/` from OpenDocument.test,
 carried here because `test/data/` is fetched by `cmake/test_data.cmake` and a

apple/CMakeLists.txt

@@ -17,8 +17,8 @@ if (BUILD_SHARED_LIBS)
 endif ()
 if (NOT ODR_BUNDLE_ASSETS)
     message(FATAL_ERROR
-            "ODR_APPLE needs ODR_BUNDLE_ASSETS=ON: the renderer's css/js and "
-            "the libmagic database ship as bundle resources")
+            "ODR_APPLE needs ODR_BUNDLE_ASSETS=ON: the renderer's css/js ship "
+            "as bundle resources")
 endif ()
 
 enable_language(OBJCXX)
@@ -64,8 +64,7 @@ set(ODR_APPLE_PUBLIC_HEADERS
         "include/OdrCoreObjC/ODRTable.h"
 )
 
-# The renderer data the root build already staged, plus the libmagic database
-# `ODR_BUNDLE_ASSETS` copied in beside it.
+# The renderer's css/js, as the root build staged them.
 file(GLOB ODR_APPLE_RESOURCES "${ODR_BUILD_ODR_DATA_PATH}/*")
 if (NOT ODR_APPLE_RESOURCES)
     message(FATAL_ERROR "no resources in ${ODR_BUILD_ODR_DATA_PATH}")

apple/README.md

@@ -32,8 +32,8 @@ for view in service.views {
 }
 ```
 
-Nothing needs configuring first. The framework points odrcore at the css, JS and
-libmagic database it carries before `main` runs — that is what the `+load` in
+Nothing needs configuring first. The framework points odrcore at the css and JS
+it carries before `main` runs — that is what the `+load` in
 `OdrCoreBootstrap.mm` is for. Override it with `GlobalParams` from
 `application(_:didFinishLaunchingWithOptions:)` if you relocated the resources.
 

apple/build_xcframework.py

@@ -110,7 +110,6 @@ def build(profile: str, conan: str, build_profile: str) -> None:
          "-DODR_TEST=OFF",
          "-DODR_JNI=OFF",
          "-DODR_PYTHON=OFF",
-         "-DODR_WITH_LIBMAGIC=ON",
          "-DODR_WITH_HTTP_SERVER=ON",
          "-DODR_BUNDLE_ASSETS=ON"])
     run(["cmake", "--build", cmake_dir, "--target", "odr_apple",
@@ -159,7 +158,6 @@ def assert_contents(framework: Path) -> None:
     required = [
         root / "Headers" / f"{FRAMEWORK}.h",
         root / "Modules" / "module.modulemap",
-        resources / "magic.mgc",
         resources / "document.css",
     ]
     missing = [path for path in required if not path.exists()]

apple/exported_symbols.txt

@@ -1,5 +1,5 @@
 # The framework's entire public surface is its ObjC classes, so export those
-# and nothing else. Everything linked in — cryptopp, libmagic, all of `odr::` —
+# and nothing else. Everything linked in — cryptopp, pugixml, all of `odr::` —
 # becomes private-extern, which keeps an app that links its own copy of any of
 # them from binding against ours.
 #

apple/include/OdrCoreObjC/ODRFile.h

@@ -56,6 +56,23 @@ typedef NS_ENUM(NSInteger, ODRFileType) {
 
   ODRFileTypeTruetypeFont,
   ODRFileTypeOpentypeFont,
+
+  ODRFileTypeWebp,
+  ODRFileTypeTaggedImageFileFormat,
+  ODRFileTypeHighEfficiencyImageFormat,
+  ODRFileTypeAv1ImageFileFormat,
+
+  ODRFileTypeMpegAudio,
+  ODRFileTypeMpeg4Audio,
+  ODRFileTypeOggAudio,
+  ODRFileTypeWaveformAudio,
+  ODRFileTypeFreeLosslessAudioCodec,
+
+  ODRFileTypeMpeg4Video,
+  ODRFileTypeQuicktimeVideo,
+  ODRFileTypeThirdGenerationPartnershipVideo,
+  ODRFileTypeMatroskaVideo,
+  ODRFileTypeAudioVideoInterleave,
 } NS_SWIFT_NAME(FileType);
 
 typedef NS_ENUM(NSInteger, ODRFileCategory) {
@@ -65,6 +82,8 @@ typedef NS_ENUM(NSInteger, ODRFileCategory) {
   ODRFileCategoryArchive,
   ODRFileCategoryDocument,
   ODRFileCategoryFont,
+  ODRFileCategoryAudio,
+  ODRFileCategoryVideo,
 } NS_SWIFT_NAME(FileCategory);
 
 typedef NS_ENUM(NSInteger, ODRFileLocation) {