Enable all non-hardware features in CI and fix clippy warnings

.github/workflows/ci.yml

@@ -21,27 +21,7 @@ jobs:
           toolchain: "1.92.0"
           components: rustfmt
 
-      - run: cargo fmt --all -- --check
-
-  clippy:
-    name: Clippy
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          submodules: recursive
-
-      - name: Install system dependencies
-        run: sudo apt-get update && sudo apt-get install -y cmake libclang-dev
-
-      - uses: dtolnay/rust-toolchain@stable
-        with:
-          toolchain: "1.92.0"
-          components: clippy
-
-      - uses: Swatinem/rust-cache@v2
-
-      - run: cargo clippy --workspace -- -D warnings
+      - run: make fmt
 
   test:
     name: Tests
@@ -60,4 +40,4 @@ jobs:
 
       - uses: Swatinem/rust-cache@v2
 
-      - run: cargo test --workspace
+      - run: make test.unit

Makefile

@@ -1,5 +1,6 @@
-CARGO_TEST_LLM_FLAGS = --lib -p llama-cpp-bindings --features tests_that_use_llms,mtmd -- --test-threads=1
-CARGO_COV_LLM_FLAGS = --lib --features tests_that_use_llms,mtmd -p llama-cpp-bindings
+FEATURES = mtmd,sampler,llguidance
+CARGO_TEST_LLM_FLAGS = --lib -p llama-cpp-bindings --features tests_that_use_llms,$(FEATURES) -- --test-threads=1
+CARGO_COV_LLM_FLAGS = --lib --features tests_that_use_llms,$(FEATURES) -p llama-cpp-bindings
 
 QWEN3_5_0_8B_ENV = \
 	LLAMA_TEST_HF_REPO=unsloth/Qwen3.5-0.8B-GGUF \
@@ -12,7 +13,7 @@ QWEN3_5_0_8B_ENV = \
 
 .PHONY: test.unit
 test.unit: clippy
-	cargo test --lib -p llama-cpp-bindings
+	cargo test --lib -p llama-cpp-bindings --features $(FEATURES)
 
 .PHONY: test.qwen3.5_0.8B
 test.qwen3.5_0.8B: clippy
@@ -42,8 +43,8 @@ test: test.unit test.llms
 
 .PHONY: fmt
 fmt:
-	cargo fmt --check
+	cargo fmt --all --check
 
 .PHONY: clippy
 clippy:
-	cargo clippy --all-targets -p llama-cpp-bindings
+	cargo clippy --all-targets -p llama-cpp-bindings --features $(FEATURES) -- -D warnings

llama-cpp-bindings/src/llguidance_sampler.rs

@@ -69,7 +69,7 @@ fn build_tok_env(model: &LlamaModel) -> Arc<ApproximateTokEnv> {
 
 // --- extern "C" vtable callbacks ---
 
-unsafe extern "C" fn llg_name(
+const unsafe extern "C" fn llg_name(
     _smpl: *const llama_cpp_bindings_sys::llama_sampler,
 ) -> *const std::os::raw::c_char {
     c"llguidance".as_ptr()
@@ -168,6 +168,10 @@ static mut LLG_SAMPLER_I: llama_cpp_bindings_sys::llama_sampler_i =
     };
 
 /// Create an llguidance-based constrained decoding sampler.
+///
+/// # Errors
+///
+/// Returns `GrammarError` if the parser factory, grammar, or parser cannot be created.
 pub fn create_llg_sampler(
     model: &LlamaModel,
     grammar_kind: &str,

llama-cpp-bindings/src/mtmd/mtmd_bitmap.rs

@@ -283,6 +283,7 @@ mod tests {
 
     #[test]
     fn from_audio_data_creates_valid_bitmap() {
+        #[allow(clippy::cast_precision_loss)]
         let audio_samples: Vec<f32> = (0..100).map(|index| (index as f32 * 0.1).sin()).collect();
         let bitmap = MtmdBitmap::from_audio_data(&audio_samples).unwrap();
 

llama-cpp-bindings/src/mtmd/mtmd_context.rs

@@ -10,15 +10,15 @@ use super::mtmd_input_chunk::MtmdInputChunk;
 use super::mtmd_input_chunks::MtmdInputChunks;
 use super::mtmd_input_text::MtmdInputText;
 
-fn tokenize_result_to_error(result: i32) -> MtmdTokenizeError {
+const fn tokenize_result_to_error(result: i32) -> MtmdTokenizeError {
     match result {
         1 => MtmdTokenizeError::BitmapCountMismatch,
         2 => MtmdTokenizeError::ImagePreprocessingError,
         _ => MtmdTokenizeError::UnknownError(result),
     }
 }
 
-fn check_encode_result(result: i32) -> Result<(), MtmdEncodeError> {
+const fn check_encode_result(result: i32) -> Result<(), MtmdEncodeError> {
     if result == 0 {
         Ok(())
     } else {

llama-cpp-bindings/src/mtmd/mtmd_error.rs

@@ -73,7 +73,7 @@ pub enum MtmdEvalError {
     /// Requested batch size exceeds the context's maximum batch size
     #[error("batch size {requested} exceeds context batch size {context_max}")]
     BatchSizeExceedsContextLimit {
-        /// The batch size requested in eval_chunks
+        /// The batch size requested in `eval_chunks`
         requested: i32,
         /// The maximum batch size configured on the context
         context_max: u32,

llama-cpp-bindings/src/mtmd/mtmd_input_chunk.rs

@@ -11,7 +11,7 @@ use super::mtmd_input_chunk_type::{MtmdInputChunkType, MtmdInputChunkTypeError};
 ///
 /// `tokens_ptr` must point to at least `n_tokens` valid `llama_token` values
 /// that remain valid for the lifetime `'chunk`.
-unsafe fn tokens_from_raw_ptr<'chunk>(
+const unsafe fn tokens_from_raw_ptr<'chunk>(
     tokens_ptr: *const llama_cpp_bindings_sys::llama_token,
     n_tokens: usize,
 ) -> Option<&'chunk [LlamaToken]> {
@@ -131,7 +131,7 @@ mod unit_tests {
     #[test]
     fn tokens_from_raw_ptr_returns_none_for_zero_count() {
         let token: llama_cpp_bindings_sys::llama_token = 42;
-        assert!(unsafe { tokens_from_raw_ptr(&token, 0) }.is_none());
+        assert!(unsafe { tokens_from_raw_ptr(&raw const token, 0) }.is_none());
     }
 
     #[test]

llama-cpp-bindings/src/mtmd/mtmd_input_chunk_type.rs

@@ -37,9 +37,9 @@ impl TryFrom<llama_cpp_bindings_sys::mtmd_input_chunk_type> for MtmdInputChunkTy
         chunk_type: llama_cpp_bindings_sys::mtmd_input_chunk_type,
     ) -> Result<Self, Self::Error> {
         match chunk_type {
-            llama_cpp_bindings_sys::MTMD_INPUT_CHUNK_TYPE_TEXT => Ok(MtmdInputChunkType::Text),
-            llama_cpp_bindings_sys::MTMD_INPUT_CHUNK_TYPE_IMAGE => Ok(MtmdInputChunkType::Image),
-            llama_cpp_bindings_sys::MTMD_INPUT_CHUNK_TYPE_AUDIO => Ok(MtmdInputChunkType::Audio),
+            llama_cpp_bindings_sys::MTMD_INPUT_CHUNK_TYPE_TEXT => Ok(Self::Text),
+            llama_cpp_bindings_sys::MTMD_INPUT_CHUNK_TYPE_IMAGE => Ok(Self::Image),
+            llama_cpp_bindings_sys::MTMD_INPUT_CHUNK_TYPE_AUDIO => Ok(Self::Audio),
             unknown => Err(MtmdInputChunkTypeError(unknown)),
         }
     }

llama-cpp-bindings/src/mtmd/mtmd_input_chunks.rs

@@ -6,7 +6,7 @@ use super::mtmd_context::MtmdContext;
 use super::mtmd_error::MtmdEvalError;
 use super::mtmd_input_chunk::MtmdInputChunk;
 
-fn check_eval_result(result: i32) -> Result<(), MtmdEvalError> {
+const fn check_eval_result(result: i32) -> Result<(), MtmdEvalError> {
     if result == 0 {
         Ok(())
     } else {
@@ -109,7 +109,7 @@ impl MtmdInputChunks {
     ) -> Result<llama_cpp_bindings_sys::llama_pos, MtmdEvalError> {
         let context_max_batch = llama_ctx.n_batch();
 
-        if n_batch > 0 && (n_batch as u32) > context_max_batch {
+        if n_batch > 0 && n_batch.cast_unsigned() > context_max_batch {
             return Err(MtmdEvalError::BatchSizeExceedsContextLimit {
                 requested: n_batch,
                 context_max: context_max_batch,

rust-toolchain.toml

@@ -1,2 +1,3 @@
 [toolchain]
 channel = "1.92.0"
+components = ["clippy", "rustfmt"]