Detect chat markers and validate chat tools once per model

.claude/rules/subproject-llama-cpp-bindings-types.md

@@ -5,5 +5,5 @@ paths:
 
 # `llama-cpp-bindings-types` Context
 
-- The purposse of `llama-cpp-bindings-types` is to provide a thin layer of types that do not need to rely on `llama.cpp` vendored library itself
+- The purposse of `llama-cpp-bindings-types` is to provide a thin layer of types that do not need to rely on the `llama.cpp` library itself
 - `llama-cpp-bindings-types` must not depend on llama.cpp bindings themselves

Cargo.toml

@@ -18,7 +18,7 @@ members = [
 
 [workspace.package]
 edition = "2024"
-version = "0.14.0"
+version = "0.15.0"
 license = "Apache-2.0"
 repository = "https://github.com/intentee/llama-cpp-bindings"
 
@@ -33,21 +33,22 @@ find_cuda_helper = "=0.2.0"
 hf-hub = "=0.5.0"
 inventory = "=0.3.24"
 libtest-mimic = "=0.8.2"
-llama-cpp-bindings = { path = "llama-cpp-bindings", version = "=0.14.0" }
-llama-cpp-bindings-build = { path = "llama-cpp-bindings-build", version = "=0.14.0" }
-llama-cpp-bindings-sys = { path = "llama-cpp-bindings-sys", version = "=0.14.0" }
-llama-cpp-bindings-types = { path = "llama-cpp-bindings-types", version = "=0.14.0" }
-llama-cpp-error-recorder = { path = "llama-cpp-error-recorder", version = "=0.14.0" }
-llama-cpp-ffi-status = { path = "llama-cpp-ffi-status", version = "=0.14.0" }
-llama-cpp-gbnf = { path = "llama-cpp-gbnf", version = "=0.14.0" }
-llama-cpp-log-decoder = { path = "llama-cpp-log-decoder", version = "=0.14.0" }
-llama-cpp-test-harness = { path = "llama-cpp-test-harness", version = "=0.14.0" }
-llama-cpp-test-harness-macros = { path = "llama-cpp-test-harness-macros", version = "=0.14.0" }
+llama-cpp-bindings = { path = "llama-cpp-bindings", version = "=0.15.0" }
+llama-cpp-bindings-build = { path = "llama-cpp-bindings-build", version = "=0.15.0" }
+llama-cpp-bindings-sys = { path = "llama-cpp-bindings-sys", version = "=0.15.0" }
+llama-cpp-bindings-types = { path = "llama-cpp-bindings-types", version = "=0.15.0" }
+llama-cpp-error-recorder = { path = "llama-cpp-error-recorder", version = "=0.15.0" }
+llama-cpp-ffi-status = { path = "llama-cpp-ffi-status", version = "=0.15.0" }
+llama-cpp-gbnf = { path = "llama-cpp-gbnf", version = "=0.15.0" }
+llama-cpp-log-decoder = { path = "llama-cpp-log-decoder", version = "=0.15.0" }
+llama-cpp-test-harness = { path = "llama-cpp-test-harness", version = "=0.15.0" }
+llama-cpp-test-harness-macros = { path = "llama-cpp-test-harness-macros", version = "=0.15.0" }
 llama-cpp-wrapper-error-fixture = { path = "llama-cpp-wrapper-error-fixture" }
-llama-cpp-wrapper-sources = { path = "llama-cpp-wrapper-sources", version = "=0.14.0" }
+llama-cpp-wrapper-sources = { path = "llama-cpp-wrapper-sources", version = "=0.15.0" }
 llguidance = "=1.7.0"
 log = "=0.4.29"
 nom = "=8.0.0"
+once_cell = "1.21"
 proc-macro2 = "=1.0.106"
 quote = "=1.0.45"
 serde = { version = "=1.0.228", features = ["derive"] }

Makefile

@@ -29,7 +29,7 @@ WRAPPER_SOURCES_CRATE_FILES = \
 EMIT_WRAPPER_BUILD_INPUTS = cargo run --quiet --package $(WRAPPER_SOURCES_CRATE) -- \
 	$(CURDIR)/llama-cpp-bindings-sys $(COMPILE_COMMANDS) $(WRAPPER_SOURCES_RESPONSE_FILE)
 
-VENDORED_SUPPRESSIONS = \
+LLAMA_CPP_AND_GSL_SUPPRESSIONS = \
 	--suppress='*:*llama-cpp-bindings-sys/llama.cpp/*' \
 	--suppress='*:*llama-cpp-bindings-sys/GSL/*'
 
@@ -115,7 +115,7 @@ lint.cpp.clang-tidy: $(COMPILE_COMMANDS) $(WRAPPER_SOURCES_RESPONSE_FILE)
 lint.cpp.cppcheck: $(COMPILE_COMMANDS) $(CPPCHECK)
 	$(CPPCHECK) --project=$(COMPILE_COMMANDS) --enable=all --inconclusive \
 		--check-level=exhaustive --error-exitcode=1 \
-		$(VENDORED_SUPPRESSIONS) \
+		$(LLAMA_CPP_AND_GSL_SUPPRESSIONS) \
 		--suppress=missingIncludeSystem
 
 .PHONY: test

llama-cpp-bindings-sys/wrapper_chat_parse.cpp

@@ -30,6 +30,61 @@ void dup_or_set_alloc_flag(const std::string & source, char ** out_dup, bool * o
     *out_dup = llama_rs_dup_string(source);
     *out_alloc_failed = (*out_dup == nullptr);
 }
+
+auto report_current_parse_exception(
+    char ** out_error,
+    llama_rs_parse_chat_message_status thrown_status) -> llama_rs_parse_chat_message_status {
+    try {
+        throw;
+    } catch (const std::bad_alloc &) {
+        return LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_OUT_OF_MEMORY;
+    } catch (const std::exception & ex) {
+        *out_error = llama_rs_dup_string(std::string(ex.what()));
+    } catch (...) {
+        *out_error = llama_rs_dup_string(std::string("unknown c++ exception"));
+    }
+    if (*out_error == nullptr) {
+        return LLAMA_RS_PARSE_CHAT_MESSAGE_ERROR_STRING_ALLOCATION_FAILED;
+    }
+    return thrown_status;
+}
+
+auto build_tools_parser(const llama_rs_chat_parser & parser, const char * tools_json) -> common_peg_arena {
+    autoparser::generation_params inputs;
+
+    if ((tools_json != nullptr) && *tools_json != '\0') {
+        inputs.tools = common_json::parse(tools_json);
+    } else {
+        inputs.tools = common_json::array();
+    }
+
+    return parser.parser.build_parser(inputs, std::string());
+}
+
+auto parse_with_tools_parser(
+    const common_peg_arena & chat_parser,
+    const char * input,
+    int is_partial,
+    llama_rs_parsed_chat_handle * out_handle,
+    char ** out_error) -> llama_rs_parse_chat_message_status {
+    try {
+        common_chat_parser_params parser_params;
+        parser_params.format = COMMON_CHAT_FORMAT_PEG_NATIVE;
+
+        common_chat_msg parsed =
+            common_chat_peg_parse(chat_parser, input, is_partial != 0, parser_params);
+
+        auto handle = std::make_unique<llama_rs_parsed_chat>();
+        handle->message = std::move(parsed);
+
+        *out_handle = handle.release();
+
+        return LLAMA_RS_PARSE_CHAT_MESSAGE_OK;
+    } catch (...) {
+        return report_current_parse_exception(
+            out_error, LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_THREW_CXX_EXCEPTION);
+    }
+}
 } // namespace
 
 extern "C" auto llama_rs_chat_parser_create(
@@ -148,42 +203,12 @@ extern "C" auto llama_rs_parse_chat_message(
     }
 
     try {
-        autoparser::generation_params inputs;
-
-        if ((tools_json != nullptr) && *tools_json != '\0') {
-            inputs.tools = common_json::parse(tools_json);
-        } else {
-            inputs.tools = common_json::array();
-        }
-
-        common_peg_arena const chat_parser = parser->parser.build_parser(inputs, std::string());
-
-        common_chat_parser_params parser_params;
-        parser_params.format = COMMON_CHAT_FORMAT_PEG_NATIVE;
-
-        common_chat_msg parsed =
-            common_chat_peg_parse(chat_parser, input, is_partial != 0, parser_params);
+        common_peg_arena const chat_parser = build_tools_parser(*parser, tools_json);
 
-        auto handle = std::make_unique<llama_rs_parsed_chat>();
-        handle->message = std::move(parsed);
-
-        *out_handle = handle.release();
-
-        return LLAMA_RS_PARSE_CHAT_MESSAGE_OK;
-    } catch (const std::bad_alloc &) {
-        return LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_OUT_OF_MEMORY;
-    } catch (const std::exception & ex) {
-        *out_error = llama_rs_dup_string(std::string(ex.what()));
-        if (*out_error == nullptr) {
-            return LLAMA_RS_PARSE_CHAT_MESSAGE_ERROR_STRING_ALLOCATION_FAILED;
-        }
-        return LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_THREW_CXX_EXCEPTION;
+        return parse_with_tools_parser(chat_parser, input, is_partial, out_handle, out_error);
     } catch (...) {
-        *out_error = llama_rs_dup_string(std::string("unknown c++ exception"));
-        if (*out_error == nullptr) {
-            return LLAMA_RS_PARSE_CHAT_MESSAGE_ERROR_STRING_ALLOCATION_FAILED;
-        }
-        return LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_THREW_CXX_EXCEPTION;
+        return report_current_parse_exception(
+            out_error, LLAMA_RS_PARSE_CHAT_MESSAGE_TOOLS_PARSER_BUILD_THREW_CXX_EXCEPTION);
     }
 }
 

llama-cpp-bindings-sys/wrapper_chat_parse.h

@@ -52,6 +52,7 @@ typedef enum llama_rs_parse_chat_message_status {
     LLAMA_RS_PARSE_CHAT_MESSAGE_ERROR_STRING_ALLOCATION_FAILED,
     LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_OUT_OF_MEMORY,
     LLAMA_RS_PARSE_CHAT_MESSAGE_LLAMA_CPP_THREW_CXX_EXCEPTION,
+    LLAMA_RS_PARSE_CHAT_MESSAGE_TOOLS_PARSER_BUILD_THREW_CXX_EXCEPTION,
 } llama_rs_parse_chat_message_status;
 
 llama_rs_parse_chat_message_status llama_rs_parse_chat_message(