diff options
| author | Onur Gündüz <ogunduz@gmail.com> | 2026-06-25 14:00:54 +0200 |
|---|---|---|
| committer | Onur Gündüz <ogunduz@gmail.com> | 2026-06-25 14:01:11 +0200 |
| commit | 13dd2132310fb4dc5aa3dd217ed62f2e8301b981 (patch) | |
| tree | 0e6f5dfdc686365af06fb59d32957e47a3ca5a79 /tools/server | |
| parent | b673eb9a72138b89f1efecf52578020cb43857ff (diff) | |
| download | ffwd-13dd2132310fb4dc5aa3dd217ed62f2e8301b981.tar.gz ffwd-13dd2132310fb4dc5aa3dd217ed62f2e8301b981.zip | |
all: migrate to yyjson
Diffstat (limited to 'tools/server')
| -rw-r--r-- | tools/server/Makefile | 4 | ||||
| -rw-r--r-- | tools/server/encode.c | 3 | ||||
| -rw-r--r-- | tools/server/handlers.c | 253 | ||||
| -rw-r--r-- | tools/server/json.c | 121 | ||||
| -rw-r--r-- | tools/server/models.c | 28 | ||||
| -rw-r--r-- | tools/server/schedule.c | 16 | ||||
| -rw-r--r-- | tools/server/server.c | 2 | ||||
| -rw-r--r-- | tools/server/server_internal.h | 40 | ||||
| -rw-r--r-- | tools/server/test_json.c | 119 |
9 files changed, 312 insertions, 274 deletions
diff --git a/tools/server/Makefile b/tools/server/Makefile index 4a11b0d..34c8608 100644 --- a/tools/server/Makefile +++ b/tools/server/Makefile @@ -15,10 +15,10 @@ TEST_BINS := test_base64 test_sbuf test_encode test_json test_schedule test_mode TEST_OBJS := $(filter-out $(OBJDIR)/main.o,$(OBJS)) LIBS := $(OUTDIR)/libffwd.a $(OUTDIR)/libae.a -LDLIBS := $(CJSON_LDFLAGS) $(BACKEND_LDFLAGS) -lm +LDLIBS := $(YYJSON_LDFLAGS) $(BACKEND_LDFLAGS) -lm CFLAGS += -std=gnu11 -pthread -CPPFLAGS += -I. -I$(LIBFFWD) -I$(AE) $(CJSON_CFLAGS) +CPPFLAGS += -I. -I$(LIBFFWD) -I$(AE) $(OBJDIR)/main.o: CFLAGS += $(VERSION_CFLAGS) diff --git a/tools/server/encode.c b/tools/server/encode.c index 61ff4ea..1674774 100644 --- a/tools/server/encode.c +++ b/tools/server/encode.c @@ -68,6 +68,9 @@ static void append_embedding_object(sbuf *b, int index, const char *embedding) { void append_embedding_value( sbuf *b, int index, const float *emb, int dims, const char *encoding, embedding_api_t api) { if (!strcmp(encoding, "float")) { + /* Floats are hand-formatted with %.9g (the shortest form that + * round-trips float32) and streamed into the sbuf, never serialized + * through yyjson's writer. */ sbuf_printf(b, "{\"object\":\"embedding\",\"index\":%d,\"embedding\":[", index); for (int i = 0; i < dims; i++) { if (i) diff --git a/tools/server/handlers.c b/tools/server/handlers.c index fed4671..2c98077 100644 --- a/tools/server/handlers.c +++ b/tools/server/handlers.c @@ -249,22 +249,21 @@ void execute_contextual_request_list(contextual_request **reqs, int n_reqs) { free(items); } -static int validate_common(cJSON *root, - cJSON *detail, +static int validate_common(yyjson_val *root, + yyjson_mut_doc *detail, model_kind expected_kind, loaded_model **out_model, int *out_dims, const char **out_encoding, http_server *s) { - cJSON *model_item = cJSON_GetObjectItemCaseSensitive(root, "model"); + yyjson_val *model_item = yyjson_obj_get(root, "model"); const char *model_id = NULL; loaded_model *m = NULL; if (!model_item) { ve_add(detail, "[\"body\",\"model\"]", "field required", "missing"); - } else if (!cJSON_IsString(model_item) || !model_item->valuestring) { + } else if (!(model_id = yyjson_get_str(model_item))) { ve_add(detail, "[\"body\",\"model\"]", "model must be a string", "type_error.string"); } else { - model_id = model_item->valuestring; m = loaded_model_for_label(s, model_id); if (!m) { ve_add(detail, "[\"body\",\"model\"]", @@ -284,12 +283,13 @@ static int validate_common(cJSON *root, return 0; } -void prepare_embedding_request(job *j, cJSON *root, http_server *s, embedding_request *out) { +void prepare_embedding_request(job *j, yyjson_doc *root_doc, http_server *s, embedding_request *out) { memset(out, 0, sizeof(*out)); out->j = j; - out->root = root; + out->root = root_doc; + yyjson_val *root = yyjson_doc_get_root(root_doc); - cJSON *detail = cJSON_CreateArray(); + yyjson_mut_doc *detail = ve_new(); loaded_model *m = NULL; int dims = 0; const char *encoding = NULL; @@ -303,47 +303,49 @@ void prepare_embedding_request(job *j, cJSON *root, http_server *s, embedding_re ? m->info->query_instruct : NULL; - cJSON *input = cJSON_GetObjectItemCaseSensitive(root, "input"); + yyjson_val *input = yyjson_obj_get(root, "input"); int n_inputs = 0; if (!input) { ve_add(detail, "[\"body\",\"input\"]", "field required", "missing"); - } else if (cJSON_IsString(input)) { - if (!input->valuestring || input->valuestring[0] == '\0') + } else if (yyjson_is_str(input)) { + const char *text = yyjson_get_str(input); + if (!text || text[0] == '\0') ve_add(detail, "[\"body\",\"input\"]", "input must not be empty", "value_error.empty"); n_inputs = 1; - } else if (cJSON_IsArray(input)) { - n_inputs = cJSON_GetArraySize(input); + } else if (yyjson_is_arr(input)) { + n_inputs = (int)yyjson_arr_size(input); if (n_inputs < 1) ve_add(detail, "[\"body\",\"input\"]", "input must contain at least 1 item", "value_error.list.min_items"); else if (n_inputs > FFWD_API_MAX_STANDARD_INPUTS) ve_add(detail, "[\"body\",\"input\"]", "input must contain at most 512 items", "value_error.list.max_items"); - cJSON *item; - int i = 0; - cJSON_ArrayForEach(item, input) { + size_t idx, max; + yyjson_val *item; + yyjson_arr_foreach(input, idx, max, item) { char loc[64]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", i); - if (!cJSON_IsString(item)) { + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", (int)idx); + if (!yyjson_is_str(item)) { ve_add(detail, loc, "input item must be a string", "type_error.string"); - } else if (!item->valuestring || item->valuestring[0] == '\0') { - ve_add(detail, loc, "input item must not be empty", "value_error.empty"); + } else { + const char *text = yyjson_get_str(item); + if (!text || text[0] == '\0') + ve_add(detail, loc, "input item must not be empty", "value_error.empty"); } - i++; } } else { ve_add(detail, "[\"body\",\"input\"]", "input must be a string or an array of strings", "type_error"); } - if (cJSON_GetArraySize(detail) == 0 && m && input) { + if (ve_count(detail) == 0 && m && input) { out->n_inputs = n_inputs; out->tokens = xcalloc((size_t)n_inputs, sizeof(*out->tokens)); out->inputs = xmalloc((size_t)n_inputs * sizeof(*out->inputs)); - int idx = 0; - if (input && cJSON_IsString(input) && input->valuestring) { - if (tokenize_input(m, j, input->valuestring, query_instruct, &out->tokens[0]) != 0) { + const char *single = yyjson_get_str(input); + if (single) { + if (tokenize_input(m, j, single, query_instruct, &out->tokens[0]) != 0) { ve_add(detail, "[\"body\",\"input\"]", "tokenization failed", "value_error.tokenization"); } else { @@ -355,16 +357,17 @@ void prepare_embedding_request(job *j, cJSON *root, http_server *s, embedding_re "value_error.context_length"); } } else { - cJSON *item; - cJSON_ArrayForEach(item, input) { + size_t idx, max; + yyjson_val *item; + yyjson_arr_foreach(input, idx, max, item) { char loc[64]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", idx); - if (!cJSON_IsString(item) || !item->valuestring) { + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", (int)idx); + const char *text = yyjson_get_str(item); + if (!text) { ve_add(detail, loc, "input item must be a string", "type_error.string"); - idx++; continue; } - if (tokenize_input(m, j, item->valuestring, query_instruct, &out->tokens[idx]) != 0) { + if (tokenize_input(m, j, text, query_instruct, &out->tokens[idx]) != 0) { ve_add(detail, loc, "tokenization failed", "value_error.tokenization"); } else { out->inputs[idx].ids = out->tokens[idx].ids; @@ -374,7 +377,6 @@ void prepare_embedding_request(job *j, cJSON *root, http_server *s, embedding_re ve_add(detail, loc, "input exceeds 32768 token limit", "value_error.context_length"); } - idx++; } } @@ -383,22 +385,26 @@ void prepare_embedding_request(job *j, cJSON *root, http_server *s, embedding_re "value_error.context_length"); } - if (cJSON_GetArraySize(detail) > 0) { + if (ve_count(detail) > 0) { job_set_422(j, detail); embedding_request_free(out); out->j = j; } else { out->ready = 1; } - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); } -void prepare_contextual_request(job *j, cJSON *root, http_server *s, contextual_request *out) { +void prepare_contextual_request(job *j, + yyjson_doc *root_doc, + http_server *s, + contextual_request *out) { memset(out, 0, sizeof(*out)); out->j = j; - out->root = root; + out->root = root_doc; + yyjson_val *root = yyjson_doc_get_root(root_doc); - cJSON *detail = cJSON_CreateArray(); + yyjson_mut_doc *detail = ve_new(); loaded_model *m = NULL; int dims = 0; const char *encoding = NULL; @@ -407,94 +413,92 @@ void prepare_contextual_request(job *j, cJSON *root, http_server *s, contextual_ out->dims = dims; out->encoding = encoding; - cJSON *input = cJSON_GetObjectItemCaseSensitive(root, "input"); + yyjson_val *input = yyjson_obj_get(root, "input"); int n_docs = 0; int total_chunks = 0; if (!input) { ve_add(detail, "[\"body\",\"input\"]", "field required", "missing"); - } else if (!cJSON_IsArray(input)) { + } else if (!yyjson_is_arr(input)) { ve_add(detail, "[\"body\",\"input\"]", "input must be an array of document chunk arrays", "type_error.array"); } else { - n_docs = cJSON_GetArraySize(input); + n_docs = (int)yyjson_arr_size(input); if (n_docs < 1) ve_add(detail, "[\"body\",\"input\"]", "input must contain at least 1 document", "value_error.list.min_items"); else if (n_docs > FFWD_API_MAX_CONTEXT_DOCS) ve_add(detail, "[\"body\",\"input\"]", "input must contain at most 512 documents", "value_error.list.max_items"); - cJSON *doc_arr; - int di = 0; - cJSON_ArrayForEach(doc_arr, input) { - if (!cJSON_IsArray(doc_arr)) { + size_t di, dn; + yyjson_val *doc_arr; + yyjson_arr_foreach(input, di, dn, doc_arr) { + if (!yyjson_is_arr(doc_arr)) { char loc[64]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", di); + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", (int)di); ve_add(detail, loc, "document must be an array of strings", "type_error.array"); - di++; continue; } - int doc_chunks = cJSON_GetArraySize(doc_arr); + int doc_chunks = (int)yyjson_arr_size(doc_arr); if (doc_chunks > FFWD_API_MAX_CONTEXT_CHUNKS - total_chunks) total_chunks = FFWD_API_MAX_CONTEXT_CHUNKS + 1; else total_chunks += doc_chunks; if (doc_chunks < 1) { char loc[64]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", di); + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", (int)di); ve_add(detail, loc, "document must contain at least 1 chunk", "value_error.list.min_items"); } - cJSON *chunk; - int ci = 0; - cJSON_ArrayForEach(chunk, doc_arr) { + size_t ci, cn; + yyjson_val *chunk; + yyjson_arr_foreach(doc_arr, ci, cn, chunk) { char loc[80]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d,%d]", di, ci); - if (!cJSON_IsString(chunk)) { + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d,%d]", (int)di, (int)ci); + if (!yyjson_is_str(chunk)) { ve_add(detail, loc, "chunk must be a string", "type_error.string"); - } else if (!chunk->valuestring || chunk->valuestring[0] == '\0') { - ve_add(detail, loc, "chunk must not be empty", "value_error.empty"); + } else { + const char *text = yyjson_get_str(chunk); + if (!text || text[0] == '\0') + ve_add(detail, loc, "chunk must not be empty", "value_error.empty"); } - ci++; } - di++; } if (total_chunks > FFWD_API_MAX_CONTEXT_CHUNKS) ve_add(detail, "[\"body\",\"input\"]", "input must contain at most 16000 chunks", "value_error.list.max_items"); } - if (cJSON_GetArraySize(detail) == 0 && m) { + if (ve_count(detail) == 0 && m) { out->n_docs = n_docs; out->docs = xcalloc((size_t)n_docs, sizeof(*out->docs)); int64_t request_tokens = 0; - cJSON *doc_arr; - int di = 0; - cJSON_ArrayForEach(doc_arr, input) { + size_t di, dn; + yyjson_val *doc_arr; + yyjson_arr_foreach(input, di, dn, doc_arr) { contextual_doc *doc = &out->docs[di]; - int n_chunks = cJSON_GetArraySize(doc_arr); + int n_chunks = (int)yyjson_arr_size(doc_arr); token_buf *chunk_tokens = xcalloc((size_t)n_chunks, sizeof(*chunk_tokens)); doc->spans = xcalloc((size_t)n_chunks, sizeof(*doc->spans)); doc->n_spans = n_chunks; int64_t doc_tokens = n_chunks > 1 ? n_chunks - 1 : 0; int doc_valid = 1; - cJSON *chunk; - int ci = 0; - cJSON_ArrayForEach(chunk, doc_arr) { + size_t ci, cn; + yyjson_val *chunk; + yyjson_arr_foreach(doc_arr, ci, cn, chunk) { char loc[80]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d,%d]", di, ci); - if (tokenize_one(m, j, chunk->valuestring, &chunk_tokens[ci]) != 0) { + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d,%d]", (int)di, (int)ci); + if (tokenize_one(m, j, yyjson_get_str(chunk), &chunk_tokens[ci]) != 0) { ve_add(detail, loc, "tokenization failed", "value_error.tokenization"); doc_valid = 0; } else { doc_tokens += chunk_tokens[ci].n_tokens; } - ci++; } if (doc_tokens > FFWD_API_MAX_ITEM_TOKENS) { char loc[64]; - snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", di); + snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", (int)di); ve_add(detail, loc, "document exceeds 32768 token limit", "value_error.context_length"); doc_valid = 0; @@ -506,24 +510,23 @@ void prepare_contextual_request(job *j, cJSON *root, http_server *s, contextual_ doc->ids = xmalloc((size_t)doc->n_tokens * sizeof(*doc->ids)); int sep_id = ffwd_tok_context_separator_id(m->tok); int pos = 0; - for (ci = 0; ci < n_chunks; ci++) { - doc->spans[ci].start = pos; - doc->spans[ci].n_tokens = chunk_tokens[ci].n_tokens; + for (int k = 0; k < n_chunks; k++) { + doc->spans[k].start = pos; + doc->spans[k].n_tokens = chunk_tokens[k].n_tokens; /* n_tokens > 0 implies ids != NULL (tokenize_one sets both * together), and the guard avoids a memcpy(dst, NULL, 0) for * an empty chunk that the static analyzer reports as UB. */ - if (chunk_tokens[ci].n_tokens > 0) - memcpy(doc->ids + pos, chunk_tokens[ci].ids, - (size_t)chunk_tokens[ci].n_tokens * sizeof(*doc->ids)); - pos += chunk_tokens[ci].n_tokens; - if (ci + 1 < n_chunks) { + if (chunk_tokens[k].n_tokens > 0) + memcpy(doc->ids + pos, chunk_tokens[k].ids, + (size_t)chunk_tokens[k].n_tokens * sizeof(*doc->ids)); + pos += chunk_tokens[k].n_tokens; + if (k + 1 < n_chunks) { doc->ids[pos++] = sep_id; } } } free_token_bufs(chunk_tokens, n_chunks); free(chunk_tokens); - di++; } if (request_tokens > FFWD_API_MAX_TOTAL_TOKENS) { ve_add(detail, "[\"body\",\"input\"]", "request exceeds 120000 token limit", @@ -534,28 +537,31 @@ void prepare_contextual_request(job *j, cJSON *root, http_server *s, contextual_ } } - if (cJSON_GetArraySize(detail) > 0) { + if (ve_count(detail) > 0) { job_set_422(j, detail); contextual_request_free(out); out->j = j; } else { out->ready = 1; } - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); } -static int -validate_rerank_model(cJSON *root, cJSON *detail, http_server *s, loaded_model **out_model) { - cJSON *item = cJSON_GetObjectItemCaseSensitive(root, "model"); +static int validate_rerank_model(yyjson_val *root, + yyjson_mut_doc *detail, + http_server *s, + loaded_model **out_model) { + yyjson_val *item = yyjson_obj_get(root, "model"); if (!item) { ve_add(detail, "[\"body\",\"model\"]", "field required", "missing"); return 0; } - if (!cJSON_IsString(item) || !item->valuestring) { + const char *model_id = yyjson_get_str(item); + if (!model_id) { ve_add(detail, "[\"body\",\"model\"]", "model must be a string", "type_error.string"); return 0; } - loaded_model *m = loaded_model_for_label(s, item->valuestring); + loaded_model *m = loaded_model_for_label(s, model_id); if (!m) { ve_add(detail, "[\"body\",\"model\"]", "value is not a valid enum member for this endpoint", "enum"); @@ -569,85 +575,89 @@ validate_rerank_model(cJSON *root, cJSON *detail, http_server *s, loaded_model * return 0; } -void prepare_rerank_request(job *j, cJSON *root, http_server *s, rerank_request *out) { +void prepare_rerank_request(job *j, yyjson_doc *root_doc, http_server *s, rerank_request *out) { memset(out, 0, sizeof(*out)); out->j = j; - out->root = root; + out->root = root_doc; + yyjson_val *root = yyjson_doc_get_root(root_doc); - cJSON *detail = cJSON_CreateArray(); + yyjson_mut_doc *detail = ve_new(); validate_rerank_model(root, detail, s, &out->model); - cJSON *query = cJSON_GetObjectItemCaseSensitive(root, "query"); - cJSON *documents = cJSON_GetObjectItemCaseSensitive(root, "documents"); + yyjson_val *query = yyjson_obj_get(root, "query"); + yyjson_val *documents = yyjson_obj_get(root, "documents"); /* query_str is set only once query passes every check; the tokenize step * below runs only when detail is empty, so it is non-NULL there. Hoisting it * also keeps the validated value explicit for the static analyzer. */ const char *query_str = NULL; if (!query) { ve_add(detail, "[\"body\",\"query\"]", "field required", "missing"); - } else if (!cJSON_IsString(query)) { + } else if (!yyjson_is_str(query)) { ve_add(detail, "[\"body\",\"query\"]", "query must be a string", "type_error.string"); - } else if (!query->valuestring || !query->valuestring[0]) { + } else if (!yyjson_get_str(query)[0]) { ve_add(detail, "[\"body\",\"query\"]", "query must not be empty", "value_error.empty"); } else { - query_str = query->valuestring; + query_str = yyjson_get_str(query); } int n_documents = 0; if (!documents) { ve_add(detail, "[\"body\",\"documents\"]", "field required", "missing"); - } else if (!cJSON_IsArray(documents)) { + } else if (!yyjson_is_arr(documents)) { ve_add(detail, "[\"body\",\"documents\"]", "documents must be an array of strings", "type_error.array"); } else { - n_documents = cJSON_GetArraySize(documents); + n_documents = (int)yyjson_arr_size(documents); if (n_documents < 1) ve_add(detail, "[\"body\",\"documents\"]", "documents must contain at least 1 item", "value_error.list.min_items"); else if (n_documents > FFWD_API_MAX_RERANK_DOCUMENTS) ve_add(detail, "[\"body\",\"documents\"]", "documents must contain at most 1000 items", "value_error.list.max_items"); - cJSON *document; - int i = 0; - cJSON_ArrayForEach(document, documents) { + size_t i, dmax; + yyjson_val *document; + yyjson_arr_foreach(documents, i, dmax, document) { char loc[72]; - snprintf(loc, sizeof(loc), "[\"body\",\"documents\",%d]", i++); - if (!cJSON_IsString(document)) + snprintf(loc, sizeof(loc), "[\"body\",\"documents\",%d]", (int)i); + if (!yyjson_is_str(document)) { ve_add(detail, loc, "document must be a string", "type_error.string"); - else if (!document->valuestring || !document->valuestring[0]) - ve_add(detail, loc, "document must not be empty", "value_error.empty"); + } else { + const char *text = yyjson_get_str(document); + if (!text || !text[0]) + ve_add(detail, loc, "document must not be empty", "value_error.empty"); + } } } - cJSON *top_n = cJSON_GetObjectItemCaseSensitive(root, "top_n"); - cJSON *top_k = cJSON_GetObjectItemCaseSensitive(root, "top_k"); + yyjson_val *top_n = yyjson_obj_get(root, "top_n"); + yyjson_val *top_k = yyjson_obj_get(root, "top_k"); if (top_n && top_k) { ve_add(detail, "[\"body\",\"top_n\"]", "top_n and top_k are aliases; provide only one", "value_error.conflict"); } - cJSON *top = top_n ? top_n : top_k; + yyjson_val *top = top_n ? top_n : top_k; out->top_n = n_documents; if (top) { - if (!cjson_is_integer(top)) { + if (!json_is_integer(top)) { ve_add(detail, top_n ? "[\"body\",\"top_n\"]" : "[\"body\",\"top_k\"]", "top_n must be an integer", "type_error.integer"); } else { - out->top_n = top->valueint; + out->top_n = (int)yyjson_get_num(top); if (out->top_n < 1 || out->top_n > n_documents) ve_add(detail, top_n ? "[\"body\",\"top_n\"]" : "[\"body\",\"top_k\"]", "top_n must be between 1 and the number of documents", "value_error.range"); } } - cJSON *return_documents = cJSON_GetObjectItemCaseSensitive(root, "return_documents"); + yyjson_val *return_documents = yyjson_obj_get(root, "return_documents"); if (return_documents) { - if (!cJSON_IsBool(return_documents)) + if (!yyjson_is_bool(return_documents)) ve_add(detail, "[\"body\",\"return_documents\"]", "return_documents must be a boolean", "type_error.bool"); else - out->return_documents = cJSON_IsTrue(return_documents); + out->return_documents = yyjson_get_bool(return_documents); } - if (cJSON_GetArraySize(detail) == 0 && out->model) { + if (ve_count(detail) == 0 && out->model) { out->n_documents = n_documents; out->documents = xcalloc((size_t)n_documents, sizeof(*out->documents)); if (tokenize_late_text(out->model, j, query_str, 1, &out->query) != 0) { @@ -659,12 +669,12 @@ void prepare_rerank_request(job *j, cJSON *root, http_server *s, rerank_request "value_error.context_length"); } - cJSON *document; - int i = 0; - cJSON_ArrayForEach(document, documents) { + size_t i, dmax; + yyjson_val *document; + yyjson_arr_foreach(documents, i, dmax, document) { char loc[72]; - snprintf(loc, sizeof(loc), "[\"body\",\"documents\",%d]", i); - if (tokenize_late_text(out->model, j, document->valuestring, 0, &out->documents[i]) != + snprintf(loc, sizeof(loc), "[\"body\",\"documents\",%d]", (int)i); + if (tokenize_late_text(out->model, j, yyjson_get_str(document), 0, &out->documents[i]) != 0) { ve_add(detail, loc, "tokenization failed", "value_error.tokenization"); } else { @@ -677,7 +687,6 @@ void prepare_rerank_request(job *j, cJSON *root, http_server *s, rerank_request else out->document_tokens = INT_MAX; } - i++; } if (out->query_tokens > FFWD_API_MAX_TOTAL_TOKENS - out->document_tokens) { ve_add(detail, "[\"body\",\"documents\"]", "request exceeds 120000 token limit", @@ -685,14 +694,14 @@ void prepare_rerank_request(job *j, cJSON *root, http_server *s, rerank_request } } - if (cJSON_GetArraySize(detail) > 0) { + if (ve_count(detail) > 0) { job_set_422(j, detail); rerank_request_free(out); out->j = j; } else { out->ready = 1; } - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); } typedef struct { @@ -764,16 +773,16 @@ void execute_rerank_request(rerank_request *r) { t0 = nstime(); sbuf b = {0}; sbuf_printf(&b, "{\"object\":\"list\",\"model\":\"%s\",\"results\":[", r->model->info->id); - cJSON *documents = cJSON_GetObjectItemCaseSensitive(r->root, "documents"); + yyjson_val *documents = yyjson_obj_get(yyjson_doc_get_root(r->root), "documents"); for (int i = 0; i < r->top_n; i++) { if (i) sbuf_putc(&b, ','); sbuf_printf(&b, "{\"index\":%d,\"relevance_score\":%.9g", ranked[i].index, (double)ranked[i].score); if (r->return_documents) { - cJSON *document = cJSON_GetArrayItem(documents, ranked[i].index); + yyjson_val *document = yyjson_arr_get(documents, (size_t)ranked[i].index); sbuf_puts(&b, ",\"document\":"); - append_json_string(&b, document->valuestring); + append_json_string(&b, yyjson_get_str(document)); } sbuf_putc(&b, '}'); } diff --git a/tools/server/json.c b/tools/server/json.c index 4812164..8f9aec2 100644 --- a/tools/server/json.c +++ b/tools/server/json.c @@ -1,8 +1,6 @@ #include "server_internal.h" #include "util.h" -#include <cjson/cJSON.h> -#include <ctype.h> #include <limits.h> #include <stdio.h> #include <stdlib.h> @@ -75,26 +73,47 @@ void job_set_error(job *j, int status, const char *message, const char *type) { j->response = json_error_body(message, type, &j->response_len); } -bool ve_add(cJSON *detail, const char *loc_json, const char *msg, const char *type) { - cJSON *obj = cJSON_CreateObject(); - cJSON *loc = cJSON_Parse(loc_json); - if (!obj || !loc) { - cJSON_Delete(obj); - cJSON_Delete(loc); +yyjson_mut_doc *ve_new(void) { + yyjson_mut_doc *detail = yyjson_mut_doc_new(NULL); + yyjson_mut_doc_set_root(detail, yyjson_mut_arr(detail)); + return detail; +} + +bool ve_add(yyjson_mut_doc *detail, const char *loc_json, const char *msg, const char *type) { + yyjson_mut_val *arr = yyjson_mut_doc_get_root(detail); + yyjson_mut_val *obj = yyjson_mut_obj(detail); + /* loc is a JSON array literal (mix of strings and integer indices) built by + * the caller; parse it and deep-copy it into the detail document so it owns + * the values once the temporary parse is freed. */ + yyjson_doc *loc_doc = yyjson_read(loc_json, strlen(loc_json), 0); + yyjson_mut_val *loc = loc_doc ? yyjson_val_mut_copy(detail, yyjson_doc_get_root(loc_doc)) : NULL; + yyjson_doc_free(loc_doc); + if (!arr || !obj || !loc) return false; - } - cJSON_AddItemToObject(obj, "loc", loc); - cJSON_AddStringToObject(obj, "msg", msg ? msg : ""); - cJSON_AddStringToObject(obj, "type", type ? type : ""); - cJSON_AddItemToArray(detail, obj); + yyjson_mut_obj_add_val(detail, obj, "loc", loc); + /* msg may point at a caller stack buffer, so copy the value strings. */ + yyjson_mut_obj_add_strcpy(detail, obj, "msg", msg ? msg : ""); + yyjson_mut_obj_add_strcpy(detail, obj, "type", type ? type : ""); + yyjson_mut_arr_add_val(arr, obj); return true; } -void job_set_422(job *j, cJSON *detail) { - cJSON *root = cJSON_CreateObject(); - cJSON_AddItemToObject(root, "detail", cJSON_Duplicate(detail, 1)); - char *s = cJSON_PrintUnformatted(root); - cJSON_Delete(root); +size_t ve_count(yyjson_mut_doc *detail) { + return yyjson_mut_arr_size(yyjson_mut_doc_get_root(detail)); +} + +void job_set_422(job *j, yyjson_mut_doc *detail) { + /* Reparent the detail array under a fresh {"detail":[...]} root in the same + * document, then serialize. The caller frees the document right after, so + * mutating its root here is safe and avoids a deep copy. */ + yyjson_mut_val *arr = yyjson_mut_doc_get_root(detail); + yyjson_mut_val *root = yyjson_mut_obj(detail); + size_t len = 0; + char *s = NULL; + if (root && arr && yyjson_mut_obj_add_val(detail, root, "detail", arr)) { + yyjson_mut_doc_set_root(detail, root); + s = yyjson_mut_write(detail, 0, &len); + } if (!s) { job_set_error(j, 500, "failed to render validation error", "server_error"); return; @@ -102,57 +121,52 @@ void job_set_422(job *j, cJSON *detail) { free(j->response); j->status = 422; j->content_type = "application/json"; - j->response = xstrdup(s); - j->response_len = strlen(j->response); - cJSON_free(s); + j->response = s; + j->response_len = len; } -cJSON *parse_json_body(job *j, cJSON *detail) { +yyjson_doc *parse_json_body(job *j, yyjson_mut_doc *detail) { if (!j->body || j->body_len == 0) { ve_add(detail, "[\"body\"]", "body required", "missing"); return NULL; } - const char *end = NULL; - cJSON *root = cJSON_ParseWithLengthOpts(j->body, j->body_len, &end, false); - if (!root) { + /* yyjson_read with default flags requires the whole buffer to be one JSON + * value (trailing whitespace is allowed, trailing content is an error), so + * it already enforces "the entire body must be consumed". */ + yyjson_doc *doc = yyjson_read(j->body, j->body_len, 0); + if (!doc) { ve_add(detail, "[\"body\"]", "invalid JSON", "json_invalid"); return NULL; } - const char *body_end = j->body + j->body_len; - while (end && end < body_end && isspace((unsigned char)*end)) - end++; - if (!end || end != body_end) { - ve_add(detail, "[\"body\"]", "invalid JSON", "json_invalid"); - cJSON_Delete(root); - return NULL; - } - if (!cJSON_IsObject(root)) { + if (!yyjson_is_obj(yyjson_doc_get_root(doc))) { ve_add(detail, "[\"body\"]", "body must be a JSON object", "type_error.object"); - cJSON_Delete(root); + yyjson_doc_free(doc); return NULL; } - return root; + return doc; } -bool cjson_is_integer(cJSON *item) { - if (!cJSON_IsNumber(item)) +bool json_is_integer(yyjson_val *item) { + if (!yyjson_is_num(item)) + return false; + double d = yyjson_get_num(item); + if (!(d >= (double)INT_MIN && d <= (double)INT_MAX)) return false; - double d = item->valuedouble; - int i = item->valueint; - return d >= (double)INT_MIN && d <= (double)INT_MAX && d == (double)i; + /* A real such as 5.0 counts as an integer; 5.5 does not. */ + return d == (double)(int)d; } -const char *encoding_from_root(cJSON *root, cJSON *detail, embedding_api_t api) { +const char *encoding_from_root(yyjson_val *root, yyjson_mut_doc *detail, embedding_api_t api) { const char *dflt = api == FFWD_API_PERPLEXITY ? "base64_int8" : "float"; - cJSON *encoding = cJSON_GetObjectItemCaseSensitive(root, "encoding_format"); + yyjson_val *encoding = yyjson_obj_get(root, "encoding_format"); if (!encoding) return dflt; - if (!cJSON_IsString(encoding) || !encoding->valuestring) { + const char *v = yyjson_get_str(encoding); + if (!v) { ve_add(detail, "[\"body\",\"encoding_format\"]", "encoding_format must be a string", "type_error.string"); return dflt; } - const char *v = encoding->valuestring; int ok = api == FFWD_API_PERPLEXITY ? (!strcmp(v, "base64_int8") || !strcmp(v, "base64_binary") || !strcmp(v, "float")) : (!strcmp(v, "float") || !strcmp(v, "base64")); @@ -171,8 +185,8 @@ const char *encoding_from_root(cJSON *root, cJSON *detail, embedding_api_t api) /* Parse the `text_type` hint. Returns 1 when the model's query instruction * should be prepended, 0 otherwise (document / absent). Models without a * published query instruction reject the field. */ -int text_type_is_query(cJSON *root, cJSON *detail, const char *query_instruct) { - cJSON *tt = cJSON_GetObjectItemCaseSensitive(root, "text_type"); +int text_type_is_query(yyjson_val *root, yyjson_mut_doc *detail, const char *query_instruct) { + yyjson_val *tt = yyjson_obj_get(root, "text_type"); if (!tt) return 0; if (!query_instruct) { @@ -180,11 +194,11 @@ int text_type_is_query(cJSON *root, cJSON *detail, const char *query_instruct) { "text_type is only supported for instruction embedding models", "extra_fields"); return 0; } - if (!cJSON_IsString(tt) || !tt->valuestring) { + const char *v = yyjson_get_str(tt); + if (!v) { ve_add(detail, "[\"body\",\"text_type\"]", "text_type must be a string", "type_error.string"); return 0; } - const char *v = tt->valuestring; if (!strcmp(v, "query")) return 1; if (!strcmp(v, "document")) @@ -194,15 +208,16 @@ int text_type_is_query(cJSON *root, cJSON *detail, const char *query_instruct) { return 0; } -int dimensions_from_root(cJSON *root, cJSON *detail, int min_dim, int max_dim, const char *encoding) { +int dimensions_from_root( + yyjson_val *root, yyjson_mut_doc *detail, int min_dim, int max_dim, const char *encoding) { int dims = max_dim; - cJSON *dimensions = cJSON_GetObjectItemCaseSensitive(root, "dimensions"); + yyjson_val *dimensions = yyjson_obj_get(root, "dimensions"); if (dimensions) { - if (!cjson_is_integer(dimensions)) { + if (!json_is_integer(dimensions)) { ve_add(detail, "[\"body\",\"dimensions\"]", "dimensions must be an integer", "type_error.integer"); } else { - dims = dimensions->valueint; + dims = (int)yyjson_get_num(dimensions); if (dims < min_dim || dims > max_dim) { char msg[96]; snprintf(msg, sizeof(msg), "dimensions must be between %d and %d", min_dim, max_dim); diff --git a/tools/server/models.c b/tools/server/models.c index 8fdc590..479ff97 100644 --- a/tools/server/models.c +++ b/tools/server/models.c @@ -89,25 +89,27 @@ static int read_query_instruct_from_files(const char *model_dir, char **out) { int rc = read_optional_text_file(model_dir, "config_sentence_transformers.json", &buf); if (rc <= 0) return rc; - cJSON *root = cJSON_Parse(buf); + yyjson_doc *doc = yyjson_read(buf, strlen(buf), 0); free(buf); - if (!root) { + if (!doc) { server_log("ffwd-server: invalid config_sentence_transformers.json: %s", model_dir); return -1; } - cJSON *prompts = cJSON_GetObjectItemCaseSensitive(root, "prompts"); - cJSON *query = prompts ? cJSON_GetObjectItemCaseSensitive(prompts, "query") : NULL; - if (query && !cJSON_IsString(query)) { - cJSON_Delete(root); + yyjson_val *root = yyjson_doc_get_root(doc); + yyjson_val *prompts = yyjson_obj_get(root, "prompts"); + yyjson_val *query = prompts ? yyjson_obj_get(prompts, "query") : NULL; + if (query && !yyjson_is_str(query)) { + yyjson_doc_free(doc); server_log("ffwd-server: prompts.query must be a string: %s", model_dir); return -1; } - if (query && query->valuestring && query->valuestring[0]) { - *out = xstrdup(query->valuestring); - cJSON_Delete(root); + const char *qs = yyjson_get_str(query); + if (qs && qs[0]) { + *out = xstrdup(qs); + yyjson_doc_free(doc); return *out ? 1 : -1; } - cJSON_Delete(root); + yyjson_doc_free(doc); return 0; } @@ -156,7 +158,7 @@ void embedding_request_free(embedding_request *r) { free(r->tokens); free(r->inputs); if (r->root) - cJSON_Delete(r->root); + yyjson_doc_free(r->root); memset(r, 0, sizeof(*r)); } @@ -169,7 +171,7 @@ void contextual_request_free(contextual_request *r) { } free(r->docs); if (r->root) - cJSON_Delete(r->root); + yyjson_doc_free(r->root); memset(r, 0, sizeof(*r)); } @@ -189,7 +191,7 @@ void rerank_request_free(rerank_request *r) { late_tokens_free(&r->documents[i]); free(r->documents); if (r->root) - cJSON_Delete(r->root); + yyjson_doc_free(r->root); memset(r, 0, sizeof(*r)); } diff --git a/tools/server/schedule.c b/tools/server/schedule.c index 9fb7db6..76da240 100644 --- a/tools/server/schedule.c +++ b/tools/server/schedule.c @@ -298,22 +298,22 @@ static int has_unfinished_jobs(const int *done, int n_jobs) { return 0; } -static int parse_job_root(job *j, cJSON **out_root) { - cJSON *detail = cJSON_CreateArray(); - cJSON *root = parse_json_body(j, detail); +static int parse_job_root(job *j, yyjson_doc **out_root) { + yyjson_mut_doc *detail = ve_new(); + yyjson_doc *root = parse_json_body(j, detail); if (!root) { job_set_422(j, detail); - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); *out_root = NULL; return -1; } - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); *out_root = root; return 0; } static void process_unknown_job(job *j) { - cJSON *root = NULL; + yyjson_doc *root = NULL; if (!j->started_ns) j->started_ns = nstime(); uint64_t t0 = nstime(); @@ -322,7 +322,7 @@ static void process_unknown_job(job *j) { if (rc != 0) return; job_set_error(j, 404, "unknown endpoint", "invalid_request_error"); - cJSON_Delete(root); + yyjson_doc_free(root); } /* Parse and tokenize one job into a heap request attached to j->prep, which the @@ -344,7 +344,7 @@ void tokenize_job(http_server *s, job *j) { else return; - cJSON *root = NULL; + yyjson_doc *root = NULL; uint64_t t0 = nstime(); int parse_rc = parse_job_root(j, &root); j->parse_ns += nstime() - t0; diff --git a/tools/server/server.c b/tools/server/server.c index f18d264..40b8e18 100644 --- a/tools/server/server.c +++ b/tools/server/server.c @@ -35,8 +35,6 @@ # include <sys/sysctl.h> #endif -#include <cjson/cJSON.h> - static volatile sig_atomic_t g_stop_requested = 0; static volatile sig_atomic_t g_signal_wfd = -1; diff --git a/tools/server/server_internal.h b/tools/server/server_internal.h index 315acdd..462af53 100644 --- a/tools/server/server_internal.h +++ b/tools/server/server_internal.h @@ -13,7 +13,7 @@ #include "ae.h" -#include <cjson/cJSON.h> +#include <yyjson.h> #include <pthread.h> #include <stdbool.h> #include <stddef.h> @@ -197,7 +197,7 @@ struct job { }; /* Parsed request state, shared by the prepare (handlers), execute (schedule), - * and render (encode) stages. Each carries its source cJSON root (freed when + * and render (encode) stages. Each carries its source JSON document (freed when * the request is freed), the resolved model, and the tokenized inputs. */ typedef struct { int *ids; @@ -206,7 +206,7 @@ typedef struct { typedef struct { job *j; - cJSON *root; + yyjson_doc *root; loaded_model *model; int dims; const char *encoding; @@ -226,7 +226,7 @@ typedef struct { typedef struct { job *j; - cJSON *root; + yyjson_doc *root; loaded_model *model; int dims; const char *encoding; @@ -242,7 +242,7 @@ typedef ffwd_late_tokens_t late_tokens; typedef struct { job *j; - cJSON *root; + yyjson_doc *root; loaded_model *model; late_tokens query; late_tokens *documents; @@ -258,13 +258,20 @@ typedef struct { void append_json_string(sbuf *b, const char *s); char *json_error_body(const char *message, const char *type, size_t *len); void job_set_error(job *j, int status, const char *message, const char *type); -bool ve_add(cJSON *detail, const char *loc_json, const char *msg, const char *type); -void job_set_422(job *j, cJSON *detail); -cJSON *parse_json_body(job *j, cJSON *detail); -bool cjson_is_integer(cJSON *item); -const char *encoding_from_root(cJSON *root, cJSON *detail, embedding_api_t api); -int text_type_is_query(cJSON *root, cJSON *detail, const char *query_instruct); -int dimensions_from_root(cJSON *root, cJSON *detail, int min_dim, int max_dim, const char *encoding); +/* Validation errors accumulate in a mutable yyjson document whose root is the + * {"detail":[...]} array: ve_new creates it, ve_add appends one entry, ve_count + * reports how many were added, and job_set_422 serializes it. Release the + * document with yyjson_mut_doc_free. */ +yyjson_mut_doc *ve_new(void); +bool ve_add(yyjson_mut_doc *detail, const char *loc_json, const char *msg, const char *type); +size_t ve_count(yyjson_mut_doc *detail); +void job_set_422(job *j, yyjson_mut_doc *detail); +yyjson_doc *parse_json_body(job *j, yyjson_mut_doc *detail); +bool json_is_integer(yyjson_val *item); +const char *encoding_from_root(yyjson_val *root, yyjson_mut_doc *detail, embedding_api_t api); +int text_type_is_query(yyjson_val *root, yyjson_mut_doc *detail, const char *query_instruct); +int dimensions_from_root( + yyjson_val *root, yyjson_mut_doc *detail, int min_dim, int max_dim, const char *encoding); /* ---- server_http.c: connection lifecycle, header parse, response framing ---- */ int set_nonblock(int fd); @@ -323,13 +330,16 @@ int load_one_model(http_server *s, const ffwd_server_model_spec_t *spec); void free_models(http_server *s); /* ---- server_handlers.c: per-endpoint prepare / batch-group / execute ---- */ -void prepare_embedding_request(job *j, cJSON *root, http_server *s, embedding_request *out); +void prepare_embedding_request(job *j, yyjson_doc *root_doc, http_server *s, embedding_request *out); int embedding_request_compatible(const embedding_request *a, const embedding_request *b); void execute_embedding_request_list(embedding_request **reqs, int n_reqs); -void prepare_contextual_request(job *j, cJSON *root, http_server *s, contextual_request *out); +void prepare_contextual_request(job *j, + yyjson_doc *root_doc, + http_server *s, + contextual_request *out); int contextual_request_compatible(const contextual_request *a, const contextual_request *b); void execute_contextual_request_list(contextual_request **reqs, int n_reqs); -void prepare_rerank_request(job *j, cJSON *root, http_server *s, rerank_request *out); +void prepare_rerank_request(job *j, yyjson_doc *root_doc, http_server *s, rerank_request *out); void execute_rerank_request(rerank_request *r); /* ---- server_schedule.c: job queue, micro-batching, completion ---- */ diff --git a/tools/server/test_json.c b/tools/server/test_json.c index 76a0f45..22b95c8 100644 --- a/tools/server/test_json.c +++ b/tools/server/test_json.c @@ -11,6 +11,10 @@ static const char *TEST_QWEN3_QUERY_INSTRUCT = "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery:"; +/* Parse a JSON literal into an immutable document so its root can be passed to + * the read-only validators (encoding_from_root, text_type_is_query). */ +static yyjson_doc *parse_obj(const char *json) { return yyjson_read(json, strlen(json), 0); } + static void test_json_error_body_escaping(void) { size_t len = 0; char *body = json_error_body("line\nquote\" tab\t ctrl\001 slash\\", "bad\ntype", &len); @@ -20,24 +24,26 @@ static void test_json_error_body_escaping(void) { TEST_ASSERT(len == strlen(body)); TEST_ASSERT(strchr(body, '\n') == NULL); TEST_ASSERT(strchr(body, '\t') == NULL); - cJSON *root = cJSON_Parse(body); + yyjson_doc *doc = yyjson_read(body, strlen(body), 0); + yyjson_val *root = yyjson_doc_get_root(doc); TEST_ASSERT(root != NULL); if (!root) { + yyjson_doc_free(doc); free(body); return; } - cJSON *err = cJSON_GetObjectItemCaseSensitive(root, "error"); - cJSON *msg = err ? cJSON_GetObjectItemCaseSensitive(err, "message") : NULL; - cJSON *type = err ? cJSON_GetObjectItemCaseSensitive(err, "type") : NULL; - if (!msg || !type || !cJSON_IsString(msg) || !cJSON_IsString(type)) { + yyjson_val *err = yyjson_obj_get(root, "error"); + yyjson_val *msg = err ? yyjson_obj_get(err, "message") : NULL; + yyjson_val *type = err ? yyjson_obj_get(err, "type") : NULL; + if (!msg || !type || !yyjson_is_str(msg) || !yyjson_is_str(type)) { TEST_ASSERT(0 && "error body fields must be strings"); - cJSON_Delete(root); + yyjson_doc_free(doc); free(body); return; } - TEST_ASSERT(strcmp(msg->valuestring, "line\nquote\" tab\t ctrl\001 slash\\") == 0); - TEST_ASSERT(strcmp(type->valuestring, "bad\ntype") == 0); - cJSON_Delete(root); + TEST_ASSERT(strcmp(yyjson_get_str(msg), "line\nquote\" tab\t ctrl\001 slash\\") == 0); + TEST_ASSERT(strcmp(yyjson_get_str(type), "bad\ntype") == 0); + yyjson_doc_free(doc); free(body); } @@ -48,7 +54,7 @@ static void test_job_set_422_replaces_response(void) { j.response_len = strlen(j.response); j.status = 500; - cJSON *detail = cJSON_CreateArray(); + yyjson_mut_doc *detail = ve_new(); ve_add(detail, "[\"body\",\"field\"]", "bad value", "value_error"); job_set_422(&j, detail); TEST_ASSERT(j.status == 422); @@ -58,80 +64,75 @@ static void test_job_set_422_replaces_response(void) { TEST_ASSERT(strstr(j.response, "bad value") != NULL); } - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); free(j.response); } static void test_encoding_from_root_family(void) { - cJSON *detail = cJSON_CreateArray(); + yyjson_mut_doc *detail = ve_new(); /* Default encoding follows the explicit serving API. */ - cJSON *empty = cJSON_CreateObject(); - TEST_ASSERT(strcmp(encoding_from_root(empty, detail, FFWD_API_OPENAI), "float") == 0); - TEST_ASSERT(strcmp(encoding_from_root(empty, detail, FFWD_API_PERPLEXITY), "base64_int8") == + yyjson_doc *empty = parse_obj("{}"); + yyjson_val *empty_root = yyjson_doc_get_root(empty); + TEST_ASSERT(strcmp(encoding_from_root(empty_root, detail, FFWD_API_OPENAI), "float") == 0); + TEST_ASSERT(strcmp(encoding_from_root(empty_root, detail, FFWD_API_PERPLEXITY), "base64_int8") == 0); - cJSON_Delete(empty); - TEST_ASSERT(cJSON_GetArraySize(detail) == 0); + yyjson_doc_free(empty); + TEST_ASSERT(ve_count(detail) == 0); /* OpenAI rejects int8 formats; Perplexity rejects the OpenAI base64. */ - cJSON *o = cJSON_CreateObject(); - cJSON_AddStringToObject(o, "encoding_format", "base64_int8"); - encoding_from_root(o, detail, FFWD_API_OPENAI); - TEST_ASSERT(cJSON_GetArraySize(detail) == 1); - cJSON_Delete(o); + yyjson_doc *o = parse_obj("{\"encoding_format\":\"base64_int8\"}"); + encoding_from_root(yyjson_doc_get_root(o), detail, FFWD_API_OPENAI); + TEST_ASSERT(ve_count(detail) == 1); + yyjson_doc_free(o); - cJSON *p = cJSON_CreateObject(); - cJSON_AddStringToObject(p, "encoding_format", "base64"); - encoding_from_root(p, detail, FFWD_API_PERPLEXITY); - TEST_ASSERT(cJSON_GetArraySize(detail) == 2); - cJSON_Delete(p); + yyjson_doc *p = parse_obj("{\"encoding_format\":\"base64\"}"); + encoding_from_root(yyjson_doc_get_root(p), detail, FFWD_API_PERPLEXITY); + TEST_ASSERT(ve_count(detail) == 2); + yyjson_doc_free(p); /* An unknown format is rejected. */ - cJSON *bad = cJSON_CreateObject(); - cJSON_AddStringToObject(bad, "encoding_format", "base64_fp16"); - encoding_from_root(bad, detail, FFWD_API_OPENAI); - TEST_ASSERT(cJSON_GetArraySize(detail) == 3); - cJSON_Delete(bad); + yyjson_doc *bad = parse_obj("{\"encoding_format\":\"base64_fp16\"}"); + encoding_from_root(yyjson_doc_get_root(bad), detail, FFWD_API_OPENAI); + TEST_ASSERT(ve_count(detail) == 3); + yyjson_doc_free(bad); - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); } static void test_text_type(void) { - cJSON *detail = cJSON_CreateArray(); + yyjson_mut_doc *detail = ve_new(); /* Absent: no-op for both model types, no error. */ - cJSON *empty = cJSON_CreateObject(); - TEST_ASSERT(text_type_is_query(empty, detail, TEST_QWEN3_QUERY_INSTRUCT) == 0); - TEST_ASSERT(text_type_is_query(empty, detail, NULL) == 0); - cJSON_Delete(empty); - TEST_ASSERT(cJSON_GetArraySize(detail) == 0); + yyjson_doc *empty = parse_obj("{}"); + yyjson_val *empty_root = yyjson_doc_get_root(empty); + TEST_ASSERT(text_type_is_query(empty_root, detail, TEST_QWEN3_QUERY_INSTRUCT) == 0); + TEST_ASSERT(text_type_is_query(empty_root, detail, NULL) == 0); + yyjson_doc_free(empty); + TEST_ASSERT(ve_count(detail) == 0); /* Qwen3: query -> 1, document -> 0, both accepted. */ - cJSON *q = cJSON_CreateObject(); - cJSON_AddStringToObject(q, "text_type", "query"); - TEST_ASSERT(text_type_is_query(q, detail, TEST_QWEN3_QUERY_INSTRUCT) == 1); - cJSON_Delete(q); - cJSON *d = cJSON_CreateObject(); - cJSON_AddStringToObject(d, "text_type", "document"); - TEST_ASSERT(text_type_is_query(d, detail, TEST_QWEN3_QUERY_INSTRUCT) == 0); - cJSON_Delete(d); - TEST_ASSERT(cJSON_GetArraySize(detail) == 0); + yyjson_doc *q = parse_obj("{\"text_type\":\"query\"}"); + TEST_ASSERT(text_type_is_query(yyjson_doc_get_root(q), detail, TEST_QWEN3_QUERY_INSTRUCT) == 1); + yyjson_doc_free(q); + yyjson_doc *d = parse_obj("{\"text_type\":\"document\"}"); + TEST_ASSERT(text_type_is_query(yyjson_doc_get_root(d), detail, TEST_QWEN3_QUERY_INSTRUCT) == 0); + yyjson_doc_free(d); + TEST_ASSERT(ve_count(detail) == 0); /* Invalid enum value on Qwen3 is reported. */ - cJSON *bad = cJSON_CreateObject(); - cJSON_AddStringToObject(bad, "text_type", "passage"); - TEST_ASSERT(text_type_is_query(bad, detail, TEST_QWEN3_QUERY_INSTRUCT) == 0); - TEST_ASSERT(cJSON_GetArraySize(detail) == 1); - cJSON_Delete(bad); + yyjson_doc *bad = parse_obj("{\"text_type\":\"passage\"}"); + TEST_ASSERT(text_type_is_query(yyjson_doc_get_root(bad), detail, TEST_QWEN3_QUERY_INSTRUCT) == 0); + TEST_ASSERT(ve_count(detail) == 1); + yyjson_doc_free(bad); /* text_type on a model with no query instruction is rejected. */ - cJSON *p = cJSON_CreateObject(); - cJSON_AddStringToObject(p, "text_type", "query"); - TEST_ASSERT(text_type_is_query(p, detail, NULL) == 0); - TEST_ASSERT(cJSON_GetArraySize(detail) == 2); - cJSON_Delete(p); + yyjson_doc *p = parse_obj("{\"text_type\":\"query\"}"); + TEST_ASSERT(text_type_is_query(yyjson_doc_get_root(p), detail, NULL) == 0); + TEST_ASSERT(ve_count(detail) == 2); + yyjson_doc_free(p); - cJSON_Delete(detail); + yyjson_mut_doc_free(detail); } int main(void) { |
