Function utf8_stream_safe_len

ds4_server.c:1013–1035

Tokenizers can split a multi-byte UTF-8 character across two tokens. If an * SSE delta ends at that boundary, some clients replace the incomplete byte * sequence with U+FFFD and later send the corrupted text back, destroying KV * cache prefix matches. Hold only the trailing incomplete character; the next * generated token will complete it. */

static void request_free(request *r) {
    ds4_tokens_free(&r->prompt);
    for (size_t i = 0; i < r->image_count; i++)
        ds4_vision_embedding_free(&r->images[i].embedding);
    free(r->images);
    free(r->image_markers);
    free(r->model);
    for (int i = 0; i < r->stops.len; i++) free(r->stops.v[i]);
    free(r->stops.v);
    free(r->raw_body);
    free(r->prompt_text);
    stop_list_clear(&r->responses_live_call_ids);
    free(r->responses_live_call_ids.v);
    free(r->responses_live_suffix_text);
    stop_list_clear(&r->anthropic_live_call_ids);
    free(r->anthropic_live_call_ids.v);
    free(r->anthropic_live_suffix_text);
    tool_schema_orders_free(&r->tool_orders);
    memset(r, 0, sizeof(*r));
}

static ds4_think_mode think_mode_from_enabled(bool enabled, ds4_think_mode effort) {
    if (!enabled || effort == DS4_THINK_NONE) return DS4_THINK_NONE;
    return effort;
}

static bool parse_reasoning_effort_name(const char *s, ds4_think_mode *out) {
    if (!s) return false;