| 1 | # Terminal text encoding / 终端文本编码 |
| 2 | |
| 3 | ## Child-process locale / 子进程 locale |
| 4 | |
| 5 | When LANG, LC_ALL and LC_CTYPE are all empty or absent, Reasonix selects an |
| 6 | installed UTF-8 locale for child processes. It prefers C.UTF-8, then en_US.UTF-8, |
| 7 | then another installed UTF-8 locale. Discovery is bounded and cached per process. |
| 8 | If discovery fails or no UTF-8 locale is installed, the environment is retained. |
| 9 | Windows continues to use its existing encoding-specific process handling. |
| 10 | |
| 11 | 当 LANG、LC_ALL、LC_CTYPE 均为空或未设置时，Reasonix 为子进程选择已安装的 UTF-8 |
| 12 | locale，优先 C.UTF-8，其次 en_US.UTF-8，再选择其他已安装的 UTF-8 locale。 |
| 13 | 探测有时间限制，并在进程内缓存。探测失败或没有可用 UTF-8 locale 时，保留原环境。 |
| 14 | Windows 继续使用既有的编码处理。 |
| 15 | |
| 16 | Explicit values, including C/POSIX and category overrides, remain unchanged. |
| 17 | Reasonix does not change the parent process environment or force LC_ALL. Thus an |
| 18 | explicit LC_ALL=C still requests byte-oriented sed/grep/sort semantics. |
| 19 | |
| 20 | 用户显式设置的值（包括 C/POSIX 和各分类覆盖）保持不变；不会修改父进程环境或强制 |
| 21 | 设置 LC_ALL。因此显式 LC_ALL=C 仍然要求 sed、grep、sort 等工具使用对应的字节语义。 |
| 22 | |
| 23 | ## Interactive input / 交互输入 |
| 24 | |
| 25 | Machine-submitted persistent-shell commands use ASCII-only byte escapes. The |
| 26 | integrated Bash terminal instead uses a private, temporary Readline configuration: |
| 27 | it includes the user's existing inputrc and then enables eight-bit input/output |
| 28 | without meta conversion. Other editing preferences remain in place. This accepts |
| 29 | Chinese and emoji even with an explicitly selected C locale. The temporary file |
| 30 | is removed on launch failure, terminal closure or process exit. |
| 31 | |
| 32 | 机器提交给持久 Shell 的命令使用纯 ASCII 字节转义。内置 Bash 终端则使用私有临时 |
| 33 | Readline 配置：先包含用户原有 inputrc，再开启八位输入输出并关闭 meta 转换，保留 |
| 34 | 其他编辑偏好。这样在显式 C locale 下也能接收中文和 emoji。临时文件会在启动失败、 |
| 35 | 终端关闭或进程退出时清理。 |
| 36 | |
| 37 | Long persistent-shell commands are sent as bounded ASCII blocks. Each block |
| 38 | must be acknowledged before the next is submitted; only the final step executes |
| 39 | the assembled command in the current shell. This avoids canonical terminal |
| 40 | input truncation while preserving variables, directory, exit status and stdin |
| 41 | isolation. Cancellation during staging retires the shell without executing the |
| 42 | partial command. |
| 43 | |
| 44 | 较长的持久 Shell 命令会拆成有长度上限的 ASCII 块，每块确认接收后才发送下一块， |
| 45 | 最后在当前 Shell 中执行完整命令。这样可避免终端规范模式截断输入，同时保留变量、 |
| 46 | 工作目录、退出码与标准输入隔离。分块传输期间取消操作会关闭该 Shell，不执行未完整 |
| 47 | 传输的命令。 |
| 48 | |
| 49 | ## Tool output / 工具输出 |
| 50 | |
| 51 | Shared live output buffers incomplete UTF-8 characters across reads before |
| 52 | emitting JSON events. Completion and cancellation flush an unfinished character |
| 53 | as one replacement character. Progress limits and retained output head/tail stop |
| 54 | at character boundaries; byte budgets are not increased. Durable protocol fields, |
| 55 | tool schemas and model prompt prefixes are unchanged. |
| 56 | |
| 57 | 共享实时输出层会缓存跨读取的不完整 UTF-8 字符，完整后才发送 JSON 事件。正常结束和 |
| 58 | 取消时，未完成字符以一个替代字符收尾。进度限额及最终保留的头尾内容都在字符边界 |
| 59 | 截断，不提高字节预算。持久协议字段、工具定义和模型提示前缀保持不变。 |
| 60 |