Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
fix(cli): preserve pasted multiline prompts
Suliman Abdulrazzaq6ada7bdf850d
diff --git a/c/coli b/c/coliindex d7c24222..c0dd2624 100755--- a/c/coli+++ b/c/coli@@ -226,6 +226,79 @@ def hline(w): return f"{C.dgray}{'─'*w}{C.r}"# ---------- util ----------def term_w(): return min(shutil.get_terminal_size((80,20)).columns, 100)+def _read_pasted_lines(stream):+ """Collect complete lines already queued after an interactive paste.++ ``input()`` is line-oriented on libedit (including the FreeBSD build), so a+ pasted block can return its first line while the remaining lines stay in+ the terminal input queue. Read only data that is immediately ready; a+ normal Enter therefore remains a one-line prompt and typing for the next+ turn is not blocked here.+ """+ lines = []+ try:+ if sys.platform == "win32":+ import msvcrt+ pending = []+ while msvcrt.kbhit():+ ch = msvcrt.getwch()+ if ch in ("\r", "\n"):+ lines.append("".join(pending))+ pending = []+ elif ch not in ("\x00", "\xe0"):+ pending.append(ch)+ if pending:+ lines.append("".join(pending))+ return lines+ import select+ while select.select([stream], [], [], 0)[0]:+ line = stream.readline()+ if not line:+ break+ lines.append(line.rstrip("\r\n"))+ except (ImportError, OSError, ValueError):+ # Some stdin implementations (notably embedded consoles) cannot be+ # polled. Keeping the first line is safer than blocking the chat loop.+ pass+ return lines++def read_prompt(prompt=""):+ """Read one prompt, retaining additional lines from a pasted block."""+ first = input(prompt)+ if not TTY:+ return first+ return "\n".join([first, *_read_pasted_lines(sys.stdin)])++def prompt_box_lines(message, width):+ """Wrap a prompt for the terminal box without losing explicit newlines."""+ lines = []+ for logical in message.splitlines() or [""]:+ lines.extend(textwrap.wrap(logical, width=max(1, width),+ replace_whitespace=False,+ drop_whitespace=False) or [""])+ return lines++def prompt_input_rows(message, columns):+ """Return the number of terminal rows occupied by the readline input."""+ rows = 0+ for index, logical in enumerate(message.splitlines() or [""]):+ prefix = 6 if index == 0 else 0 # `` │ › `` before the first line+ rows += max(1, (prefix + len(logical) + columns - 1) // columns)+ return rows++def redraw_prompt_box(message, width):+ """Replace readline's raw echo with a wrapped, multiline prompt box."""+ clean = message.strip("\r\n")+ columns = shutil.get_terminal_size((80,20)).columns+ used = prompt_input_rows(clean, columns)+ sys.stdout.write(f"\x1b[{used}A\x1b[0J")+ inner = width - 3+ for index, line in enumerate(prompt_box_lines(clean, inner)):+ prefix = f"{C.teal}{C.b}›{C.r}" if index == 0 else " "+ print(f" {C.dgray}│{C.r} {prefix} {line}{' '*(inner-len(line))}{C.dgray}│{C.r}")+ print(f" {C.dgray}╰{'─'*width}╯{C.r}")+ return clean+def model_arch(model):try:with open(os.path.join(model, "config.json"), encoding="utf-8") as f:@@ -925,15 +998,16 @@ def chat_attached(a, base, model_id):while True:if TTY:print(f" {C.dgray}╭{'─'*w}╮{C.r}")- try: msg=input(f" {C.dgray}│{C.r} {C.teal}{C.b}›{C.r} ")+ try: msg=read_prompt(f" {C.dgray}│{C.r} {C.teal}{C.b}›{C.r} ")except EOFError: print(); break- print(f" {C.dgray}╰{'─'*w}╯{C.r}")+ try: msg=redraw_prompt_box(msg, w)+ except Exception: print(f" {C.dgray}╰{'─'*w}╯{C.r}")else:try: msg=input()except EOFError: break- msg=msg.strip()+ msg=msg.strip("\r\n")if msg in (":q",":quit","exit"): break- if not msg: continue+ if not msg.strip(): continueif msg==":reset": msgs=[]; print(f" {C.dim}✦ new conversation{C.r}\n"); continuemsgs.append({"role":"user","content":msg})body=json.dumps({"model":model_id,"messages":msgs,"stream":True,@@ -1107,32 +1181,20 @@ def cmd_chat(a):except Exception: passprint(f" {C.dim}type and press Enter · Ctrl-C stops the answer · :more continues · :reset clears memory · :q exits{C.r}\n")w=term_w()-4- def user_box(msg):- """ri-disegna il messaggio dentro una box che si ADATTA su piu' righe:- l'input grezzo (che sborda) viene cancellato e sostituito dal testo avvolto."""- cols=shutil.get_terminal_size((80,20)).columns- used=max(1, (6+len(msg)+cols-1)//cols) # righe occupate dall'input (" │ › "+msg)- sys.stdout.write(f"\x1b[{used}A\x1b[0J") # su di N righe e pulisci fino in fondo- inner=w-3 # spazio utile: " │ › "+testo+"│" = w+4 colonne- lines=textwrap.wrap(msg, inner) or [""]- for i,ln in enumerate(lines):- pre = f"{C.teal}{C.b}›{C.r}" if i==0 else " "- print(f" {C.dgray}│{C.r} {pre} {ln}{' '*(inner-len(ln))}{C.dgray}│{C.r}")- print(f" {C.dgray}╰{'─'*w}╯{C.r}")try:while True:if TTY:print(f" {C.dgray}╭{'─'*w}╮{C.r}")- try: msg=input(f" {C.dgray}│{C.r} {C.teal}{C.b}›{C.r} ")+ try: msg=read_prompt(f" {C.dgray}│{C.r} {C.teal}{C.b}›{C.r} ")except EOFError: print(); break- try: user_box(msg.strip())+ try: msg=redraw_prompt_box(msg, w)except Exception: print(f" {C.dgray}╰{'─'*w}╯{C.r}")else:try: msg=input()except EOFError: break- msg=msg.strip()+ msg=msg.strip("\r\n")if msg in (":q",":quit","exit"): break- if not msg: continue+ if not msg.strip(): continueif msg==":reset":p.stdin.write(b"\x02RESET\n"); p.stdin.flush()stream_turn(p, END, lambda b: None)diff --git a/c/tests/test_cli_output.py b/c/tests/test_cli_output.pyindex 9659e940..bd48d0d7 100644--- a/c/tests/test_cli_output.py+++ b/c/tests/test_cli_output.py@@ -63,6 +63,43 @@ class CliOutputLanguageTest(unittest.TestCase):self.assertIn("set COLI_MODEL or use --model", result.stderr)+class InteractivePromptTest(unittest.TestCase):+ """Pasted prompts stay intact and render predictably in the TUI box."""++ @classmethod+ def setUpClass(cls):+ loader = SourceFileLoader("coli_prompt_under_test", str(CLI))+ spec = importlib.util.spec_from_loader(loader.name, loader)+ cls.coli = importlib.util.module_from_spec(spec)+ loader.exec_module(cls.coli)++ def test_prompt_box_preserves_newlines_and_indentation(self):+ lines = self.coli.prompt_box_lines("int main() {\n return 0;\n}", 20)+ self.assertEqual(lines, ["int main() {", " return 0;", "}"])++ def test_prompt_box_wraps_long_lines_without_collapsing_whitespace(self):+ lines = self.coli.prompt_box_lines(" return a_long_name;", 10)+ self.assertEqual(lines, [" return", " a_long_na", "me;"])++ def test_prompt_input_rows_accounts_for_explicit_lines(self):+ self.assertEqual(self.coli.prompt_input_rows("one\ntwo", 80), 2)+ self.assertEqual(self.coli.prompt_input_rows("x" * 80, 80), 2)++ def test_read_prompt_collects_lines_already_queued_after_first(self):+ import io+ import select++ stream = io.StringIO("second\nthird\n")+ with mock.patch.object(self.coli, "TTY", True), \+ mock.patch.object(self.coli.sys, "platform", "freebsd"), \+ mock.patch("builtins.input", return_value="first"), \+ mock.patch.object(self.coli.sys, "stdin", stream), \+ mock.patch.object(select, "select", side_effect=(+ ([stream], [], []), ([stream], [], []), ([], [], []),+ )):+ self.assertEqual(self.coli.read_prompt(), "first\nsecond\nthird")++class ChatCapForwardingTest(unittest.TestCase):"""#379/#386 r2 (F9): `coli chat` on a non-glm model spawns openai_serveras its local server. An explicit --cap must ride along on that command