j(<dZddlmZddlZddlZddlZddlZddlZddlm Z m Z m Z ddl m Z ddlmZdd Zd d Zd!dZd"dZGddeZdZdZdZd#dZd$dZdS)%u` The seam where Node stops being necessary. ⛔ THIS IS THE KEYSTONE OF THE WHOLE DETACHMENT, and its shape is the one decision that makes the rest small. `_pw/_impl` is not replaced - 799 of its methods never touch the channel at all, and throwing them away would mean rewriting working logic for nothing. What is replaced is the ONE thing under them: `PipeTransport`, which spawns `node cli.js run-driver` and speaks the Playwright protocol over a pipe. The seam is `Transport`, an abstract class with five methods. Everything above it - channels, guids, `ChannelOwner`, the sync and async facades, the 823 generated API methods - keeps working unchanged, because from up there nothing can tell whether the answer crossed a pipe or came from the next object in the same process. **THE PROTOCOL, MEASURED AND NOT DEDUCED.** What crosses this seam is recorded by `scripts/capture_protocol.py` against the real Node driver, and the server is written against that recording. Reading `_impl` tells you which initializer fields are CONSUMED; it cannot tell you which are SENT, and the gap between those two sets is exactly where a future code path breaks. Four message shapes, in both directions: -> {"id": N, "guid": G, "method": M, "params": P, "metadata": {...}} <- {"id": N, "result": R} an answer <- {"id": N, "error": {"error": {...}}} a failure <- {"guid": G, "method": "__create__", "params": {"type": T, "guid": G2, "initializer": I}} <- {"guid": G, "method": M, "params": P} an event ⛔ **THREADS, AND WHY THIS IS NOT OPTIONAL.** `send()` is called ON THE ASYNCIO LOOP. Our Juggler operations block: `goto` waits for a load state, `click` waits for actionability. Running them inline would block the loop, and under the sync facade - which drives the loop from a greenlet - that is a deadlock, not a slowdown. So `send()` only enqueues; a small pool of worker threads runs the operations and posts every reply back with `loop.call_soon_threadsafe`, which is the only way to touch the loop from another thread. ⛔ **AND ORDERING IS NOT GUARANTEED, WHICH IS CORRECT.** Replies are matched by `id`, never by arrival order, so a slow `goto` cannot hold up a `title()` on another page. What must stay ordered is `__create__` before any message that names the guid it creates, and that is the dispatcher's job, not the pool's. ) annotationsN)AnyDictOptional) parse_error) Transportfailure BaseExceptionreturnDict[str, str]c t|jt|pt|jdt jt|||jdddS)zThe `{name, message, stack}` shape both an error REPLY and a directly-translated exception are built from - one function, so the two do not drift the way rule 16 forbids.i`N)namemessagestack)type__name__strjoin tracebackformat_exception __traceback__r s XC:\Projects\Historical\venv\Lib\site-packages\invisible_playwright/_juggler/transport.py_error_payloadr9ss W &w<<94==#93 MM7G$9;;<<.jobs PT  //OOOOO  < < <1'::  //0A60:<<<<<<<<< > F#Vn  "CM #rr r c\|dS||dt|iddS)u⛔ The shape matters: `Connection.dispatch` reads `msg["error"]["error"]["message"]`, and a flatter error is silently turned into a generic one, which is how a precise server-side reason becomes "Protocol error" by the time a user sees it.Nerror)rur)r~r)r?rr s rrzzInProcessTransport.reply_errorsM > F ~g667       r)r.r/r0rr r")rArBrCrr r")rGrrHrr r )rGrrHrr r"rP)rrr r")rrr!rr r")rrr r r r")r __module__ __qualname____doc__r[r3rFrRrVrbrerkrirprsr]r~ryrz __classcell__)r@s@rr,r,YsOMM G88882####J"$$$$#### : : : :2222 !!!! 6666*          rr,INVPW_TRANSPORTdriverjugglerrctjtptS)zDWhich transport this process asked for. Anything but `driver` is us.)osenvironrv CHOICE_ENVJUGGLERstriplowerrrchosenr3s3 JNN: & & 1' 8 8 : : @ @ B BBrr.r/'InProcessTransport'cttkrtdtzddlm}t ||S)z2The transport a `Connection` should speak through.athe Node driver was removed on 2026-08-28: this package no longer ships `_driver/` and no longer downloads node. What replaced it is the in-process Python server, which is now the only transport - unset %s and it runs. To get the old arm back for a COMPARISON (the only thing it was still for): check out the last commit that carried it into a git worktree and point INVPW_DRIVER_TREE at it. Both `judge_both_transports.py` and `diff_protocol.py` read that variable.) JugglerServer)rDRIVERrUrr0rr,)r.rs rmake_transportr8sa xx6 58B BCC C&%%%%% dMMOO 4 44r)r r r r )r r r r )rr r!rr r")rr r r r r")r r)r.r/r r)r __future__rasynciorr5r8rtypingrrr&invisible_playwright._pw._impl._helperr)invisible_playwright._pw._impl._transportrrrr&r*r,rrrrrrrrrsc**V#""""" &&&&&&&&&&>>>>>>??????    0000""""&&&& B B B B B B B B j  CCCC 5 5 5 5 5 5r