j*9dZddlmZddlZddlZddlmZmZmZm Z ddl m Z ddl m Z Gdd eZGd d ZGd d ZdS)uProtocol objects: guids, parentage, `__create__`, and method routing. This is the other half of `transport.py`. The transport moves messages; this decides what answers them. **THE MODEL, WHICH IS SMALLER THAN IT LOOKS.** The Playwright protocol is not RPC on a flat namespace: every message names a `guid`, and that guid is an OBJECT with a type, a parent and an initializer. The client mirrors that tree locally - `Connection._objects` - and builds a `ChannelOwner` for each one the moment it sees a `__create__`. So a server that answers correctly has to do three things and only three: 1. Hand out guids and remember what each one is. 2. Announce every new object with `__create__` BEFORE anything names it, parented to an object the client already knows. 3. Route `{guid, method, params}` to the object that owns it. ⛔ **ORDER IS PART OF THE PROTOCOL, NOT AN OPTIMISATION.** `__create__` for a child must reach the client before the reply that returns a channel to it, and before any event fired on it. `Connection.dispatch` looks the guid up in a plain dict and raises `Cannot find object` if it is not there yet - so an out-of-order create is not a race that usually works, it is a hard failure. Everything that creates an object therefore emits the create first and returns the channel after, in that order, from the same thread. ⛔ **AND A CHANNEL IS `{"guid": ...}`, NOT THE OBJECT.** `_connection.py` replaces guids with channels on the way in and channels with guids on the way out. A server that returns a bare guid string where a channel is expected does not fail at the seam: the client stores a string where a `ChannelOwner` should be, and the failure surfaces later, in a property access far from the cause. ) annotationsN)AnyCallableDictOptional) perimeter)TargetClosedErrorceZdZdZdS)ProtocolExceptionuA failure the client should see as a protocol error, with its reason. ⛔ Prefer this over a bare `Exception` for anything the caller could have caused: the message travels all the way to the user, and "Element is not visible" is a different day than "TypeError". N)__name__ __module__ __qualname____doc__YC:\Projects\Historical\venv\Lib\site-packages\invisible_playwright/_juggler/dispatcher.pyr r .srr cheZdZUdZdZiZded< dddZeddZ dddZ dd dZ d!dZ dS)" Dispatchera+One protocol object. A subclass declares its type and the methods it answers. Methods are looked up by name in `METHODS`, which is explicit on purpose: routing by `getattr(self, method)` would expose every attribute of the object to anything that can put a string in a message. UnknownDict[str, str]METHODSNserver'Server'parentOptional['Dispatcher'] initializerOptional[Dict]guid Optional[str]returnNonec||_||_|pi|_|p||j|_d|_||||dS)NF) rrrnew_guidTYPErdisposedregisterannounce)selfrrrrs r__init__zDispatcher.__init__Gsj  &,"6FOODI66  rcd|jiS)zAHow this object is referred to inside a result or an initializer.r)rr)s rchannelzDispatcher.channelTs ""rmethodstrparamscP|j|j||piddS)z%Fire a protocol event on this object.rr.r0N)rsend_uprr)r.r0s remitzDispatcher.emitYs= TY&'-|55 6 6 6 6 6rreasonc,|jrdSd|_|rd|ini}|j|jd|d|j||j|D]#}d|_|j|$dS)uTell the client this object is gone, and forget its whole subtree. ⛔ ONE MESSAGE, MANY OBJECTS, AND THAT ASYMMETRY IS THE PROTOCOL. The client drops the children of a disposed object by itself - a `__dispose__` per child would be traffic saying what the guid tree already says, and the driver does not send it either: measured 2026-08-28, it disposes Page, BrowserContext, Browser, ElementHandle and APIRequestContext, and never a Frame. So the children are dropped HERE, silently, and the wire stays identical. ⛔ FORGETTING THEM IS NOT OPTIONAL. Without the cascade this server's guid registry kept every frame of every closed page: measured at one `FrameDispatcher` per page, for the life of the browser. Nothing crashed, which is why it outlived the subscriber leak it was found beside - and why `test_a_long_session_does_not_accumulate_subscribers` now asserts on both registries rather than the noisy one. NTr6 __dispose__r2)r&rr3r unregisterdescendants_of)r)r6r0childs rdisposezDispatcher.dispose^s$ =  F '-5(F##2 TY-'-// 0 0 0 t$$$[//55 * *E!EN K " "5 ) ) ) ) * *rrrc|j|}|3tj|}t |p|jd|d|dt |||S)Nz has no method z, and z| is INSIDE the perimeter this package covers - so this is a gap, not a decision. See 32-stacco-da-playwright.md section 6.5.)rgetr refusalr r%getattr)r)r.r0nameoutsides rcallzDispatcher.call|s|'' <  '//G#0IIIvvvvvv/11 1 #wtT""6***r)NN) rrrrrrrr r!r")r!rN)r.r/r0rr!r")r6r r!r"r.r/r0rr!r) r rrrr%r__annotations__r*propertyr-r5r<rCrrrrr7s D!G    04'+     ###X#66666 *****<++++++rrceZdZdZd*dZd+dZd,d Zd-dZd.dZd/dZ d.dZ d0dZ d1dZ d2dZ dZd2dZd3dZd2d Zd4d#Zd*d$Zd5d%Zd6d(Zd)S)7Serverz;The guid registry, and the entry point the transport calls.r!r"ci|_i|_i|_tj|_d|_d|_g|_d|_ d|_ dS)N) _objects _disposed _counters threadingLock_lock _transport root_guid _on_shutdown_twins_deliver_eventr,s rr*zServer.__init__sY/1 +-57^%% #"$15 $(r transportrc||_dSrD)rR)r)rWs rattachz Server.attachs #robjectsDict[str, Any] deliver_eventc"||_||_dS)zGive this server a live view of the client's guid -> object registry, and its event-delivery entry point. See `InProcessTransport.bind_impl_objects`, which calls this - the docstring for WHY lives there, next to its one caller.N)rUrV)r)rZr\s r bind_twinszServer.bind_twinss  +rrr/c:|jpi|S)zThe live impl-side object for a guid this server already created, or None if nothing is bound yet or the guid is unknown.)rUr>)r)rs rtwinz Server.twins !r&&t,,,rfnargsc"|j|g|RdS)zCall `fn(*args)` as an event, under the client's own rules for what that means (see `Connection.deliver_event`). MUST be called from the asyncio loop thread - use `call_soon` to get there first.N)rVr)rarbs rdeliverzServer.delivers% B&&&&&&&rc(|jj|g|RS)zA fused dispatcher's way to run a blocking call off the loop. Delegates to the transport, which owns the worker pool; see `InProcessTransport.run_blocking` for why this exists and what it guards against.)rR run_blockingrds rrgzServer.run_blockings! ,t+B66666rc,|jj|g|RdS)zA fused dispatcher's way to hand an EVENT to the asyncio loop from the connection's read-loop thread. See `InProcessTransport.call_soon`.N)rR call_soonrds rrizServer.call_soons' "!",t,,,,,,rmessagercL|j|j|dSdSrD)rR emit_message)r)rjs rr3zServer.send_ups. ? & O ( ( 1 1 1 1 1 ' &r type_namec|j5|j|tjd}d|t |fzcdddS#1swxYwYdS)Nrz%s@%d)rQrN setdefault itertoolscountlowernext)r)rmcounters rr$zServer.new_guids Z @ @n// 9?1;M;MNNGioo//g?? @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @sAA((A,/A,objrcd|j5||j|j<ddddS#1swxYwYdSrD)rQrLrr)rus rr'zServer.registers| Z * *&)DM#( # * * * * * * * * * * * * * * * * * *s %))ic|j5|j|jdd|j|j<t |j|jkrV|jtt|jt |j|jkVddddS#1swxYwYdSrD) rQrLpoprrMlenDISPOSED_MEMORYrsiterrws rr9zServer.unregisters Z ? ? M  ch - - -'+DN38 $dn%%(<<<""4T^(<(<#=#=>>>dn%%(<<< ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?sB#B88B<?B<listc|j5t|j}dddn #1swxYwYg}|D]}|j}t }|gt ||vrV||ur||n<|t ||j}|t ||vV|S)uEverything registered whose parent chain reaches `obj`. ⛔ WALKED RATHER THAN INDEXED, on purpose. A children list on every dispatcher is a second place that knows the tree, and the tree already lives in `parent`; keeping the two in step is precisely the kind of duplication that lets one of them go stale. The walk is over the live registry, which this cascade is what keeps small. N) rQr}rLvaluesrsetidappendadd)r)ru everythingout candidatewalkerseens rr:zServer.descendants_ofsZ 6 6dm224455J 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6# ' 'I%F55D$F4)?)?S==JJy)))F$$$ $F4)?)?  s ';??c|j |jjn|j}||d|j|j|jdddS)zZEmit `__create__` for a new object, parented where the client can already find it.N __create__)typerrr2)rrrSr3r%r)r)ru parent_guids rr(zServer.announcesb*-)?cjooT^  ""x&)o77       rhookCallable[[], None]c:|j|dSrD)rTrr)rs r on_shutdownzServer.on_shutdown s   &&&&&rct|jD]} |#t$rYwxYw|jdSrD)reversedrT Exceptionclearrs rshutdownzServer.shutdownsmT.//  D       !!!!!s # 00cP|d}|d}|dpi}||jkr|||S|j5|j|}dddn #1swxYwY|tt j|}|rt||j5||jv}dddn #1swxYwY|rtdtd|d|d| ||S)Nrr.r0z/Target page, context or browser has been closedz no object z to answer z: it was never created) r>rS handle_rootrQrLr r?r rMr rC)r)rjrr.r0rurBgones rhandlez Server.handles{{6""X&&X&&," 4> ! !##FF33 3 Z * *-##D))C * * * * * * * * * * * * * * * ;  '//G 1'000 . .t~- . . . . . . . . . . . . . . . G'EGGG##44!"" "xx'''s$*BBB C  C$'C$r.r0c&td|z)Nzthe root has no method %r)r r4s rrzServer.handle_root6s ;f DEEErN)r!r")rWrr!r")rZr[r\rr!r")rr/r!r)rarrbrr!r")rarrbrr!r)rjrr!r")rmr/r!r/)rurr!r")rurr!r})rrr!r")rjrr!rrE)r rrrr*rYr^r`rergrir3r$r'r{r9r:r(rrrrrrrrIrIsEE((((<$$$$,,,,---- '''' 7777---- 2222@@@@ **** O????.     '''' " " " "((((6FFFFFFrrI)r __future__rrprOtypingrrrrrKr connectionr rr rrIrrrrs&>#"""""000000000000*)))))  T+T+T+T+T+T+T+T+niFiFiFiFiFiFiFiFiFiFr