jش4dZddlmZddlZddlZddlZddlmZmZm Z m Z ddl m Z ddl Z ddlmZmZGdd eZd Zd ZGd d e Ziddddddddddddddddddd d!d"d#d$d%d&d'd(d)d*d+d,d-d.d/d0d1d2Zdmd8Zejd9Zd:Zed;d46!!!!!!!!!!!!rr )zhttps://api.ipify.orgzhttps://icanhazip.comzhttps://checkip.amazonaws.com) socks5:// socks4://socks://cDeZdZUdZded<ded<ded<ded<d d Zd S) _Attemptz/One endpoint tried, and what came back from it.rurlfloatsecondsrdetailrcJd|jdd|jdd|jd|jS)Nz z<32 z5.2fzs : )r*r,rr-rs rlinez _Attempt.lineEs6SDHSSS4<SSSSSdkSSSrN)rr)r r!r"r#__annotations__r2rrrr)r)=sZ99 HHHNNN IIIKKKTTTTTTrr) proxy_auth)z;the proxy refused the credentials - check username/passwordzhsomething on this network demanded proxy credentials, so a transparent proxy is intercepting the requestproxy_rejected)ztthe proxy answered but refused to open the tunnel - check the plan, the target, or an allowlist on the provider sidezhsomething on this network refused to open the tunnel, so a transparent proxy is intercepting the request socks_missing)>a socks:// proxy needs PySocks - pip install 'requests[socks]'r7 dns_failure)z9the proxy HOSTNAME does not resolve - check it for a typozzthe echo endpoint hostname does not resolve, so this host has no working DNS - not a proxy problem, there is no proxy hereconnect_failed)z4nothing accepted a connection at the proxy host:portzjno route to the echo endpoint from this host - the network is down or something is refusing the connection read_timeout)z9the proxy accepted the connection and then never answeredzthe geoip database knows the IP but carries no timezone for itrGno_coordinates_for_ip):the geoip database knows the IP but carries no coordinatesrIinvalid_timezone)Ythe geoip database returned a zone this system's tz database does not know - check tzdatarK)Aunrecognised failure - the repr on the line above is all there isrL)8the endpoints failed for DIFFERENT reasons, listed belowrM)rmixedrrproxiedboolrcbt|}|s|S|r|dn|dS)zThe advice for one class, for the path the request actually took. Falls back to the class name rather than inventing text: a class with no entry is a bug in the table, and saying its name is more honest than guessing what it means. rr )_REMEDYget)rrOpairs r_remedyrUs9 ;;t  D   *47747*rz#Tunnel connection failed:\s*(\d{3}))NameResolutionErrorzgetaddrinfo failedgaierrorc#>K|]}tj|VdSr) ipaddress ip_network).0ns r r]s=  !i*1--      r) z 10.0.0.0/8z 172.16.0.0/12z192.168.0.0/16z 127.0.0.0/8z169.254.0.0/16z 0.0.0.0/8z 100.64.0.0/10z::1/128zfc00::/7z fe80::/10z::/128z fec0::/10exc BaseException'tuple[str, str]'ct|jd|t}|r+|d}|dkrdd|dfSdd|dfSt |t jjrdvrd St |t jj r'tt|d d d d }d d|fSt |tr7t |t j sddt|d ddfStfdtDrdSt |t jjrddt|d ddfSt |t jjrdSt |t jjt jjt jjfrdSdt+|d dfS)aName the failure class behind one failed attempt. A PURE function - no network, no clock, no state, and no notion of whether a proxy is in the path - which is what lets the whole classification be checked against known-bad inputs without a proxy. The class names the MECHANISM; the caller pairs it with the path to produce advice, because the caller is the only place that already knows the path. Classified on MEASURED evidence rather than on the exception type, because the type does not separate the cases. Measured 2026-09-02: a closed proxy port, a proxy hostname that does not resolve, a blackholed proxy address AND a 407 all arrive as `requests.exceptions.ProxyError`, while the same closed port behind socks5:// arrives as `ConnectTimeout` instead. What distinguishes them lives in the chained cause, which reaches `str(exc)`. ORDER IS LOAD-BEARING in three places, and each is marked below. The general rule: any branch testing a SUBCLASS must precede the branch testing its base, and `requests.exceptions` has two inheritance edges that are easy to miss - `SSLError` and `ProxyError` both subclass `ConnectionError`, and `InvalidSchema` subclasses both `RequestException` and `ValueError`. r0r 407r4zproxy answered z to CONNECTr5SOCKS)r6zPySocks is not installedresponseN status_coder<zendpoint answered HTTP r>zreply was not an IP (F)c3 K|]}|vV dSrr)r[tokentexts rr]z_classify..s' 2 2U5D= 2 2 2 2 2 2r)r8za hostname did not resolver;z TLS failed ()r:z,connected, then no reply before the deadline)r9zcould not open a connectionrn)typer _TUNNEL_STATUSsearchgroup isinstancerequests exceptions InvalidSchema HTTPErrorgetattr ValueErrorRequestExceptionrany _DNS_TOKENSSSLError ReadTimeout ProxyErrorConnectTimeoutConnectionErrorrepr)r^statuscoderjs @r _classifyrs!,3ii ) )C ) )D  " "4 ( (F E||A 5==!D4!D!D!DD D!D4!D!D!DDD #x*899;goo::#x*455GwsJ55}dKK$&F&F&FFF#z""N:c8;T+U+UN#%MSXXcrc]%M%M%MMM 2 2 2 2k 2 2 222;:: #x*344=>??> d3iio %%rproxyOptional[Dict[str, str]]c|sdS|dpd}t|o|dkS)NFserverz direct://)rSstriprPlower)rrs r _proxy_is_setr/sO uii!!'R . . 0 0F << 9FLLNNk99rDict[str, str]cz|dpd}|}|ds|drd}n2|drd}n|drd }nd }d |vr|d d d n|}d |vrt d|d|dpd}|dpd}|r't |dd t |dd}nd}|d ||}||dS)aTranslate our proxy dict into a ``requests`` proxies mapping. SOCKS5 uses the ``socks5h`` scheme so DNS is resolved proxy-side (matches ``network.proxy.socks_remote_dns=True`` in the Firefox path). HTTP/HTTPS pass through unchanged. Credentials are URL-encoded. rrr%r'socks5hr&socks4zhttps://httpshttpz://r :z proxy server zC has no port. An endpoint needs host:port - e.g. socks5://host:1080usernamepassword)safe@)rr)rSrr startswithsplitrvr) rrlowscheme host_portuserpwdauthr*s r_proxies_for_requestsr6sii!!'R . . 0 0F ,,..C ~~k""cnnZ&@&@  $ $  # #-2f__ UA&&q))&I )  )F ) ) )** * 99Z &BD ))J   %2C 2&&&??s)<)<)<???  ) ) )i ) )C# & &&rg$@g.@)timeoutbudgetrr+rc |rt|nd}g}tj|z}tD]j}|tjz }|dkrnJtj} t ||} t j||| dz | dz f} | | j } tj | nc#t$rV} t| \} }|t|tj|z | |Yd} ~ d} ~ wwxYwt!fdt"Dr>|t|tj|z d| dh| cS||tjz z }t%|}t%t}t'|}d|D}|sd } n*t%|d kr|} nd } |rd nd }d|dt+| |d|d|d|dd|dd g}|d|D||kr|d||z dt/d|| t3|)aReturn the public egress IP. Routes the request through ``proxy`` when given (SOCKS support requires ``requests[socks]`` / PySocks); with ``proxy=None`` it makes a direct request that sees the host's own public IP. Tries each echo endpoint in turn; raises :class:`GeoTimezoneError` if none return a valid IP. ``timeout`` bounds ONE request; ``budget`` bounds the whole step. Both are needed, and having only the first is what made this the slowest thing in a launch: three endpoints tried in sequence at ten seconds each is a thirty-second worst case that nothing capped, and one launch in six spent 35s here. A per-request timeout says how long to wait for a server; it cannot say how long the caller is willing to wait in total. The remaining budget is now handed to each request, so a slow first endpoint shortens the second rather than adding to it, and the step returns or raises within ``budget`` however many endpoints the list grows to. Nr)proxiesrc3 K|]}|vV dSrr)r[netparseds rr]z%discover_egress_ip..s'66v}666666rr?z! is not a routable public addressch|] }|j Srrr[as r z%discover_egress_ip..s & & &QV & & &rr@r rNzthrough the proxyzdirectly (no proxy set)z!could not discover the egress IP r0ztried z of z endpoint(s) in z.1fzs of a gzs budgetc3>K|]}|VdSr)r2rs rr]z%discover_egress_ip..s*,,a,,,,,,rz NOTE: the budget ran out with z@ endpoint(s) never tried, so the redundancy never came into play r)rtime monotonic_IP_ECHO_ENDPOINTSminrqrSraise_for_statusrjrrY ip_address Exceptionrappendr)rx _NOT_ROUTABLElenrPpoprUextendr jointuple)rrrrrdeadliner* remainingstartedslice_respipr^rr-spenttriedtotalrOkindsthroughlinesrs @rdiscover_egress_ipr\s=./4=#E***G!#H~&(H!,,t~/// >> E."" ),,F<Wvz6A:.FD  ! ! # # #""B)"--FF   $S>>LD& OOdn..8$GG I I I HHHH   6666 666 6 6  OOHT^%%/888:: ; ; ;  h!1!11 2E MME " # #E7mmG & &X & & &E   # Uqyy{{%,K!!2KG PGOOwtW7M7MOOWWWEWW5WWWVWWWW E LL,,8,,,,,, u}}   =uu} = = = > > > 499U++$x Q Q QQs*A0C D;%A D66D;r mmdb_pathr'Optional[Dict[str, Any]]'c ddl} |t||j5}||}dddn #1swxYwYnW#t t t|dt f$r.}td|dtddd |d |d}~wwxYwt|tr|ndS) uThe ONE place that opens the database and reads a record. The three functions below - timezone, locale and coordinates - read the SAME record of the SAME IP, and before this function two of them repeated the same three lines. Adding the third would have made three copies of how the database is read, which is rule 16 broken while applying it. Returns ``None`` when the IP is not there: the caller decides whether that is fatal (the timezone, which behind a proxy must fail loudly) or not (the locale, which has a declared fallback). ⛔ THE MODE IS EXPLICIT, AND ``MODE_MMAP`` ON PURPOSE ([B227], 2026-09-25). The default picks the C extension, which on Windows opens the file with a path in BYTES and cannot find a name that is not ASCII. The cache lives under the user's profile, so every Windows user called José, Müller or with a CJK account name got ``FileNotFoundError`` on a database that exists, and behind a proxy that refused the launch outright. Measured on the same file: ``MODE_AUTO`` and ``MODE_MMAP_EXT`` fail, ``MODE_MMAP``, ``MODE_FILE`` and ``MODE_MEMORY`` answer. ``MODE_MMAP`` opens the file from Python and maps it like the extension does; the handful of lookups a launch makes do not notice the pure-Python reader. A database that is there but cannot be read is said as such (``geoip_unreadable``), not as the bare ``OSError``, which named a missing file and sent the reader looking for a download problem. rN)modeInvalidDatabaseErrorzthe geoip database at z could not be read - rCFr0r) maxminddb open_databaser MODE_MMAPrSOSErrorrvrur rUrpdict)rrrreaderrecordr^s r _geo_recordrsC6 .  $ $S^^):M $ N N $RXZZ^^F $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ Z I5z B B D... ;Y ; ;)511 ; ;58 ; ;#%%%+. ..  -- 76647s:)AA AAAAA&B1)B,,B1'tuple[float, float]'c@t||}|std|dd|dpi}|d|d}}||td |d d t|t|fS) uMap ``ip`` -> (latitude, longitude) from the same record as the timezone. ⛔ It is the ONE source of the declared position: the engine no longer asks the hardware anything (no WiFi, no GPS, no cell) and asks Google nothing. The position comes out of the proxy's exit IP, exactly as the timezone and the language do, so the three cannot contradict one another. ⛔ THE ACCURACY DOES NOT COME FROM HERE, and that is not an oversight: measured 2026-08-20 on a real record, ``location`` carries ``latitude``, ``longitude`` and ``time_zone`` and **not** ``accuracy_radius``. Declaring it is a separate decision and lives in the Profile: a position derived from an IP with GPS-grade accuracy would be incoherent by construction. Raises :class:`GeoTimezoneError` - the same class as the timezone, because it is the same failure - when the IP is missing or the record carries no coordinates. **No fallback is invented**: without a declaration the engine refuses, which is rule 7. egress IP " not present in the geoip databaserDrlocationlatitude longitudeNzno coordinates for egress IP  in the geoip databaserH)rr rSr+)rrrloclatlons rip_to_coordinatesrs&Y ' 'F ! ? ? ? ?!!! ! **Z &BCwwz""CGGK$8$8C {ck FB F F F(*** * ::uSzz !!rc ^t||}|std|dd|dpid}|std|dd d d lm}m} ||n/#|t f$r }td |d |d|d|d}~wwxYw|S)a#Map ``ip`` to its IANA timezone using the offline mmdb. Reads the standard MaxMind ``location.time_zone`` field and validates it against the system tz database. Raises :class:`GeoTimezoneError` if the IP is absent from the DB or the zone is missing / not a valid IANA name. rrrDrr time_zonezno timezone for egress IP rrFr)ZoneInfoZoneInfoNotFoundErrorz$geoip returned an invalid IANA zone z for r0rJN)rr rSzoneinforrrv)rrrtzrrr^s rip_to_timezoners*Y ' 'F ! ? ? ? ?!!! ! **Z &B + +K 8 8B ' C C C C%''' '98888888. !: .... I2 I Ib I IC I I#%%%*- .. Is2 A>> B* B%%B* Optional[str]ct||}|r,|dpidpdnd}|pdS)amThe ISO 3166 country of ``ip`` (MaxMind ``country.iso_code``), upper case, or None when the DB does not know the IP or the record has no code. Only the COUNTRY: which language a browser there reports is not a geo question but a Firefox one (which build people there install), and it is answered in one place, `_locale._COUNTRY_FIREFOX_BUILDS`. countryiso_coderN)rrSupper)rrrccs r ip_to_countryr/sWY ' 'FBH P6::i &B + +J 7 7 =2bB 88:: rc(eZdZUdZded<ded<dS)_EgresszWhere a session comes from: the address and its country, together, because the language decision needs both (the country picks the builds, the address picks one of them).rrrN)r r!r"r#r3rrrrr<s.'' GGGLLLLLrrT) may_discoverdiscovery_failure egress_iprrOptional[BaseException]Optional[_Egress]cddlm} |}| |rt|std}|/|t |jd|nd}t ||dSt||}|t |d|dSt||S#t$r2}t |t |jd|Yd}~dSd}~wwxYw)aEThe egress address and country, for ``locale="auto"``; None when the country cannot be known. PRIVATE, and only :func:`invisible_core._locale.decide_session_locale` calls it: the country is an INPUT to the decision, not a value any consumer may read. Until 36.x a tag took its place (the public `resolve_session_locale`), and every consumer that read the raw tag derived values the table then contradicted. Behind a proxy it reuses the already-discovered ``egress_ip`` (no extra round-trip); without a proxy it discovers the host's public IP when it was not handed one and ``may_discover`` allows it. On any failure it returns None and says so on stderr; the decision then takes en-US (never breaks launch - locale is cosmetic, unlike timezone which traps a foreign-proxy mismatch). r ensure_geoip_mmdbNr0zno egress IP was resolvedzthe geo DB has no country for ) downloadrrrrlr _warn_locale_fallbackrrr) rrrrrrwhyrr^s r_egress_countryrEs0&,+++++ :,:}U/C/C:#D))B :'3,--6MM:KMMM3  "% - - -4$5$5$7$788 ? !%)N")N)N O O O4r7### eS (:%C%Cc%C%CDDDttttts$AB-B B C&'CCrrcddl}t|rdnd}dd|dD}t d|d||j dS) a)Say that the locale was NOT resolved, on stderr, every time. This used to be two bare returns. The docstring called locale "cosmetic", and for a lone session it nearly is - but the timezone is resolved from the SAME egress IP and does not fall back, so a failure here produces a session whose timezone says one country and whose language says the United States. That pairing is a cross-field inconsistency of exactly the kind the timezone trap exists to prevent, and it was reaching users with no signal at all. The OUTCOME is deliberately unchanged: raising here would break launches that work today, on a field that is recoverable by passing `locale=` explicitly. What changes is that it stops being invisible - an absent signal must be loud, never silent. rNzbehind a proxyz with no proxyrc3 K|] }d|zV dS)z Nr)r[r2s rr]z(_warn_locale_fallback..s&AAv}AAAAAArz5invisible-core: could not resolve the session locale z; falling back to en-US. The timezone is still resolved from the egress IP, so this session may pair a non-US timezone with a US language - pass locale="xx-XX" to set it explicitly. )file)sysrrrprintstderr)rrrwherer-s rrrys JJJ -e 4 4 I  /E YYAA4AAA A AF         Z rcheZdZUdZded<ded<dZded<dZded <d Zd ed <dZd ed<ddZ dS) SessionGeouGeo facts resolved once per session from a single egress round-trip. ``timezone`` follows the precedence in the module docstring. ⛔ ``egress_ip`` IS A FACT, NOT A DECISION: the address this session really leaves from, discovered once, with or without a proxy. Whether that value is then DECLARED to the engine as an srflx is decided by :meth:`srflx_to_declare`, and without a proxy the answer is no. Until 2026-08-26 this field was ``None`` without a proxy, and the "no" was expressed by that very ``None``. One field for two meanings: the effect was that ``prepare_session_geo`` discovered the address for the timezone, **threw it away**, and ``resolve_session_locale`` had to discover it again. Two identical requests to an external service, from the real address, before the browser exists - where a real user makes zero. The fact is carried now, and the "no" lives where the decision already lived. rtimezonerrNzOptional[float]rrFrPsrflx_suppressedzOptional['SessionLocale']localerc"|jrdn|jS)uThe address the engine must announce as its srflx, or ``None``. ⛔ THE ONE PLACE where this question is answered. The two env builders - ``launch.build_launch_env`` in the core and ``_session.build_env`` in the wrapper - call it rather than recomputing it: they were already two landing points, and the same rule written twice could have drifted. N)r rr1s rsrflx_to_declarezSessionGeo.srflx_to_declares,@tt$.@r)rr) r r!r"r#r3rrr r r rrrrrs$MMM!%H$$$$!%I%%%%#"""")-F,,,,AAAAAArrc|sdSt|sdSddlm}|sdS ddlm}|||}n#t $rYdSwxYw|dduo|dduS) u.Declare a synthetic srflx, or let the real one through? The criterion comes from a real detector's code, read rather than deduced: its configuration contains no STUN at all, only TURN, and the allocation's username is the same identifier its verification POST sends to the backend. So in an honest browser **an srflx can only come from a successful allocation**, and the client-side candidate and the server-side proof coincide by construction. It is that implication a declared candidate breaks. Da qui la regola, che ha tre rami: * **no proxy**: the real STUN answers with the address traffic actually leaves from, so the srflx is born correct and with its own allocation. Nothing is declared. Until 2026-08-26 this branch was not written here: the same outcome fell out of ``egress_ip`` being ``None`` without a proxy - the decision was encoded in the ABSENCE of a fact. The two are separated now, and this branch says out loud what used to be obtained as a side effect. * **UDP demonstrated and coherent** (UDP leaves from the same address as TCP): the real srflx will already be born with the right address. Nothing is declared: declaring would add a candidate with no matching allocation. * **everything else**: the exit IP is declared, which is the behaviour it has always had. Without an srflx the detector writes *"Javascript is manipulated"*, which accuses the browser; with one it writes *"VPN/PROXY detected"*, which accuses the network. The difference between those two sentences is this line. ⛔ AND THE PROBE CANNOT FAIL A LAUNCH. Whatever goes wrong - network, a timeout, a field that is not there - falls back to the cautious branch. A capability is only exploited once it is DEMONSTRATED. FTr )UDP_GOES_THROUGH_SOCKS) capability)known_tcp_exitudpudp_matches_tcp)r_proxyr _capabilityrrrS)rrrrcs r_srflx_suppressedrsB u   t/..... !u++++++ JuY 7 7 7 uu 55<<4  DAEE*;$<$<$DDs5 AAcddlm} |S#t$r:}|rd|dnd|d}t|t d|d|d|d }~wwxYw) a4The mmdb, or an error saying the DATABASE failed and not the network. The distinction this exists to make: acquiring the database and looking an address up in it are separate failures with separate remedies, and until 2026-09-02 the acquisition sat inside the same expression as the lookup, under one blanket `except`. A failed download therefore reached the caller wearing the face of a proxy that would not answer, and the reflex it invited was to go and check the proxy, which was working. Both TIMEZONE resolvers go through here rather than keeping a copy each. `_coordinate` in `prepare_session_geo` still acquires the database on its own line, and that is deliberate: coordinates are best-effort and swallow their own failure, so routing them through a helper that RAISES would turn a tolerated gap into a refused launch. `proxied` selects the advice, and is passed rather than recomputed: the caller has already decided it with `_proxy_is_set`, and deciding it twice is the duplication that rule 16 names. r rzthe egress IP is z#, so the network path is fine, but z, but rBr0rN)rrrr rU)rrOrr^founds r_geoip_databasers(,+++++/  """ ///@LRLLLL!?R!?!?!?  Dg17;; D Ds D D$&&&+. //s  A5AAautorr c  ddlm |pd}t|}d}d}|r* t |}n#t $r }|}Yd}~nd}~wwxYwd fd }|rZ|d krB||\} } t||| | t||t|||d | Sd} |r|nt d} | |ptd || \} } t| t| |} n>#t $r1}|rtddt|| |d| cYd}~Sd}~wwxYwt| | | | t|| t|| |dS)aResolve the session timezone, egress IP AND language in ONE round-trip. The egress IP is discovered once and reused for the timezone mapping (when ``timezone`` is ``""``/``"auto"``), the WebRTC public-IP override and the language. Timezone precedence is identical to :func:`resolve_session_timezone`; the egress IP is best-effort for the WebRTC side (a discovery failure that the timezone path doesn't need won't break the launch - but if the timezone path *does* need it behind a proxy, that path still fails loudly). ``locale`` is what the caller asked for, a tag or "auto", and the returned ``SessionGeo.locale`` is the decision (`decide_session_locale`). It is decided HERE so that a consumer makes one call and gets the three facts that must agree with one another, instead of repeating an ``if locale == "auto"`` branch of its own: until 36.x every consumer did, five copies across four packages. r rrNr'Optional[str]'r)'tuple[Optional[float], Optional[float]]'c`|sdS t|S#t$rYdSwxYw)aCoordinates are BEST-EFFORT, the zone is not, and the gap is meant. A wrong timezone behind a proxy is the `tz_mismatch` trap and has to fail the launch. A MISSING position, on the other hand, is not a contradiction: it is a browser nobody has asked where it is yet, and the engine refuses it cleanly. Failing a launch over that would be more brittle without being more faithful. )NN)rr)rrs r _coordinatez(prepare_session_geo.._coordinate]sS : $R):):)<)<== =   :: s  --rT)rrrrzegress IP discovery failedF)r )rrr)rrrr) rrrrrrrrrr r rr)rrr r proxy_setr egress_errr^rrrrzoners @rprepare_session_geor#9sO(,+++++ .b   ! !Be$$I $I&*J *511II   JJJJJJ   TbhhjjF"";y))S"ic+E9==!&IU/3zSSSTT TB$# AYY);D)A)A :N 01M N N N;r??Sb/"i"@"@AA $$$   "d7 bE!,#,#,#$$$ $ $ $ $ $ $ $ dBS*;E2*F*Ff%eTTT V VVs6A AAAAD E&E E Ec|pd}|r|dkr|St|} t|r|nd}t |t ||S#t $r|rYdSwxYw)aMap the user's ``timezone`` setting to a concrete IANA zone (or ``""``). Timezone-only path (no WebRTC side effects): an explicit IANA zone wins and triggers NO network call; ``""``/``"auto"`` resolve from the egress IP. The launch path uses :func:`prepare_session_geo` instead (which additionally returns the egress IP for WebRTC); this standalone resolver is kept for third-party integrations that only want the zone. See the module docstring for the precedence table. rrN)rrrrrrr)rrrr rs rresolve_session_timezoner%s .b   ! !B bhhjjF"" e$$I  < = =b/"i"@"@AAA    rrs0A44 BB)rrrOrPrr)r^r_rr`)rrrrP)rrrrr)rrrr+rr+rr)rrrrrr)rrrrrr)rrrrrr)rrrrrr) rrrrrrPrrrr)rrrrrr)rrrrrrP)rrrOrPrr)r)rrrrr rrr)rrrrrr)-r# __future__rrYrertypingrrrr urllib.parserrq_localer r RuntimeErrorr r_SOCKS_SCHEMESr)rRrUcompilermryrrrrrrrrrrrrrrrrr#r%rrrr.s*#""""" 222222222222++++++++ ! ! ! ! !| ! ! !" 8 T T T T Tz T T T>W -W  - W JW =W $0%W ,H-W 2M3W <3=W FBGW PFQW Z#[W n&oW x9yW B>CW LJMW RFSW XYW bMDiW W W t + + + +"BCCH    8      C&C&C&C&L::::#'#'#'#'N'+hR hRhRhRhRhRhRV(8(8(8(8V""""B8    j-1AE111111h    F?A?A?A?A?A?A?A?AD8E8E8E8Ev////BMSRVRVRVRVRVjr