Rand Stats




Web API for Raku (RakuWAPI)


This is a Proposed Draft.

Version 0.9.Draft


This document standardizes an API for web application and server framework developers in Raku. It provides a standard protocol by which web applications and application servers may communicate with each other.

This standard has the following goals:

Aside from that is the underlying assumption that this is a simple interface and ought to at least somewhat resemble work in the standards it is derived from, including Rack, WSGI, PSGI, CGI, and others.


1.0 Glossary

A RakuWAPI application is a Raku routine that expects to receive an environment from an application server and returns a response each time it is called by the server.

A Web Server is an application that processes requests and responses according to a web-related protocol, such as HTTP or WebSockets or similar protocol.

The origin is the external entity that makes a given request and/or expects a response from the application server. This can be thought of generically as a web browser, bot, or other user agent.

An application server is a program that is able to provide an environment to a RakuWAPI application and process the value returned from such an application.

The application server might be associated with a web server, might itself be a web server, might process a protocol used to communicate with a web server (such as CGI or FastCGI), or may be something else entirely not related to a web server (such as a tool for testing RakuWAPI applications).

Middleware is a RakuWAPI application that wraps another RakuWAPI application for the purpose of performing some auxiliary task such as preprocessing request environments, logging, postprocessing responses, etc.

A framework developer is a developer who writes an application server.

An application developer is a developer who writes a RakuWAPI application.

A sane Supply is a Supply object that follows the emit*-done/quit protocol, i.e., it will emit 0 or more objects followed by a call to the done or quit handler. See Supply for details.

1.1 Type Constraints

The following type constraints are defined for use with this document.

    subset NonEmptyStr of Str where { !.defined || .chars > 0 };
    subset PathStr of Str where { !.defined || $_ ~~ any('', m{ ^ "/" }) };
    subset PositiveInt of Int where { !.defined || $_ > 0 };
    subset Supplierish of Any where { !.defined || ?.can('emit').grep(*.arity == 2) };

Any place a type is used in this document, the implementation is free to use any subtype (either subset or sub-class) of that type in place of the named type so long as the type constraint is guaranteed to hold for the subtype. For example, if an Int is required, it would be permissible to use an IntStr instead.


This specification is divided into three layers:

Each layer has a specific role related to the other layers. The server layer is responsible for managing the application lifecycle and performing communication with the origin. The application layer is responsible for receiving metadata and content from the server and delivering metadata and content back to the server. The middleware layer is responsible for enhancing the application or server by providing additional services and utilities.

This specification goes through each layer in order. In the process, each section only specifies the requirements and recommendations for the layer that section describes. When other layers a mentioned outside of its section, the specification is deliberately vague to keep all specifics in the appropriate section.

To aid in reading this specification, the numbering subsections of 2.0, 2.1, and 2.2 are matched so that you can navigate between them to compare the requirements of each layer. For example, 2.0.1 describes the environment the server provides, 2.1.1 describes how the application interacts with that environment, and 2.1.1 describes how middleware may manipulate that environment.

2.0 Layer 0: Server

A RakuWAPI application server is a program capable of running RakuWAPI applications as defined by this specification.

A RakuWAPI application server implements some kind of web service. For example, this may mean implementing an HTTP or WebSocket service or a related protocol such as CGI, FastCGI, SCGI, etc. An application server also manages the application lifecycle and executes the application, providing it with a complete environment, and processing the response from the application to determine how to respond to the origin.

One important aspect of this specification that is not defined is the meaning of a server error. At times it is suggested that certain states be treated as a server error, but what that actually means to a given implementation is deliberatly undefined. That is a complex topic which varies by implementation and by the state the server is in when such a state is discovered. The server SHOULD log such events and SHOULD use the appropriate means of communication provided to notify the application that a server error has occurred while responding.

2.0.0 Application Definition

A RakuWAPI application is defined as a class or object, which must be implemented according to a particular interface. The application server MUST provide a means by which an application is loaded. The application server SHOULD be able to load them by executing a RakuWAPI script file.

For example, here is a simple application:

    use v6;
    sub app(%env) {
        start {
            200, [ Content-Type => 'text/plain' ], [ 'Hello World!' ]

For full details on how an application is defined, see Section 2.2.0. For details on how a server interacts with the application, see Section 2.0.4.

2.0.1 The Environment

The environment is delivered to the application via hashes, they MUST be Associative. The application server makes the environment available to the application at runtime. The environment is used to:

Each variable or key in the environment is described as belonging to one of two roles:

Calls to the runtime routine MUST be provided with all required environment variables belonging to either of these roles in the passed environment hash. However, calls to the configuration routine (see 2.0.4) MUST include the configuration envrionment and SHOULD NOT include runtime environment in the passed environment hash.

The server MAY provide variables in the environment in either role in addition to the ones defined here, but they MUST contain a period and SHOULD be given a unique prefix to avoid name clashes.

The following prefixes are reserved and SHOULD NOT be used unless defined by this specification and only according to the definition given here.

In the tables below, a type constraint is given for each variable. The application server MUST provide each key as the named type. All variables given in the tables with and MUST be provided. Configuration Environment

The configuration environment MUST be made available to the application during every call made to the application, both to the configuration routine and the runtime routine.

wapi.versionVersion:DThis is the version of this specification, v0.9.Draft.
wapi.errorsSupplierish:DThe error stream for logging.
wapi.multithreadBool:DTrue if the app may be simultaneously invoked in another thread in the same process.
wapi.multiprocessBool:DTrue if the app may be simultaneously invoked in another process.
wapi.run-onceBool:DTrue if the server expects the app to be invoked only once during the life of the process. This is not a guarantee.
wapi.protocol.supportSet:DThis is a Set of strings naming the protocols supported by the application server.
wapi.protocol.enabledSetHash:DThis is the set of enabled protocols. The application may modify this set with those found in wapi.protocol.support to enable/disable protocols the server is permitted to use. Runtime Environment

Many of the runtime environment variables are derived from the old Common Gateway Interface (CGI). This environment MUST be given when the application is being called, i.e., whenever the runtime routine is called.

REQUEST_METHODNonEmptyStr:DThe HTTP request method, such as "GET" or "POST".
SCRIPT_NAMEPathStr:DThis is the initial portion of the URI path that refers to the application.
PATH_INFOPathStr:DThis is the remainder of the request URI path within the application. This value SHOULD be URI decoded by the application server according to RFC 3875
REQUEST_URIStr:DThis is the exact URI sent by the client in the request line of the HTTP request. The application server SHOULD NOT perform any decoding on it.
QUERY_STRINGStr:DThis is the portion of the requested URL following the ?, if any.
SERVER_NAMENonEmptyStr:DThis is the name of the web server.
SERVER_PORTPositiveInt:DThis is the port number of the web server.
SERVER_PROTOCOLNonEmptyStr:DThis is the server protocol sent by the client, e.g. "HTTP/1.0" or "HTTP/1.1".
CONTENT_LENGTHInt:_This corresponds to the Content-Length header sent by the client. If no such header was sent the application server MUST set this key to the Int type value.
CONTENT_TYPEStr:_This corresponds to the Content-Type header sent by the client. If no such header was sent the application server MUST set this key to the Str type value.
HTTP_*Str:_The remaining request headers are placed here. The names are prefixed with HTTP_, in ALL CAPS with the hyphens ("-") turned to underscores ("_"). Multiple incoming headers with the same name MUST have their values joined with a comma (", ") as described in RFC 2616. The HTTP_CONTENT_LENGTH and HTTP_CONTENT_TYPE headers MUST NOT be set.
Other CGI KeysStr:_The server MUST attempt to provide as many other CGI variables as possible, but no others are required or formally specified.
wapi.url-schemeStr:DEither "http" or "https".
wapi.inputSupply:DThe input stream for reading the body of the request, if any.
wapi.readyPromise:DThis is a vowed Promise that MUST be kept by the server as soon as the server has tapped the application's output Supply and is ready to receive emitted messages. The value of the kept Promise is irrelevent. The server SHOULD NOT break this Promise.
wapi.body.encodingStr:DName of the encoding the server will use for any strings it is sent.
wapi.protocolStr:DThis is a string naming the response protocols the server is expecting from the application for call.

In the environment, either SCRIPT_NAME or PATH_INFO must be set to a non-empty string. When REQUEST_URI is "/", the PATH_INFO SHOULD be "/" and SCRIPT_NAME SHOULD be the empty string. SCRIPT_NAME MUST NOT be set to "/".

2.0.2 The Input Stream

The input stream is set in the wapi.input key of the runtime environment. This represents the request payload sent from the origin. Unless otherwise indicated as part of the protocol definition, the server MUST provide a sane Supply that emits Blob objects containing the content of the request payload. It MAY emit nothing and just signal done if the request payload is empty.

2.0.3 The Error Stream

The error stream MUST be given in the configuration environment via wapi.errors. This MUST be a Supplierish (see Section 1.1) object the server provides for emitting errors. This is a defined object that has an emit method that has the same signature as Supplier.emit.

The application MAY call emit on this object zero or more times, passing any object that may be stringified. The server SHOULD write these log entries to a suitable log file or to $*ERR or wherever appropriate. If written to a typical file handle, it should automatically append a newline to each emitted message.

2.0.4 Application Lifecycle

After the application has been defined, it will be called each time the application server needs to respond to the origin. What that means will vary depending on which protocol is needed to appropriately respond to the origin.

These requirements, however, are in held in common regardless of protocol, the application server requirements are as follows:

The server MUST NOT call the application with wapi.protocol set to a protocol that has been previously disabled by the application via the wapi.protocol.enabled setting.

For details on how each protocol handles the application call, see section 4.

2.1 Layer 1: Middleware

RakuWAPI middleware is simply an application that wraps another application. Middleware is used to perform any kind of pre-processing, post-processing, or side-effects that might be added onto an application. Possible uses include logging, encoding, validation, security, debugging, routing, interface adaptation, and header manipulation.

For example, in the following snippet, mw is a middleware application that adds a custom header:

    sub app(%env) { start { 200, [ Content-Type => 'text/plain' ], [ 'Hello World' ] } }
    sub mw(&wrappee is copy, %config) returns Callable {
        &wrappee = wrappee(%config) if &wrappee.returns ~~ Callable;
        sub (%env) {
                -> $p {
                    my @r = $p.result;
                    @r[1].push: 'WAPI-Used' => 'True';
    my &mw-app = &mw.assuming(&app);

2.1.0 Middleware Definition

The way in which middleware is defined and applied is left up to the middleware author. The example in the previous section uses a combination of priming and defining a closure. This is, by no means, the only way to define RakuWAPI middleware in Raku.

What is important in middleware definition is the following:

Any general purpose middleware should be defined as a configuration routine.

Otherwise, There Is More Than One Way To Do It.

2.1.1 The Environment

Middleware applications MAY set or modify the environment (both configuration and runtime environment) as needed. Middleware applications MUST maintain the typing required for the server in Sections and above, as modified by extensions and the application protocol in use.

Whenever setting new variables in the environment, the variables MUST contain a period and SHOULD use a unique prefix to avoid name clashes with servers, other middleware, and applications.

2.1.2 The Input Stream

An application server is required to provide the input stream as a Supply emitting Blobs. Middleware MUST provide the same.

However, it is possible that the middleware will consume the data in wapi.input for the purpose of parsing or providing the data contained in another form. In such a case, the middleware SHOULD provide an empty, already closed, Supply. The middleware should provide the data in another environment variable.

2.1.3 The Error Stream

See sections 2.0.3 and 2.2.3.

2.1.4 Application Lifecycle

Middleware MUST return a valid response to the server according to the value set in wapi.protocol by the server (or whatever middleware came before it).

See sections 2.0.4 and 2.2.4. Middleware MUST adhere to all requirements of the application as respects the server (2.2.4) and all requirements of the server as respects the application (2.0.4).

See section 4 for details on protocol handling.

2.2 Layer 2: Application

A RakuWAPI application is a Raku routine. The application MUST be Callable. An application may be defined as either a runtime routine or a configuration routine. A configuration routine receives a RakuWAPI configuration environment and returns a runtime routine. A runtime routine receives a RakuWAPI runtime environment and responds to it by returning a response.

As an example, a simple Hello World RakuWAPI application defined with a runtime routine could be implemented as follows:

    sub app(%env) {
        start {
            200, [ Content-Type => 'text/plain' ], [ 'Hello World' ]

Or, a slightly more complex Hello World application could be implemented using a configuration routine instead like so:

    sub app-config(%config) returns Callable {
        %config<wapi.protocol.enabled> ∩= set('request-response');
        sub app(%env) {
            start {
                200, [ Content-Type => 'text/plain' ], [ 'Hello World' ]

This second application makes sure that only the request-response protocol is enabled before returning an application only capable of responding that way (see Section 4).

2.2.0 Defining an Application

An application is defined in one of two mechanisms, as mentioned in the previous section:

During application defintion, the application MAY also instantiate and apply middleware to be used by the application. Runtime Routine

To define an application as a runtime routine, the application is defined as a Callable (typically a Routine). This application MUST accept a single parameter as its argument. When the application server handles a request, the server will set this parameter to the runtime environment for the request. The application MUST NOT have a return type set to a Callable.

The application SHOULD respond to the caller, i.e., the application server or outer middleware, according to the wapi.protocol string set in the passed environment.

Here, for example, is a RakuWAPI application that calculates and prints the Nth Lucas number depending on the value passed in the query string. This assumes a request-response protocol (see Section 4.0).

    sub lucas-runtime(%env) {
        start {
            my $n = %env<QUERY_STRING>.Int;
            my $lucas-number := 2, 1, * + * ... *;
            200, [ Content-Type => 'text/plain' ], [ $lucas-number[$n] ];

This application is vulnerable, however, to problems if the server might call the application with a different protocol. Configuration Routine

An application SHOULD return a configuration routine, which is defined just like the runtime routine, but it must constrain its return type to something Callable. This application MUST accept a single parameter as its argument, just like the runtime routine. This single parameter will be set to the configuration environment when called by the application server. This routine will also be called prior to any request or other contact from an origin has occurred, which gives the application the opportunity to communicate with the server early.

The application SHOULD modify the configuration environment to suit the needs of the application. The application SHOULD end the routine by returning a runtime routine (see the previous section).

Here is the example from the previous section, but using a configuration routine to guarantee that the only protocol the application can use to contact it is the request-response protocol (see Section 4.0). It ends by returning the &lucas-app subroutine defined in the previous section:

    sub lucas-config(%config) returns Callable {
        # Only permit the request-response protocol
        %config<wapi.protocol.enabled> ∩= set('request-response');


2.2.1 The Environment

Calls to the configuration routine of the application (if defined) will receive the configuration environment as defined in Section Calls to the runtime routine of the application will receive the combination of the runtime environment and configuration environment as defined in Sections and Additional variables may be provided by your application server and middleware in either environment hash.

The application itself MAY store additional values in the environment as it sees fit. This allows the application to communicate with a server or middleware or even with itself. When the application modifies the environment, the variables set MUST contain a period and SHOULD start with a unique name that is not wapi. or wapix. as these are reserved.

2.2.2 The Input Stream

Some calls to your application may be accompanied by a request payload. For example, a POST or PUT request sent by an origin using HTTP will typically include such a payload. The application MAY choose to read the payload using the sane Supply provided in the wapi.input variable of the runtime environment. This will provide a stream lf Blobs.

2.2.3 The Error Stream

The application server is required to provide a wapi.errors variable in the environment with a Supplierish object (see Section 1.1). The application MAY emit any errors or messages here using any object that stringifies. The application SHOULD NOT terminate such messages with a newline as the server will do so if necessary.

2.2.4 Application Call

To handle requests from the origin, the application server will make calls to the application routine. The application SHOULD return a valid response to the server. The response required will depend on what string is set in wapi.protocol within the runtime environment, so the application SHOULD check that on every call if it may vary.

The application SHOULD attempt to return a value as quickly as possible via the runtime routine. For protocols that require the application to return a Promise, the application SHOULD wrap the entire body of the call in a start block to minimize the time the server will be waiting on the application.

See section 4 on how different application protocols are handled.

3 Extensions

In addition to the standard specification, there are a number of extensions that servers or middleware MAY choose to implement. They are completely optional and applications and middleware SHOULD check for their presence before using them. Such checks SHOULD be performed as early as possible.

Unless stated otherwise, all environment variables described here are set in the runtime environment, which is passed as the single argument with each call to the runtime routine.

3.0 Header Done

The wapix.header.done environment variable, if provided, MUST be a vowed Promise. This Promise MUST be kept when the server is done sending or processing the response header. The Promise MUST be broken if the server is unable or unwilling to send the application provided headers.

This is not an exhaustive list, but here are a few possible reasons why this Promise MAY be broken:

When broken, this Promise SHOULD be broken with a helpful diagnostic exception.

3.1 Body Done

The wapix.body.done environment variable, if provided, MUST be a vowed Promise. This Promise MUST be kept when the server is done sending or processing the response body. The Promise MUST be broken if the server is unable or unwilling to send the complete application body.

This is not an exhaustive list, but here are a few possible reasons why this Promise MAY be broken:

In particular, wapix.body.done MUST be broken if wapix.header.done is broken (assuming both extensions are implemented).

When broken, the Promise SHOULD be broken with a helpful diagnostic exception.

3.2 Raw Socket

The wapix.io environment variable, if provided, SHOULD be the socket object used to communicate to the client. This is the interface of last resort as it sidesteps the entire RakuWAPI interface. It may be useful in cases where an application wishes to control details of the socket itself.

If your application requires the use of this socket, please file an issue describing the nature of your application in detail. You may have a use-case that a future revision to RakuWAPI can improve.

This variable MAY be made available as part of the configuration environment.

3.3 Logger

The wapix.logger environment variable, if provided, MUST be a Routine defined with a signature as follows:

    sub (Str:D $message, Str:D :$level = 'info');

When called application MUST provide a $level that is one of: "debug", "info", "warn", "error", "fatal".

Te wapix.logger environment variable SHOULD be provided in the configuration environment.

3.4 Sessions

This extension implements basic session handling that allows certain data to persist across requests. Session data SHOULD be associated with a particular origin.

The wapix.session environment variable, if provided, MUST be an Associative. This hash maps arbitrary keys and values that may be read and written to by an application. The application SHOULD only use Str keys and values. The details of persisting this data is up to the application server or middleware implementing the session extension.

The wapix.session.options environment variable, if provided, MUST be an Associative. This variable uses implementation-specific keys and values to communicate between the application and the extension implementation. This allows the application a channel by which to instruct the session handler how to operate.

3.5 Harakiri Mode

The wapix.harakiri environment variable, if provided, MUST be a Bool. If set to True it signals to the application that the server supports harakiri mode, which allows the application to ask the server to terminate the current work when the request is complete. This variable SHOULD be set in the configuration environment.

The wapix.harakiri.commit environment variable MAY be set to a True value by the application to signal to the server that the current worker should be killed after the current request has been processed.

3.6 Cleanup Handlers

The wapix.cleanup environment variable, if provided, MUST be a Bool. If set to True it tells the application that the server supports running cleanup handlers after the request is complete. This variable SHOULD be set in the configuration environment.

The wapix.cleanup.handlers environment variable MUST be provided if the wapix.cleanup flag is set. This MUST an Array. The application adds cleanup handlers to the array by putting Callables into the Array (usually by pushing). Each handler will be given a copy of the %env as the first argument. The server MUST run these handlers, but only after the application has completely finished returning the response and any response payload.

If the server supports the harakiri extension, it SHOULD allow the cleanup handlers to invoke harakiri mode by setting wapix.harakiri.commit (see 3.5).

3.7 Output Block Detection

The wapix.body.backpressure environment variable, if provided, MUST be a Bool flag. It is set to True to indicate that the RakuWAPI server provide response backpressure detection by polling for non-blocking I/O problems. In this case, the server MUST provide the other two environment variables. If False or not defined, the server does not provide these two environment variables. This variable SHOULD be defined in the configuration environment.

The wapix.body.backpressure.supply environment variable MUST be provided if wapix.body.backpressure is True. When provided, it MUST be a live Supply that periodically emits True and False values. True is emitted when the server polls for backpressure and detects a blocked output socket. False is emitted when the server polls for backpressure and detects the previously blocked socket is no longer blocked.

The wapix.body.backpressure.test environment variable MUST be provided if wapix.body.backpressure is True. When provided, it MUST be a Bool that is True while output has last been detected as blocked and False otherwise. This can be useful for detecting the initial state before the backpressure supply has emitted any value or just as a way to poll the last known status of the socket.

3.8 Protocol Upgrade

The wapix.net-protocol.upgrade environment variable MUST be provided in the configuration environment, if the server implements the protocol upgrade extension. It MUST be the Set of names of protocols the server supports for upgrade.

When the client makes a protocol upgrade request using an Upgrade header, the application MAY request that the server negotiate the upgrade to one of these supported protocols by sending a WAPIx-Upgrade header back to the server with the named protocol. The application MAY send any other headers related to the Upgrade and MAY send a message payload if the upgrade allows it. These SHOULD override any server supplied values or headers.

The server MUST negotiate the new protocol and enable any environment variables required for interacting through that protocol. After the handshake or upgrade negoatiation is complete, the server MUST make a new call to the application with a new environment to process the remainder of the network request with the origin.

3.8.0 HTTP/2 Protocol Upgrade

The workings of HTTP/2 are similar enough to HTTP/1.0 and HTTP/1.1 that use of a protocol upgrade may not be necessary in most or all use-cases. However, servers MAY choose to delegate this to the application using the protocol upgrade extension.

Servers that support this protocol upgrade MUST place the name "h2c" and/or "h2" into the wapix.net-protocol.upgrade set, for support of HTTP/2 over cleartext connections and HTTP/2 over TLS, respectively.

The application MUST NOT request an upgrade using the WAPIx-Upgrade header for "h2c" unless the wapi.url-scheme is "http". Similarly, the application MUST NOT request an upgrade for "h2" unless the wapi.url-scheme is "https". The application server SHOULD enforce this requirement for security reasons.

The application MUST NOT tap the wapi.input stream when performing this upgrade. The application SHOULD NOT return a message payload aside from an empty Supply.

3.8.1 WebSocket Protocol Upgrade

Servers that support the WebSocket protocol upgrade MUST place the name "ws" into the wapix.net-protocol.upgrade set.

The application MUST NOT tap the wapi.input stream when performing this upgrade. The application SHOULD NOT return a message payload aside from an empty Supply.

3.9 Transfer Encoding

This extension is only for HTTP/1.1 protocol connections. When the server supports this extension, it MUST provide a wapix.http11.transfer-encoding variable containing a Set naming the transfer encodings the server supports as strings. This SHOULD be set in the configuration environment.

When the application returns a header named WAPIx-Transfer-Encoding with the name of one of the supported transfer encoding strings, the server MUST apply that transfer encoding to the message payload. If the connection is not HTTP/1.1, the server SHOULD ignore this header.

3.9.0 Chunked Encoding

When the server supports and the application requests "chunked" encoding. The application server SHOULD treat each emitted Str or Blob as a chunk to be encoded according to RFC7230. It MUST adhere to requirements of RFC7230 when sending the response payload to the origin.

3.9.1 Other Encodings

All other encodings should be handled as required by the relevant rules for HTTP/1.1.

3.10 HTTP/2 Push Promises

When the SERVER_PROTOCOL is "HTTP/2", servers SHOULD support the HTTP/2 push promises extension. However, applications SHOULD check to make sure that the wapix.h2.push-promise variable is set to a defined value before using this extension.

This extension is implemented by providing a variable named wapix.h2.push-promise. When provided, this MUST be a Supplier.

When the application wishes to invoke a server push, it MUST emit a message describing the request the server is pushing. The application server will receive this request and make a new, separate call to the application to fulfill that request.

Push-promise messages are sent as an Array of Pairs. This is a set of headers to send with the PUSH_PROMISE frame, including HTTP/2 pseudo-headers like ":path" and ":authority".

Upon receiving a message to wapix.h2.push-promise, the server SHOULD schedule a followup call to the application to fulfill the push-promise as if the push-promise were an incoming request from the client. (The push-promise could be canceled by the client, so the call to the application might not actually happen.)

4 Application Protocol Implementation

One goal of RakuWAPI application servers is to allow the application to focus on building web applications without having to implement the mundane details of web protocols. In times past, this was simply a matter of implementing HTTP/1.x or some kind of front-end to HTTP/1.x (such as CGI or FastCGI). While HTTP/1.x is still relevant to the web today, new protocols have also become important to modern web applications, such as HTTP/2 and WebSocket.

These protocols may have different interfaces that do not lend themselves to the request-response pattern specifed by PSGI. Therefore, we provide a means by which servers and applications may implement these alternate protocols, which each may have different requirements. These protocols are called application protocols to differentiate them from network protocols. For example, rather than providing a protocol for HTTP, we provide the "request-response" protocol. The underlying network protocol may be HTTP/1.0, HTTP/1.1, HTTP/2 or it may be something else that operates according to a similar pattern.

The application and application server MUST communicate according to the application protocol used for the current application call. For many applications, just implementing the basic request-response protocol is enough. However, to allow for more complete applications, RakuWAPI provides additional tools to help application and application server to communicate through a variety of situations. This is handled primarily via the wapi.protocol, wapi.protocol.support, and wapi.protocol.enabled values in the environment.

The application SHOULD check the value in wapi.protocol. If the application needs to make a decision based upon the network protocol, the application SHOULD check the SERVER_PROTOCOL.

The application SHOULD check wapi.protocol.support to discover which protocols are supported by the application server. An application that is not able to support all supported protocols SHOULD modify wapi.protocol.enabled to only include protocols supported by the application as early as possible.

The application server MUST provide the wapi.protocol.support and wapi.protocol.enabled values as part of the configuration environment. The application server MUST NOT use any protocol that is not a member of the wapi.protocol.enabled set.

This specification defines the following protocols:

It is recommended that an application server that implements all of these protocols only enable the request-response protocol within wapi.protocol.enabled by default. This allows simple RakuWAPI applications to safely operate without having to perform any special configuration.

4.0 Request-Response Protocol

The "request-response" protocol SHOULD be used for any HTTP-style client-server web protocol, this include HTTP/1.x and HTTP/2 connections over plain text and TLS or SSL.

4.0.0 Response

Here is an example application that implements the "request-response" protocol:

    sub app(%env) {
        start {
            [ Content-Type => 'text/plain' ],
            supply {
                emit "Hello World"

The runtime routine for the application MUST return a Promise. This Promise MUST be kept with a Capture (or something that becomes one on return) and MAY be broken. The Capture MUST contain 3 positional elements, which are the status code, the list of headers, and the response payload.

Here is a more interesting example application of this interface:

    sub app(%env) {
        start {
            my $n = %env<QUERY_STRING>.Int;
            [ Content-Type => 'text/plain' ],
            supply {
                my $acc = 1.FatRat;
                for 1..$n -> $v {
                    emit $acc *= $v;
                    emit "\n";

The example application above will print out all the values of factorial from 1 to N where N is given in the query string. The header may be returned immediately and the lines of the body may be returned as the values of factorial are calculated by the application server.

And here is an example demonstrating a couple ways in which coercion can be used by an application to simplify the result:

    sub app(%env) {
        my enum HttpStatus (OK => 200, NotFound => 404, ServerError => 500);
        start {
            OK, [ Content-Type => 'text/plain' ], [ 'Hello World' ]

In this example, the status is returned using an enumeration which coerces to an appropriate integer value. The payload is returned as a list, which is automatically coerced into a Supply.

Applications SHOULD return a Promise as soon as possible. It is recommended that applications wrap all operations within a start block to make this automatic.

Application servers SHOULD NOT assume that the returned Promise will be kept. It SHOULD assume that the Promise has been vowed and MUST NOT try to keep or break the Promise from the application.

4.0.1 Response Payload

The response payload, in the result of the application's promise, must be a sane Supply. This supply MUST emit nothing for requests whose response must be empty.

For any other request, the application MAY emit zero or more messages in the returned payload Supply. The messages SHOULD be handled as follows:

4.0.2 Encoding

The application server SHOULD handle encoding of strings or stringified objects emitted to it. When performing encoding, the application server SHOULD honor the charset set within the Content-Type header, if given. If it does not honor the charset or no charset is set by the application, it MUST encode any strings in the response payload according to the encoding named in wapi.body.encoding.

4.0.3 Request-Response Lifecycle

The application server performs the following during a single request-response interaction:

The server processes requests from an origin, passes the processed request information to the application by calling the application, waits for the application's response, and then returns the response to the origin. In the simplest example this means handling an HTTP roundtrip. Yet, it may also mean implementing a related protocol like CGI or FastCGI or SCGI or something else entirely.

In the modern web, an application may want to implement a variety of complex HTTP interactions. These use-cases are not described by the typical HTTP request-response roundtrip with which most web developers are familiar. For example, an interactive Accept-Continue response or a data stream to or from the origin or an upgrade request to switch protocols. As such, application servers SHOULD make a best effort to be implemented in such a way as to make this variety applications possible.

The application server SHOULD pass control to the application as soon as the headers have been received and the environment can be constructed. The application server MAY continue processing the request payload while the application server begins its work. The server SHOULD NOT emit the request payload to wapi.input yet.

Once the application has returned the its Promise to respond, the server SHOULD wait until the promise is kept. Once kept, the server SHOULD tap the response payload as soon as possible. After tapping the Supply, the application server SHOULD keep the Promise in wapi.ready (the application server SHOULD NOT break this Promise). Once that promise has been kept, only then SHOULD the server start emitting the contents of the request payload, if any, to wapi.input.

The server SHOULD return the application response headers back to the origin as soon as they are received. After which, the server SHOULD return each chunk emitted by the response body from the application as soon as possible.

This particular order of operations during the application call will ensure that the application has the greatest opportunity to perform well and be able to execute a variety of novel HTTP interactions.

4.0.4 HTTP/1.1 Keep Alive

When an application server supports HTTP/1.1 with keep-alive, each request sent on the connection MUST be handled as a separate call to the application.

4.0.4 HTTP/2 Handling

When a server supports HTTP/2 it SHOULD implement the HTTP/2 Push Promise Extension defined in section 3.10. An application server MAY want to consider implementing HTTP/2 protocol upgrades using the extension described in section 3.8 or MAY perform such upgrades automatically instead.

The application MUST be called once for each request or push-promise made on an HTTP/2 stream.

4.1 Framed-Socket Protocol

The "framed-socket" protocol is appropriate for WebSocket-style peer-to-peer TCP connections. The following sections assume a WebSocket connection, but might be adaptable for use with other framed message exchanges, such as message queues or chat servers.

4.1.0 Response

Any application server implementing WebSocket MUST adhere to all the requirements described above with the following modifications when calling the application for a WebSocket:

The application MUST return a Promise that is kept with just a Supply (i.e., not a 3-element Capture as is used with the request-response protocol). The application MAY break this Promise. The application will emit frames to send back to the origin using the promised Supply.

4.1.1 Response Payload

Applications MUST return a sane Supply that emits an object for every frame it wishes to return to the origin. The application MAY emit zero or more messages to this supply. The application MAY emit done to signal that the connection is to be terminated with the client.

The messages MUST be framed and returned to the origin by the application server as follows, based on message type:

4.1.2 Encoding

The application server SHOULD handle encoding of strings or stringified objects emitted to it. The server MUST encode any strings in the message payload according to the encoding named in wapi.body.encoding.

4.2 PSGI Protocol

To handle legacy applications, this specification defines the "psgi" protocol. If wapi.protocol.enabled has both "psgi" and "request-response" enabled, the "request-response" protocol SHOULD be preferred.

4.2.0 Response

The application SHOULD return a 3-element Capture directly. The first element being the numeric status code, the second being an array of pairs naming the headers to return in the response, and finally an array of values representing the response payload.

4.2.1 Payload

The payload SHOULD be delivered as an array of strings, never as a supply.

4.2.2 Encoding

String encoding is not defined by this document.

4.3 Socket Protocol

The "socket" protocol can be provided by an application server wishing to allow the application to basically take over the socket connection with the origin. This allows the application to implement an arbitrary protocol. It does so, however, in a way that is aimed toward providing better portability than using the socket extension.

The socket provided sends and receives data directly to and from the application without any framing, buffering, or modification.

The "socket" protocol SHOULD only be used when it is enabled in wapi.protocol.enabled and no other protocol enabled there can fulfill the current application call.

4.3.0 Response

Here's an example application that implements the "socket" protocol to create a naïve HTTP server:

    sub app(%env) {
        start {
            supply {
                whenever %env<wapi.input> -> $msg {
                    my $req = parse-http-request($msg);

                    if $req.method eq 'GET' {
                        emit "200 OK HTTP/1.0\r\n";
                        emit "\r\n";
                        emit "Custom HTTP Server";

The socket protocol behaves very much like "framed-socket", but with fewer specified details in the environment. Some of the mandatory environment are not mandatory for the socket protocol.

The following parts of the environment SHOULD be provided as undefined values:

The wapi.protocol must be set to "socket".

The application server SHOULD provide data sent by the origin to the application through the wapi.input supply as it arrives. It MUST still be a sane supply.

The application server MAY tunnel the connection through another socket, stream, file handle, or what-not. That is, the application MAY NOT assume the communication is being performed over any particular medium. The application server SHOULD transmit the data to the origin as faithfully as possible, keeping the intent of the application as much as possible.

The application server SHOULD forward on data emitted by the application in the returned payload supply as it arrives. The application server SHOULD send no other data to the origin over that socket once the application is handed control, unless related to how the data is being tunneled or handled by the server. The application server SHOULD close the connection with the origin when the application server sends the "done" message to the supply. Similarly, the application server SHOULD send "done" the wapi.input supply when the connection is closed by the client or server.

4.3.1 Response Payload

Applications MUST return a sane Supply which emits a string of bytes for every message to be sent. The messages returned on this stream MUST be Blob objects.

4.3.2 Encoding

The application server SHOULD reject any object other than a Blob sent as part of the message payload. The application server is not expected to perform any encoding or stringification of messages.










This second revision eliminates the legacy standard and requires that all P6SGI responses be returned as a Promise. The goal is to try and gain some uniformity in the responses the server must deal with.


This is the first published version. It was heavily influenced by PSGI and included interfaces based on the standard, deferred, and streaming responses of PSGI. Instead of callbacks, however, it used Promise to handle deferred responses and Channel to handle streaming. It mentioned middleware in passing.