Scroll to navigation

http(3tcl) Tcl Bundled Packages http(3tcl)


NAME

http - Client-side implementation of the HTTP/1.1 protocol

SYNOPSIS

package require http ?2.9?

::http::config ?-option value ...?

::http::geturl url ?-option value ...?

::http::formatQuery key value ?key value ...?

::http::quoteString value

::http::reset token ?why?

::http::wait token

::http::status token

::http::size token

::http::code token

::http::ncode token

::http::meta token

::http::data token

::http::error token

::http::cleanup token

::http::register proto port command

::http::registerError port ?message?

::http::unregister proto

EXPORTED COMMANDS

Namespace http exports the commands config, formatQuery, geturl, quoteString, register, registerError, reset, unregister, and wait.

It does not export the commands cleanup, code, data, error, meta, ncode, size, or status.



DESCRIPTION

The http package provides the client side of the HTTP/1.1 protocol, as defined in RFC 7230 to RFC 7235, which supersede RFC 2616. The package implements the GET, POST, and HEAD operations of HTTP/1.1. It allows configuration of a proxy host to get through firewalls. The package is compatible with the Safesock security policy, so it can be used by untrusted applets to do URL fetching from a restricted set of hosts. This package can be extended to support additional HTTP transport protocols, such as HTTPS, by providing a custom socket command, via ::http::register.

The ::http::geturl procedure does a HTTP transaction. Its options determine whether a GET, POST, or HEAD transaction is performed. The return value of ::http::geturl is a token for the transaction. The value is also the name of an array in the ::http namespace that contains state information about the transaction. The elements of this array are described in the STATE ARRAY section.

If the -command option is specified, then the HTTP operation is done in the background. ::http::geturl returns immediately after generating the HTTP request and the callback is invoked when the transaction completes. For this to work, the Tcl event loop must be active. In Tk applications this is always true. For pure-Tcl applications, the caller can use ::http::wait after calling ::http::geturl to start the event loop.

Note: The event queue is even used without the -command option. As a side effect, arbitrary commands may be processed while http::geturl is running.

COMMANDS

::http::config ?options?
The ::http::config command is used to set and query the name of the proxy server and port, and the User-Agent name used in the HTTP requests. If no options are specified, then the current configuration is returned. If a single argument is specified, then it should be one of the flags described below. In this case the current value of that setting is returned. Otherwise, the options should be a set of flags and values that define the configuration:
The Accept header of the request. The default is */*, which means that all types of documents are accepted. Otherwise you can supply a comma-separated list of mime type patterns that you are willing to receive. For example, “image/gif, image/jpeg, text/*”.
Specifies whether HTTP/1.1 transactions on a persistent socket will be pipelined. See the PERSISTENT SOCKETS section for details. The default is 1.
Specifies whether requests that use the POST method will always use a fresh socket, overriding the -keepalive option of command http::geturl. See the PERSISTENT SOCKETS section for details. The default is 0.
The name of the proxy host, if any. If this value is the empty string, the URL host is contacted directly.
The proxy port number.
The command is a callback that is made during ::http::geturl to determine if a proxy is required for a given host. One argument, a host name, is added to command when it is invoked. If a proxy is required, the callback should return a two-element list containing the proxy server and proxy port. Otherwise the filter should return an empty list. The default filter returns the values of the -proxyhost and -proxyport settings if they are non-empty.

The ::http::geturl command runs the -proxyfilter callback inside a catch command. Therefore an error in the callback command does not call the bgerror handler. See the ERRORS section for details.

Specifies what to do if a POST request over a persistent connection fails because the server has half-closed the connection. If boolean true, the request will be automatically retried; if boolean false it will not, and the application that uses http::geturl is expected to seek user confirmation before retrying the POST. The value true should be used only under certain conditions. See the PERSISTENT SOCKETS section for details. The default is 0.
The encoding used for creating the x-url-encoded URLs with ::http::formatQuery and ::http::quoteString. The default is utf-8, as specified by RFC 2718. Prior to http 2.5 this was unspecified, and that behavior can be returned by specifying the empty string ({}), although iso8859-1 is recommended to restore similar behavior but without the ::http::formatQuery or ::http::quoteString throwing an error processing non-latin-1 characters.
The value of the User-Agent header in the HTTP request. In an unsafe interpreter, the default value depends upon the operating system, and the version numbers of http and Tcl, and is (for example) “Mozilla/5.0 (Windows; U; Windows NT 10.0) http/2.9.0 Tcl/8.6.9”. A safe interpreter cannot determine its operating system, and so the default in a safe interpreter is to use a Windows 10 value with the current version numbers of http and Tcl.
If the value is boolean true, then by default requests will send a header “Accept-Encoding: gzip,deflate,compress”. If the value is boolean false, then by default this header will not be sent. In either case the default can be overridden for an individual request by supplying a custom Accept-Encoding header in the -headers option of http::geturl. The default is 1.
::http::geturl url ?options?
The ::http::geturl command is the main procedure in the package. The -query option causes a POST operation and the -validate option causes a HEAD operation; otherwise, a GET operation is performed. The ::http::geturl command returns a token value that can be used to get information about the transaction. See the STATE ARRAY and ERRORS section for details. The ::http::geturl command blocks until the operation completes, unless the -command option specifies a callback that is invoked when the HTTP transaction completes. ::http::geturl takes several options:
Specifies whether to force interpreting the URL data as binary. Normally this is auto-detected (anything not beginning with a text content type or whose content encoding is gzip or compress is considered binary data).
The block size used when reading the URL. At most size bytes are read at once. After each block, a call to the -progress callback is made (if that option is specified).
Copy the URL contents to channel name instead of saving it in state(body).
Invoke callback after the HTTP transaction completes. This option causes ::http::geturl to return immediately. The callback gets an additional argument that is the token returned from ::http::geturl. This token is the name of an array that is described in the STATE ARRAY section. Here is a template for the callback:

proc httpCallback {token} {

upvar #0 $token state
# Access state as a Tcl array }

The ::http::geturl command runs the -command callback inside a catch command. Therefore an error in the callback command does not call the bgerror handler. See the ERRORS section for details.

Invoke callback whenever HTTP data is available; if present, nothing else will be done with the HTTP data. This procedure gets two additional arguments: the socket for the HTTP data and the token returned from ::http::geturl. The token is the name of a global array that is described in the STATE ARRAY section. The procedure is expected to return the number of bytes read from the socket. Here is a template for the callback:

proc httpHandlerCallback {socket token} {

upvar #0 $token state
# Access socket, and state as a Tcl array
# For example...
...
set data [read $socket 1000]
set nbytes [string length $data]
...
return $nbytes }

The http::geturl code for the -handler option is not compatible with either compression or chunked transfer-encoding. If -handler is specified, then to work around these issues http::geturl will reduce the HTTP protocol to 1.0, and override the -zip option (i.e. it will not send the header "Accept-Encoding: gzip,deflate,compress").

If options -handler and -channel are used together, the handler is responsible for copying the data from the HTTP socket to the specified channel. The name of the channel is available to the handler as element -channel of the token array.

The ::http::geturl command runs the -handler callback inside a catch command. Therefore an error in the callback command does not call the bgerror handler. See the ERRORS section for details.

This option is used to add headers not already specified by ::http::config to the HTTP request. The keyvaluelist argument must be a list with an even number of elements that alternate between keys and values. The keys become header field names. Newlines are stripped from the values so the header cannot be corrupted. For example, if keyvaluelist is Pragma no-cache then the following header is included in the HTTP request:

Pragma: no-cache
If boolean true, attempt to keep the connection open for servicing multiple requests. Default is 0.
Force the HTTP request method to type. ::http::geturl will auto-select GET, POST or HEAD based on other options, but this option overrides that selection and enables choices like PUT and DELETE for WebDAV support.

It is the caller's responsibility to ensure that the headers and request body (if any) conform to the requirements of the request method. For example, if using -method POST to send a POST with an empty request body, the caller must also supply the option “-headers {Content-Length 0}”.

Pass an specific local address to the underlying socket call in case multiple interfaces are available.
The callback is made after each transfer of data from the URL. The callback gets three additional arguments: the token from ::http::geturl, the expected total size of the contents from the Content-Length meta-data, and the current number of bytes transferred so far. The expected total size may be unknown, in which case zero is passed to the callback. Here is a template for the progress callback:

proc httpProgress {token total current} {

upvar #0 $token state }
Select the HTTP protocol version to use. This should be 1.0 or 1.1 (the default). Should only be necessary for servers that do not understand or otherwise complain about HTTP/1.1.
This flag (if the value is non-empty) causes ::http::geturl to do a POST request that passes the string query verbatim to the server as the request payload. The content format (and encoding) of query is announced by the request header Content-Type which is set by the option -type. Any value of -type is permitted, and it is the responsibility of the caller to supply query in the correct format.

If -type is not specified, it defaults to application/x-www-form-urlencoded, which requires query to be an x-url-encoding formatted query-string (this -type and query format are used in a POST submitted from an html form). The ::http::formatQuery procedure can be used to do the formatting.

The block size used when posting query data to the URL. At most size bytes are written at once. After each block, a call to the -queryprogress callback is made (if that option is specified).
This flag causes ::http::geturl to do a POST request that passes the data contained in channelID to the server. The data contained in channelID must be an x-url-encoding formatted query unless the -type option below is used. If a Content-Length header is not specified via the -headers options, ::http::geturl attempts to determine the size of the post data in order to create that header. If it is unable to determine the size, it returns an error.
The callback is made after each transfer of data to the URL (i.e. POST) and acts exactly like the -progress option (the callback format is the same).
Whether to enforce RFC 3986 URL validation on the request. Default is 1.
If milliseconds is non-zero, then ::http::geturl sets up a timeout to occur after the specified number of milliseconds. A timeout results in a call to ::http::reset and to the -command callback, if specified. The return value of ::http::status is timeout after a timeout has occurred.
Use mime-type as the Content-Type value, instead of the default value (application/x-www-form-urlencoded) during a POST operation.
If boolean is non-zero, then ::http::geturl does an HTTP HEAD request. This request returns meta information about the URL, but the contents are not returned. The meta information is available in the state(meta) variable after the transaction. See the STATE ARRAY section for details.
::http::formatQuery key value ?key value ...?
This procedure does x-url-encoding of query data. It takes an even number of arguments that are the keys and values of the query. It encodes the keys and values, and generates one string that has the proper & and = separators. The result is suitable for the -query value passed to ::http::geturl.
::http::quoteString value
This procedure does x-url-encoding of string. It takes a single argument and encodes it.
::http::reset token ?why?
This command resets the HTTP transaction identified by token, if any. This sets the state(status) value to why, which defaults to reset, and then calls the registered -command callback.
::http::wait token
This is a convenience procedure that blocks and waits for the transaction to complete. This only works in trusted code because it uses vwait. Also, it is not useful for the case where ::http::geturl is called without the -command option because in this case the ::http::geturl call does not return until the HTTP transaction is complete, and thus there is nothing to wait for.
::http::data token
This is a convenience procedure that returns the body element (i.e., the URL data) of the state array.
::http::error token
This is a convenience procedure that returns the error element of the state array.
::http::status token
This is a convenience procedure that returns the status element of the state array.
::http::code token
This is a convenience procedure that returns the http element of the state array.
::http::ncode token
This is a convenience procedure that returns just the numeric return code (200, 404, etc.) from the http element of the state array.
::http::size token
This is a convenience procedure that returns the currentsize element of the state array, which represents the number of bytes received from the URL in the ::http::geturl call.
::http::meta token
This is a convenience procedure that returns the meta element of the state array which contains the HTTP response headers. See below for an explanation of this element.
::http::cleanup token
This procedure cleans up the state associated with the connection identified by token. After this call, the procedures like ::http::data cannot be used to get information about the operation. It is strongly recommended that you call this function after you are done with a given HTTP request. Not doing so will result in memory not being freed, and if your app calls ::http::geturl enough times, the memory leak could cause a performance hit...or worse.
::http::register proto port command
This procedure allows one to provide custom HTTP transport types such as HTTPS, by registering a prefix, the default port, and the command to execute to create the Tcl channel. E.g.:

package require http
package require tls
::http::register https 443 ::tls::socket
set token [::http::geturl https://my.secure.site/]
::http::registerError port ?message?
This procedure allows a registered protocol handler to deliver an error message for use by http. Calling this command does not raise an error. The command is useful when a registered protocol detects an problem (for example, an invalid TLS certificate) that will cause an error to propagate to http. The command allows http to provide a precise error message rather than a general one. The command returns the value provided by the last call with argument message, or the empty string if no such call has been made.
::http::unregister proto
This procedure unregisters a protocol handler that was previously registered via ::http::register, returning a two-item list of the default port and handler command that was previously installed (via ::http::register) if there was such a handler, and an error if there was no such handler.

ERRORS

The ::http::geturl procedure will raise errors in the following cases: invalid command line options, an invalid URL, a URL on a non-existent host, or a URL at a bad port on an existing host. These errors mean that it cannot even start the network transaction. It will also raise an error if it gets an I/O error while writing out the HTTP request header. For synchronous ::http::geturl calls (where -command is not specified), it will raise an error if it gets an I/O error while reading the HTTP reply headers or data. Because ::http::geturl does not return a token in these cases, it does all the required cleanup and there is no issue of your app having to call ::http::cleanup.

For asynchronous ::http::geturl calls, all of the above error situations apply, except that if there is any error while reading the HTTP reply headers or data, no exception is thrown. This is because after writing the HTTP headers, ::http::geturl returns, and the rest of the HTTP transaction occurs in the background. The command callback can check if any error occurred during the read by calling ::http::status to check the status and if its error, calling ::http::error to get the error message.

Alternatively, if the main program flow reaches a point where it needs to know the result of the asynchronous HTTP request, it can call ::http::wait and then check status and error, just as the callback does.

The ::http::geturl command runs the -command, -handler, and -proxyfilter callbacks inside a catch command. Therefore an error in the callback command does not call the bgerror handler. When debugging one of these callbacks, it may be convenient to report errors by using a catch command within the callback command itself, e.g. to write an error message to stdout.

In any case, you must still call ::http::cleanup to delete the state array when you are done.

There are other possible results of the HTTP transaction determined by examining the status from ::http::status. These are described below.

If the HTTP transaction completes entirely, then status will be ok. However, you should still check the ::http::code value to get the HTTP status. The ::http::ncode procedure provides just the numeric error (e.g., 200, 404 or 500) while the ::http::code procedure returns a value like “HTTP 404 File not found”.
If the server closes the socket without replying, then no error is raised, but the status of the transaction will be eof.
The error message will also be stored in the error status array element, accessible via ::http::error.
A timeout occurred before the transaction could complete
user-reset

Another error possibility is that ::http::geturl is unable to write all the post query data to the server before the server responds and closes the socket. The error message is saved in the posterror status array element and then ::http::geturl attempts to complete the transaction. If it can read the server's response it will end up with an ok status, otherwise it will have an eof status.

STATE ARRAY

The ::http::geturl procedure returns a token that can be used to get to the state of the HTTP transaction in the form of a Tcl array. Use this construct to create an easy-to-use array variable:

upvar #0 $token state

Once the data associated with the URL is no longer needed, the state array should be unset to free up storage. The ::http::cleanup procedure is provided for that purpose. The following elements of the array are supported:

This is boolean true if (after decoding any compression specified by the “Content-Encoding” response header) the HTTP response is binary. It is boolean false if the HTTP response is text.
The contents of the URL. This will be empty if the -channel option has been specified. This value is returned by the ::http::data command.
The value of the charset attribute from the Content-Type meta-data value. If none was specified, this defaults to the RFC standard iso8859-1, or the value of $::http::defaultCharset. Incoming text data will be automatically converted from this charset to utf-8.
A copy of the Content-Encoding meta-data value.
The current number of bytes fetched from the URL. This value is returned by the ::http::size command.
If defined, this is the error string seen when the HTTP transaction was aborted.
The HTTP status reply from the server. This value is returned by the ::http::code command. The format of this value is:

HTTP/1.1 code string

The code is a three-digit number defined in the HTTP standard. A code of 200 is OK. Codes beginning with 4 or 5 indicate errors. Codes beginning with 3 are redirection errors. In this case the Location meta-data specifies a new URL that contains the requested information.

The HTTP protocol returns meta-data that describes the URL contents. The meta element of the state array is a list of the keys and values of the meta-data. This is in a format useful for initializing an array that just contains the meta-data:

array set meta $state(meta)

Some of the meta-data keys are listed below, but the HTTP standard defines more, and servers are free to add their own.

The type of the URL contents. Examples include text/html, image/gif, application/postscript and application/x-tcl.
The advertised size of the contents. The actual size obtained by ::http::geturl is available as state(currentsize).
An alternate URL that contains the requested data.
The error, if any, that occurred while writing the post query data to the server.
See description in the chapter ERRORS above for a list and description of status. During the transaction this value is the empty string.
A copy of the Content-Length meta-data value.
A copy of the Content-Type meta-data value.
The requested URL.

PERSISTENT CONNECTIONS

BASICS

See RFC 7230 Sec 6, which supersedes RFC 2616 Sec 8.1.

A persistent connection allows multiple HTTP/1.1 transactions to be carried over the same TCP connection. Pipelining allows a client to make multiple requests over a persistent connection without waiting for each response. The server sends responses in the same order that the requests were received.

If a POST request fails to complete, typically user confirmation is needed before sending the request again. The user may wish to verify whether the server was modified by the failed POST request, before sending the same request again.

A HTTP request will use a persistent socket if the call to http::geturl has the option -keepalive true. It will use pipelining where permitted if the http::config option -pipeline is boolean true (its default value).

The http package maintains no more than one persistent connection to each server (i.e. each value of “domain:port”). If http::geturl is called to make a request over a persistent connection while the connection is busy with another request, the new request will be held in a queue until the connection is free.

The http package does not support HTTP/1.0 persistent connections controlled by the Keep-Alive header.

SPECIAL CASES

This subsection discusses issues related to closure of the persistent connection by the server, automatic retry of failed requests, the special treatment necessary for POST requests, and the options for dealing with these cases.

In accordance with RFC 7230, http::geturl does not pipeline requests that use the POST method. If a POST uses a persistent connection and is not the first request on that connection, http::geturl waits until it has received the response for the previous request; or (if http::config option -postfresh is boolean true) it uses a new connection for each POST.

If the server is processing a number of pipelined requests, and sends a response header “Connection: close” with one of the responses (other than the last), then subsequent responses are unfulfilled. http::geturl will send the unfulfilled requests again over a new connection.

A difficulty arises when a HTTP client sends a request over a persistent connection that has been idle for a while. The HTTP server may half-close an apparently idle connection while the client is sending a request, but before the request arrives at the server: in this case (an “asynchronous close event”) the request will fail. The difficulty arises because the client cannot be certain whether the POST modified the state of the server. For HEAD or GET requests, http::geturl opens another connection and retransmits the failed request. However, if the request was a POST, RFC 7230 forbids automatic retry by default, suggesting either user confirmation, or confirmation by user-agent software that has semantic understanding of the application. The http::config option -repost allows for either possibility.

Asynchronous close events can occur only in a short interval of time. The http package monitors each persistent connection for closure by the server. Upon detection, the connection is also closed at the client end, and subsequent requests will use a fresh connection.

If the http::geturl command is called with option -keepalive true, then it will both try to use an existing persistent connection (if one is available), and it will send the server a “Connection: keep-alive” request header asking to keep the connection open for future requests.

The http::config options -pipeline, -postfresh, and -repost relate to persistent connections.

Option -pipeline, if boolean true, will pipeline GET and HEAD requests made over a persistent connection. POST requests will not be pipelined - if the POST is not the first transaction on the connection, its request will not be sent until the previous response has finished. GET and HEAD requests made after a POST will not be sent until the POST response has been delivered, and will not be sent if the POST fails.

Option -postfresh, if boolean true, will override the http::geturl option -keepalive, and always open a fresh connection for a POST request.

Option -repost, if true, permits automatic retry of a POST request that fails because it uses a persistent connection that the server has half-closed (an “asynchronous close event”). Subsequent GET and HEAD requests in a failed pipeline will also be retried. The -repost option should be used only if the application understands that the retry is appropriate - specifically, the application must know that if the failed POST successfully modified the state of the server, a repeat POST would have no adverse effect.

PROTOCOL UPGRADES

The HTTP/1.1 Connection and Upgrade client headers inform the server that the client wishes to change the protocol used over the existing connection (RFC 7230). This mechanism can be used to request a WebSocket (RFC 6455), a higher version of the HTTP protocol (HTTP 2), or TLS encryption. If the server accepts the upgrade request, its response code will be 101.

To request a protocol upgrade when calling http::geturl, the -headers option must supply appropriate values for Connection and Upgrade, and the -command option must supply a command that implements the requested protocol and can also handle the server response if the server refuses the protocol upgrade. For upgrade requests http::geturl ignores the value of option -keepalive, and always uses the value 0 so that the upgrade request is not made over a connection that is intended for multiple HTTP requests.

The Tcllib library websocket implements WebSockets, and makes the necessary calls to commands in the http package.

There is currently no native Tcl client library for HTTP/2.

The Upgrade mechanism is not used to request TLS in web browsers, because http and https are served over different ports. It is used by protocols such as Internet Printing Protocol (IPP) that are built on top of http(s) and use the same TCP port number for both secure and insecure traffic.

In browsers, opportunistic encryption is instead implemented by the Upgrade-Insecure-Requests client header. If a secure service is available, the server response code is a 307 redirect, and the response header Location specifies the target URL. The browser must call http::geturl again in order to fetch this URL. See https://w3c.github.io/webappsec-upgrade-insecure-requests/

EXAMPLE

This example creates a procedure to copy a URL to a file while printing a progress meter, and prints the meta-data associated with the URL.

proc httpcopy { url file {chunk 4096} } {

set out [open $file w]
set token [::http::geturl $url -channel $out \
-progress httpCopyProgress -blocksize $chunk]
close $out
# This ends the line started by httpCopyProgress
puts stderr ""
upvar #0 $token state
set max 0
foreach {name value} $state(meta) {
if {[string length $name] > $max} {
set max [string length $name]
}
if {[regexp -nocase ^location$ $name]} {
# Handle URL redirects
puts stderr "Location:$value"
return [httpcopy [string trim $value] $file $chunk]
}
}
incr max
foreach {name value} $state(meta) {
puts [format "%-*s %s" $max $name: $value]
}
return $token } proc httpCopyProgress {args} {
puts -nonewline stderr .
flush stderr }

SEE ALSO

safe(3tcl), socket(3tcl), safesock(3tcl)

KEYWORDS

internet, security policy, socket, www

2.9 http