This document is intended for people who suddenly have to or want to implement an application that automates something usually done with a browser, but are missing the background to understand what they actually need to do. It provides guidance on the steps required to implement a program that interacts with a web site which is designed to be used with a browser. It does not save you from eventually learning the background of what you are doing, but it should help you to get started quickly and learn the details later.
This document has evolved from discussions on the HttpClient mailing lists. Although it refers to HttpClient, the concepts described here apply equally to HttpComponents or SUN's HttpURLConnection or any other HTTP communication library for any programming language. So you might find it useful even if you're not using Java and HttpClient.
The existence of this document does not imply that the HttpClient community feels responsible for teaching you how to program a client HTTP application. It is merely a way for us to reduce the noise on the mailing list without just leaving the newbies out in the cold.
Let's assume that you have some kind of repetitive, web-based task that you want to automate. Something like:
At this time, we don't have a specific example which could be developed into a sample application. So this document is all bla-bla, and you will have to work out the details - all the details - yourself. Such is life.
This scenario describes a hobbyist usage of HTTP, in other words: a bad practice. Web sites are designed for user interaction, not as an application programming interface (API). The interface of a web site is the user interface displayed by a browser. The HTTP communication between the browser and the server is an internal API, subject to change without notice.
A web site can be redesigned at any point in time. The server then sends different documents and a browser will display the new content. The user easily adjusts to click the appropriate links, and the browser communicates via HTTP as specified by the new documents from the server. Your application that only mimicks a browser will simply break.
Nevertheless, implementing this scenario will help you to get familiar with HTTP communication. It is also "good enough" for hobbyists applications, for example if you want to download the latest installment of your favorite daily webcomic to install it as the screen background. There is no big damage if such an application breaks.
If you want to implement a solid application, you should use only published APIs. For example, to check for new mail on your webmail account, you should ask the webmail provider for POP or IMAP access. These are standardized protocols supported my most EMail client applications. If you want to have a newsticker, look for RSS feeds from the provider and applications that display them.
As another example, if you want to perform a web search, there are search companies that provide an API for using their search engines. Unlike the examples before, such APIs are proprietary. You will still have to implement an application, but then you are using a published API that the provider will not change without notice.
HttpClient is not a browser. Here's the difference.
In the center of the figure you can find internal components. Browsers have a cache of recently accessed documents and image files. They need to remember cookies and passwords entered by the user. Such information can be kept in memory or stored persistently in the file system at the bottom of the figure, to be available again when the browser is restarted. Certificates for secure communication are almost always stored persistently. To the right of the figure is the network. Browsers support many protocols on different levels of abstraction. There are application protocols such as FTP and HTTP to retrieve documents from servers, and transport layer protocols such as TLS/SSL and Socks to establish connections for the application protocols.
One characteristic of browsers that is not shown in the figure is tolerance for bad input. There needs to be tolerance for invalid user input to make the browser user friendly. There also needs to be tolerance for malformed documents retrieved from servers, and for flaws in server behavior when executing protocols, to make as many websites as possible accessible to the user.
HTTP ClientHTTP Client" />
The figure shows some of the components you will find in a browser, and highlights the scope of HttpClient. The primary responsibility of HttpClient is the HTTP protocol, executed directly or through an HTTP proxy. It provides interfaces and default implementations for cookie and password management, but not for persisting such data. User interfacing, HTML parsing, plugins or non-HTTP application level protocols are not in the scope of HttpClient. It does provide interfaces to plug in transport layer protocols, but it does not implement such protocols.
All the rest of a browser's functionality you require needs to be provided by your application. HttpClient executes HTTP requests, but it will not and can not assemble them. Since HttpClient does not interface with the user, nor interpret content such as HTML files, there is little or no tolerance for bad data passed to the API. There is some tolerance for flaws in server behavior, but there are limits to the deviations HttpClient can handle.
This section introduces some important terms you have to know to understand the rest of this document.
consists of a header section and an optional entity. There are two kinds of messages, requests and responses. They differ in the format of the first line, but both can have header fields and an optional entity.
is sent from a client to a server. The first line includes the URI for which the request is sent, and a method that the server should execute for the client.
is sent from a server to a client in response to a request. The first line includes a status code that tells about success or failure of the request. HTTP defines a set of status codes, like 200 for success and 404 for not found. Other protocols based on HTTP can define additional status codes.
is an operation requested from the server. HTTP defines a set of operations, the most frequent being GET and POST. Other protocols based on HTTP can define additional methods.
are name-value pairs, where both name and value are text. The name of a header field is not case sensitive. Multiple values can be assigned to the same name. RFC 2616 defines a wide range of header fields for handling various aspects of the HTTP protocol. Other specifications, like RFC 2617 and RFC 2965, define additional headers. Some of the defined headers are for general use, others are meant for exclusive use with either requests or responses, still others are meant for use only with an entity.
is data sent with an HTTP message. For example, a response can contain the page or image you are downloading as an entity, or a request can include the parameters that you entered into a web form. The entity of an HTTP message can have an arbitrary data format, which is usually specified as a MIME type in a header field.
is a series of requests from a single source to a server. The server can keep session data, and needs to recognize the session to which each incoming request belongs. For example, if you execute a web search, the server will only return one page of search results. But it keeps track of the other results and makes them available when you click on the link to the "next" page. The server needs to know from the request that it is you and your session for which more results are requested, and not me and my session. That's because I searched for something else.
are the preferred way for servers to track sessions. The server supplies a piece of data, called a cookie, in response to a request. The server expects the client to send that piece of data in a header field with each following request of the same session. The cookie is different for each session, so the server can identify to which session a request belongs by looking at the cookie. If the cookie is missing from a request, the server will not respond as expected.
Create and execute a GET request for the login page. Just use the link you would type into the browser as the URL. This is what a browser does when you enter a URL in the address bar or when you click on a link that points to another web page.
Inspect the response from the server:
It should be sent as the entity of the response to your request. The entity is also referred to as the response body.
Cookies are sent in a header field named Set-Cookie or Set-Cookie2. It is possible that you don't get a session cookie until you log in. If there is no session cookie in the response, you'll have to do perform step 2 later, after you reach the point where the cookie is set.
If you do not get the page you expect, check the URL you are requesting. If it is correct, the server may use a browser detection. You will have to set the header field User-Agent to a value used by a popular browser to pretend that the request is coming from that browser.
If you can't get the login page, get the home page instead now. Get the login page in the next step, when you establish the session.
Create and execute another GET request for a page. You can simply request the login page again, or some other page of which you know the URL. Do NOT try to get a page which would be returned in response to submitting a web form. Use something you can reach simply by clicking on a link in the browser. Something where you can see the URL in the browser status line while the mouse pointer is hovering over the link.
This step is important when developing the application. Once you know that your application does establish the session correctly, you may be able to remove it. Only if you couldn't get the login page directly and had to get the home page first, you know you have to leave it in.
Inspect the request being sent to the server.
You can see what is sent to the server by enabling the wire log for HttpClient. You only need to see the request headers, not the body. The session cookie should be sent in a header field called Cookie. There may be several of those, and other cookies might be sent as well.
Inspect the response from the server:
You should not get another session cookie. If you get the same session cookie as before, the server behaves a little strange but that should not be a problem. If you get a new session cookie, then the server did not recognize the session for the request. Usually, this happens if the request did not contain the session cookie. But servers might use other means to track sessions, or to detect session hijacking.
If the session cookie is not sent in the request, one of two things has gone wrong. Either the cookie was not detected in the previous response, or the cookie was not selected for being sent with the new request.
If the cookie is accepted from the previous response but still not sent with the new request, make sure that HttpClient uses the same cookie store object. Unless you explicitly manage cookie store objects (not recommended for newbies!), this will be the case if you use the same HttpClient object to execute both requests.
If the cookie is still not sent with the request, make sure that the URL you are requesting is in the scope for the cookie. Cookies are only sent to the domain and path specified in the cookie scope. A cookie for host "jakarta.apache.org" will not be sent to host "tomcat.apache.org". A cookie for domain ".apache.org" will be sent to both. A cookie for host "apache.org", without the leading dot, will not be sent to "jakarta.apache.org". The latter case can be resolved by using a different cookie spec that adds the leading dot. In the other cases, use a URL that in the cookie scope to establish the session.
If the session cookie is sent with the request, but a new session cookie is set in the response anyway, check whether there are cookies other than the session cookie in the request. Some servers are incapable of detecting multiple cookies sent in individual header fields. HttpClient can be advised to put all cookies into a single header field.
If that doesn't help, you are in trouble. The server may use additional means to track the session, for example the header field named Referer. Set that field to the URL of the previous request. (see this mail)
If that doesn't help either, you will have to compare the request from your application to a corresponding one generated by a browser. The instructions in step 5 for POST requests apply for GET requests as well. It's even simpler with GET, since you don't have an entity.
Now it is time to analyze the form defined in the HTML markup of the page. A form in HTML is a set of name-value-pairs called parameters, where some of the values can be entered in the browser. By analyzing the HTML markup, you can learn which parameters you have to define and how to send them to the server.
Look for the form tag in the page source. There may be several forms in the page, but they can not be nested. Locate the form you want to submit. Locate the matching /form tag. Everything in between the two may be relevant. Let's start with the attributes of the form tag:
specifies the method used for submitting the form. If it is GET or not specified at all, then you need to create a GET request. The parameters will be added as a query string to the URL. If the method is POST, you need to create a POST request. The parameters will be put in the entity of the request, also referred to as the request body. How to do that is discussed in step 5.
specifies the URL to which the request has to be sent. Do not try to get this URL from the address bar of your browser! A browser will automatically follow redirects and only displays the final URL, which can be different from the URL in this attribute. It is possible that the URL includes a query string that specifies some parameters. If so, keep that in mind.
specifies the MIME type for the entity of the request generated by the form. The two common cases are url-encoded (default) and multipart-mime. Note that these terms are just informally used here, the exact values that need to be written in an HTML document are specified elsewhere. This attribute is only used for the POST method. If the method is GET, the parameters will always be url-encoded, but not in an entity.
specifies the character set that the browser should allow for user input. It will not be discussed here, but you will have to consider this value if you experience charset related problems.
Except for optional query parameters in the action attribute, the parameters of a form are specified by HTML tags between form and /form. The following is a list of tags that can be used to define parameters. Except where stated otherwise, they have a name attribute which specifies the name of the parameter. The value of the parameter usually depends on user input.
<input type="text" name="..."> <input type="password" name="...">
specify single-line input fields. Using the return key in one of these fields will submit the form, so the value really is a single line of input from the user.
<input type="text" readonly name="..." value="..."> <input type="hidden" name="..." value="...">
specify a parameter that can not be changed by the user. The value of the parameter is given by the value attribute.
<input type="radio" name="..." value="..."> <input type="checkbox" name="..." value="...">
specify a parameter that can be included or omitted. There usually is more than one tag with the same name. For radio buttons, only one can be selected and the value of the parameter is the value of the selected radio button. For checkboxes, more than one can be selected. There will be one name-value-pair for each selected checkbox, with the same name for all of them.
<input type="submit" name="..." value="..."> <button type="submit" name="..." value="...">
specify a button to submit the form. The parameter will only be added to the form if that button is used to submit. If another button is used, or the form is submitted by pressing the return key in a text input field, the parameter is not part of the submitted form data. If the name attribute is missing, no parameter is added to the form data for that button.
<textarea name="..."> <textarea value="..." readonly>
specify a multi-line input field. In the readonly case, the value of the parameter is the text between the textarea and /textarea tags.
<select name="..." multiple>}}} <option value="...">...</option>}}} <option value="...">...</option>}}} ... </select>
specify a selection list or drop-down menu. If the multiple attribute is not present, only one option can be selected. There will be one name-value-pair for each selected option, with the same name for all of them. If there is no value attribute, the value for that option is the text between option and /option.
<input type="image" name="...">
specifies an image that can be clicked to submit the form. If that image is clicked to submit the form, two parameters are added to the form data. The name attribute is suffixed with ".x" and ".y", the values for the parameters are the relative coordinates of the mouse pointer within the image at the time of the click, in pixel. If the name attribute is missing, no parameters will be added to the form data.
<input type="file" name="...">
specifies a file selection box. The user can select a file that should be sent as part of the form data. This is only possible if the encoding is multipart-mime. Unlike other parameters, the file is not mapped to a simple name-value-pair. File upload is not a topic for beginners.
After you have determined the action URL and name-value-pairs of a form, you should exit the program you used to get the HTML source, start it again and repeat the analysis with the new page.
Most parameters will be the same for both pages. But some parameters, in particular those from hidden input fields, may change from session to session, or even with every request. The same can be the case with the action URL.
Parameters that remain the same can be hard-coded in your program. If parameters change (except for user input), then your application has to request the page with the form and extract the dynamic parameters at runtime. If you're lucky you can locate them by simple string searches. If you're unlucky, you need an HTML parser to make sense of the page. HTML parsing is out of scope for HttpClient, but you'll find some HTML parsers mentioned in the mailing list archives.
Note that a redesign of the form on the server can break your application at any time. Whenever that happens, you have to repeat the analysis with the new form returned by the server after the redesign, and adjust your application accordingly.
After analyzing the form, it is time to create a request that matches what a browser would generate. If the method is GET, just add the name-value-pairs for all parameters to the query string. If the method is POST, things are a little more complicated.
It depends on the server how closely you have to match browser behavior. For example, a servlet will not distinguish between parameters in the query string and url-encoded parameters of the entity. But other server side code might make that distinction. The safe way is always to match browser behavior exactly.
HttpClient supports both encoding types, url-encoded and multipart-mime. To send parameters url-encoded, use the POST request and add the parameters directly there. To send parameters in multipart-mime, collect the parameters in a multipart-encoded request entity and add set the entity for the POST request. You will also find support for file upload in the multipart package. Note that these techniques are mutually exclusive, they can not be combined. Parameters defined in the query string of the URL can remain there.
Send the request. Inspect the response from the server:
That is called a redirect. Follow redirects to the ultimate page and inspect that response. See step 6 on following redirects.
If the server response to your POST request indicates a problem, try to enable or disable the expect-continue handshake, or switch the protocol version to HTTP/1.0. If that doesn't help...
Inspect the request you are sending:
There is a variety of sniffer programs you can use to grep the browser request. Some of them are mentioned in the responses to on the mailing list.
Candidates for problems are missing or wrong parameters, and differences in the header fields. The parameters are all up to you. As a general rule for the header fields, you should send the same as the browser does. The order of the fields does not matter.
But there's a caveat: some header fields are controlled by HttpClient and can not be set explicitly. Other header fields are used to indicate capabilities which a browser has, but your application probably has not. For these, the request from your application has to and should differ. Here is a possibly incomplete list of headers that need special consideration:
controlled by HttpClient. The value is usually obtained from the URL you are posting to. It is possible to set a different value, called a "virtual host".
controlled by HttpClient. The values are obtained from the request entity.
usually controlled by HttpClient to handle connection keep-alive. Leave it alone or set the value to "close".
used to indicate the capability to process compressed responses. Do not set this, unless you are prepared to implement decompression.
It is quite common for servers to respond with a 303 or 307 status code to a POST request. These redirects indicate that your application has to send another request to retrieve the actual result of the operation you have triggered with the POST request.
HttpClient can be configured to follow some redirects automatically. Others it is not allowed to follow automatically, since RFC 2616 specifies that a user interaction should take place. We will make sure that HttpClient is compliant with this requirement, but we can't stop you from implementing a different behavior in your application. The Location header field in the redirect response indicates the URL from which to fetch the actual page. It is common practice that servers return a relative URL as the location, although the specification requires an absolute URL.
Note that there may be more than one redirect in succession. Your application then has to follow the redirect for a redirect, but make sure that you do not enter an infinite loop. If you find that there are more than two redirects in succession, something probably is fishy.
Your application can send as many GET and POST requests and follow as many redirects as is required. But you should remember that there is a session tracked by the server. Once your application is done, and if the web site does provide a logout link, you should send a final request to log out. This will tell the server that the session data can be discarded. If the server prevents multiple logins with the same user ID and your application has to run repeatedly, logout may even be required.
ReferenceMaterials: a list of technical specifications for HTTP and related stuff.