diff options
Diffstat (limited to 'doc/white-papers/mail')
| -rw-r--r-- | doc/white-papers/mail/camel.sgml | 339 | ||||
| -rw-r--r-- | doc/white-papers/mail/ibex.sgml | 158 |
2 files changed, 0 insertions, 497 deletions
diff --git a/doc/white-papers/mail/camel.sgml b/doc/white-papers/mail/camel.sgml deleted file mode 100644 index a339909f54..0000000000 --- a/doc/white-papers/mail/camel.sgml +++ /dev/null @@ -1,339 +0,0 @@ -<!doctype article PUBLIC "-//Davenport//DTD DocBook V3.0//EN" [ -<!entity Evolution "<application>Evolution</application>"> -<!entity Camel "Camel"> -]> - -<article class="whitepaper" id="camel"> - - <artheader> - <title>The &Camel; Messaging Library</title> - - <authorgroup> - <author> - <firstname>Dan</firstname> - <surname>Winship</surname> - <affiliation> - <address> - <email>danw@helixcode.com</email> - </address> - </affiliation> - </author> - - <author> - <firstname>Bertrand</firstname> - <surname>Guiheneuf</surname> - <affiliation> - <address> - <email>bertrand@helixcode.com</email> - </address> - </affiliation> - </author> - </authorgroup> - - <copyright> - <year>2000</year> - <holder>Helix Code, Inc.</holder> - </copyright> - - </artheader> - - <sect1 id="introduction"> - <title>Introduction</title> - - <para> - &Camel; is a generic messaging library. It is being used as the - back end for the mail component of &Evolution;. The name - "&Camel;" is an acronym; it refers to the fact that the - library is capable of going several days without food or water. - It means : Camel's Acronym Makes Everyone Laugh. - </para> - - <para> - &Camel;'s initial design is heavily based on Sun's - <trademark>JavaMail</trademark> API. It uses the Gtk+ object - system, and many of its classes are direct analags of JavaMail - classes. Its design has also been influenced by the features of - IMAP, and the limitations of the standard UNIX mbox mail store, - which set some of the boundaries on its requirements and - extensibility. - </para> - - <para> - &Camel; sees all message repositories as stores containing - folders. These folders in turn contain the messages the client - actually accesses. The use of such a unified interface allows - the client applications to be very extensible. &Camel; includes - an external provider mechanism which allows applications to - dynamically load and use protocols which were not available when - the application was initially written. - </para> - - <para> - The abstract store/folder mechanism is a powerful and versatile - way of accessing messages. No particular asumptions are made on - the client side, thus allowing new ways of managing the - messages. For example, the messages stored in the folders don't - necessarily have to share some common physical location. The - folder can be a purely virtual folder, containing only - references to the actual messages. This is used by the "vFolder" - provider, which allows you select messages meeting particular - criteria and deal with them as a group. - </para> - - <para> - In addition to these possibilities, &Camel; has full MIME - support. &Camel; MIME messages are lightweight objects - representing the MIME skeleton of the actual message. The data - contained in the subparts are never stored in memory except when - they are actually needed. The application, when accessing the - various MIME objects contained in the message (text parts, - attachments, embedded binary objects ...) asks &Camel; for a - stream that it can read data from. This scheme is particularly - useful with the IMAP provider. IMAP has strong MIME support - built-in, which allows &Camel; to download only the parts of - messages that it actually needs: attachments need not be - downloaded until they are viewed, and unnecessary - "multipart/alternative" parts will never be read off the server. - </para> - </sect1> - - <sect1 id="overview"> - <title>Overview</title> - - <graphic format="gif" fileref="camel"></graphic> - - <para> - To begin using &Camel;, an application first creates a - <classname>CamelSession</classname> object. This object is used - to store application defaults, and to coordinate communication - between providers and the application. - </para> - - <para> - A <classname>CamelProvider</classname> is a dynamically-loadable - module that provides functionality associated with a specific - service. Examples of providers are IMAP and SMTP. Providers - include subclasses of the various other &Camel; classes for - accessing and manipulating messages. - </para> - - <para> - <classname>CamelService</classname> is an abstract class for - describing a connection to a local or remote service. It - currently has two subclasses: <classname>CamelStore</classname>, - for services that store messages (such as IMAP servers and mbox - files), and <classname>CamelTransport</classname>, for services - that deliver messages (such as SMTP, or a local MTA). A provider - could also be both a store and a transport, as in the case of - NNTP. - </para> - - <para> - A <classname>CamelStore</classname> contains some number of - <classname>CamelFolder</classname> objects, which in turn - contain messages. A <classname>CamelFolder</classname> provides - a <classname>CamelFolderSummary</classname> object, which - includes details about the subject, date, and sender of each - message in the folder. The folder also includes the messages - themselves, as subclasses of <classname>CamelMedium</classname>. - </para> - - <para> - Email messages are represented by the - <classname>CamelMimeMessage</classname> class, a subclass of - <classname>CamelMedium</classname>. This class includes - operations for accessing RFC822 and MIME headers, accessing - subparts of MIME messages, encoding and decoding Base64 and - Quoted-Printable, etc. - </para> - - <para> - <classname>CamelTransport</classname> includes methods for - delivering messages. While the abstract - <function>CamelTransport::send</function> method takes a - <classname>CamelMedium</classname>, its subclasses may only be - able to deliver messages of specific - <classname>CamelMedium</classname> subclasses. For instance, - <classname>CamelSendmailTransport</classname> requires a - <classname>CamelMimeMessage</classname>, because it needs a - message that includes a "To:" header. A hypothetical - <classname>CamelNNTPTransport</classname> would need a - <classname>CamelNewsMessage</classname>, which would have a - "Newsgroups:" header. - </para> - - <para> - The content of messages are referred to using - <classname>CamelStream</classname> and its subclasses. In the - case of an mbox-based store, the - <classname>CamelStream</classname> would abstract the operation - of reading the correct section of the mbox file. For IMAP, - reading off the <classname>CamelStream</classname> might result - in commands being issued to the remote IMAP server and data - being read off a socket. - </para> - - <para> - The final major class in &Camel; is - <classname>CamelException</classname>, which is used to - propagate information about errors. Many methods take a - <classname>CamelException</classname> as an argument, which the - caller can then check if an error occurs. It includes both a - numeric error code which can be interpreted by the program, and - a text error message that can be displayed to the user. - </para> - </sect1> - - <sect1 id="classes"> - <title>Major Subcomponents</title> - - <sect2 id="store"> - <title>The Message Store</title> - - <para> - A <classname>CamelStore</classname> inherits the ability to - connect and authenticate to a service from its parent class, - <classname>CamelService</classname>. It then adds the ability - to retrieve folders. A store must contain at least one folder, - which can be retrieved with - <function>CamelStore::get_default_folder</function>. There are - also methods to retrieve the "top-level" folder (for - hieararchical stores), and to retrieve an arbitrary folder by - name. - </para> - - <para> - All <classname>CamelFolder</classname>s must implement certain - core operations, most notably generating a summary and - retrieving and deleting messages. A - <classname>CamelFolder</classname> must assign a permanently - unique identifier to each message it contains. Messages can - then be retrieved via - <function>CamelFolder::get_message_by_uid</function>. Alternately, - within a single mail-reading session, messages can be referred - to by their linear position within the store using - <function>CamelFolder::get_message_by_number</function>. - </para> - - <para> - Folders must also implement the - <function>get_parent_folder</function> and - <function>list_subfolders</function> methods. For stores that - don't allow multiple folders, they would return NULL and an - empty list, respectively. Stores that do allow multiple - folders will also define methods for creating and deleting - folders, and for moving messages between them (assuming the - folders are writable). - </para> - - <para> - Folders that support searching can define the - <function>search_by_expression</function> method. For mbox - folders, this is implemented by indexing the messages with the - ibex library and using that to search them later. For IMAP - folders, this uses the IMAP SEARCH command. Other folder types - might not be able to implement this functionality, in which - case users would not be able to do full-content searches on - them. - </para> - </sect2> - - <sect2 id="messages"> - <title>Messages</title> - - <para> - As mentioned before, messages are represented by subclasses of - <classname>CamelMedium</classname>. - <classname>CamelMedium</classname> itself is a subclass of - <classname>CamelDataWrapper</classname>, a generic class for - connecting a typed data source to a data sink. - <classname>CamelMedium</classname> adds the concept of message - headers versus message body. - (<classname>CamelDataWrapper</classname> has one other - important subclass, <classname>CamelMultipart</classname>, - which is used to provide separate access to the multiple - independent parts of a multipart MIME type.) - <classname>CamelMedium</classname>'s subclasses provide more - specialized handling of various headers: - <classname>CamelMimePart</classname> adds special handling for - the &ldquot;Content-*&rdquot; headers in MIME messages, and - its subclass <classname>CamelMimeMessage</classname> adds - handling for the RFC822 headers. - </para> - - <graphic format="gif" fileref="mimemessage"></graphic> - - <para> - Consider a message with two parts: a text part (in both plain - text and HTML), and an attached image: - - <programlisting> - - From: Dan Winship <danw@helixcode.com> - To: Matt Loper <matt@helixcode.com> - Subject: the Camel white paper - MIME-Version: 1.0 - Content-Type: multipart/mixed; - boundary="jhTYrnsRrdhDFGa" - - This is a multi-part message in MIME format. - --jhTYrnsRrdhDFGa - Content-Type: multipart/alternative; - boundary="sFSenbAFDSgDfg" - - --sFSenbAFDSgDfg - Content-Type: text/plain - - Hey, Matt - - Check out this graphic... - - -- Dan - - --sFSenbAFDSgDfg - Content-Type: text/html - - Hey, Matt<br> - <br> - Check out this graphic...<br> - <br> - -- Dan<br> - <br> - --sFSenbAFDSgDfg-- - - --jhTYrnsRrdhDFGa - Content-Type: image/png - Content-Transfer-Encoding: base64 - - F4JLw0ORrkRa8AwAMQJLAaI3UDIGsco9RAaB92... - --jhTYrnsRrdhDFGa-- - </programlisting> - - <para> - In &Camel;, this would be represented as follows: - </para> - - <graphic fileref="samplemsg"></graphic> - </sect2> - - <sect2 id="streams"> - <title>Streams</title> - - <para> - Streams are a generic data transport layer. Two basic stream - classes are <classname>CamelStreamFs</classname>, for - reading and writing files, and - <classname>CamelStreamMem</classname>, for reading from and - writing to objects that are already in memory. - </para> - - <para> - Streams can also be chained together. So a CamelMimePart - containing base64-encoded data can filter its output through - a CamelStreamB64. Other parts of the application that want - to read its data will never need to even realize that the - original data was encoded. - </para> - </sect2> - -</article> diff --git a/doc/white-papers/mail/ibex.sgml b/doc/white-papers/mail/ibex.sgml deleted file mode 100644 index dcb8f5ca4b..0000000000 --- a/doc/white-papers/mail/ibex.sgml +++ /dev/null @@ -1,158 +0,0 @@ -<!doctype article PUBLIC "-//Davenport//DTD DocBook V3.0//EN" [ -<!entity Evolution "<application>Evolution</application>"> -<!entity Camel "Camel"> -<!entity Ibex "Ibex"> -]> - -<article class="whitepaper" id="ibex"> - - <artheader> - <title>Ibex: an Indexing System</title> - - <authorgroup> - <author> - <firstname>Dan</firstname> - <surname>Winship</surname> - <affiliation> - <address> - <email>danw@helixcode.com</email> - </address> - </affiliation> - </author> - </authorgroup> - - <copyright> - <year>2000</year> - <holder>Helix Code, Inc.</holder> - </copyright> - - </artheader> - - <sect1 id="introduction"> - <title>Introduction</title> - - <para> - &Ibex; is a library for text indexing. It is being used by - &Camel; to allow it to quickly search locally-stored messages, - either because the user is looking for a specific piece of text, - or because the application is contructing a vFolder or filtering - incoming mail. - </para> - </sect1> - - <sect1 id="goals"> - <title>Design Goals and Requirements for Ibex</title> - - <para> - The design of &Ibex; is based on a number of requirements. - - <itemizedlist> - <listitem> - <para> - First, obviously, it must be fast. In particular, searching - the index must be appreciably faster than searching through - the messages themselves, and constructing and maintaining - the index must not take a noticeable amount of time. - </para> - </listitem> - - <listitem> - <para> - The indexes must not take up too much space. Many users have - limited filesystem quotas on the systems where they read - their mail, and even users who read mail on private machines - have to worry about running out of space on their disks. The - indexes should be able to do their job without taking up so - much space that the user decides he would be better off - without them. - </para> - - <para> - Another aspect of this problem is that the system as a whole - must be clever about what it does and does not index: - accidentally indexing a "text" mail message containing - uuencoded, BinHexed, or PGP-encrypted data will drastically - affect the size of the index file. Either the caller or the - indexer itself has to avoid trying to index these sorts of - things. - </para> - </listitem> - - <listitem> - <para> - The indexing system must allow data to be added to the index - incrementally, so that new messages can be added to the - index (and deleted messages can be removed from it) without - having to re-scan all existing messages. - </para> - </listitem> - - <listitem> - <para> - It must allow the calling application to explain the - structure of the data however it wants to, rather than - requiring that the unit of indexing be individual files. - This way, &Camel; can index a single mbox-format file and - treat it as multiple messages. - </para> - </listitem> - - <listitem> - <para> - It must support non-ASCII text, given that many people send - and receive non-English email, and even people who only - speak English may receive email from people whose names - cannot be written in the US-ASCII character set. - </para> - </listitem> - </itemizedlist> - - <para> - While there are a number of existing indexing systems, none of - them met all (or even most) of our requirements. - </para> - </sect1> - - <sect1 id="implementation"> - <title>The Implementation</title> - - <para> - &Ibex; is still young, and many of the details of the current - implementation are not yet finalized. - </para> - - <para> - With the current index file format, 13 megabytes of Info files - can be indexed into a 371 kilobyte index file—a bit under - 3% of the original size. This is reasonable, but making it - smaller would be nice. (The file format includes some simple - compression, but <application>gzip</application> can compress an - index file to about half its size, so we can clearly do better.) - </para> - - <para> - The implementation has been profiled and optimized for speed to - some degree. But, it has so far only been run on a 500MHz - Pentium III system with very fast disks, so we have no solid - benchmarks. - </para> - - <para> - Further optimization (of both the file format and the in-memory - data structures) awaits seeing how the library is most easily - used by &Evolution;: if the indexes are likely to be kept in - memory for long periods of time, the in-memory data structures - need to be kept small, but the reading and writing operations - can be slow. On the other hand, if the indexes will only be - opened when they are needed, reading and writing must be fast, - and memory usage is less critical. - </para> - - <para> - Of course, to be useful for other applications that have - indexing needs, the library should provide several options, so - that each application can use the library in the way that is - most suited for its needs. - </para> - </sect1> -</article> |