Module Mimestring

module Mimestring: sig .. end
Low-level functions to parse and print mail and MIME messages

Mimestring contains a lot of functions to scan and print strings formatted as MIME messages. For a higher-level view on this topic, see the Netmime module.


Parsing and Printing Mail Headers

val scan_header : ?downcase:bool ->
?unfold:bool ->
?strip:bool ->
string -> start_pos:int -> end_pos:int -> (string * string) list * int
let params, header_end_pos = scan_header s start_pos end_pos:

Scans the mail header that begins at position start_pos in the string s and that must end somewhere before position end_pos. It is intended that in end_pos the character position following the end of the body of the MIME message is passed.

Returns the parameters of the header as (name,value) pairs (in params), and in header_end_pos the position of the character following directly after the header (i.e. after the blank line separating the header from the body).

The following normalizations have already been applied:

The default is to apply all three normalizations (D), (U), and (S) (for historic reasons). The three arguments downcase, unfold, and strip control which normalizations are performed (and for historic reasons, too, this is not what you would expect - backwards compatibility can sometimes be a burden):

This means that unfold not only removes CR/LF from the field value, but also removes whitespace at the beginning and at the end of the field value. strip causes not to remove CR/LF if it occurs somewhere within the field value, but all whitespace (including CR/LF) at the beginning of the field value and at the end of the field value is still deleted. Note that if you only want (S) you have to pass ~unfold:false and ~strip:true.

The rules to postprocess mail messages in MIME format are not applied (e.g. encoding transformations as indicated by RFC 2047).

The function fails if the header violates the header format strongly. (Some minor deviations are tolerated, e.g. it is sufficient to separate lines by only LF instead of CRLF.)

The Format of Mail Messages

Messages consist of a header and a body; the first empty line separates both parts. The header contains lines "param-name: param-value" where the param-name must begin on column 0 of the line, and the ":" separates the name and the value. So the format is roughly:

 param1-name: param1-value
 paramN-name: paramN-value

(Where "_" denotes an empty line.)

This function wants in start_pos the position of the first character of param1-name in the string, and in end_pos the position of the character following body. It returns as header_end_pos the position where body begins. Furthermore, in params all parameters are returned the function finds in the header.


Note that parameter values are restricted; you cannot represent arbitrary strings. The following problems can arise:

This implementation of a mail scanner tolerates a number of deviations from the standard: long lines are not rejected; 8 bit values are generally accepted; lines may be ended only with LF instead of CRLF.

Furthermore, the transformations (D), (U), and (S) can be performed resulting in values that are simpler to process.


This function can parse all mail headers that conform to RFC 822 or RFC 2822.

But there may be still problems, as RFC 822 allows some crazy representations that are actually not used in practice. In particular, RFC 822 allows it to use backslashes to "indicate" that a CRLF sequence is semantically meant as line break. As this function normally deletes CRLFs, it is not possible to recognize such indicators in the result of the function.

val read_header : ?downcase:bool ->
?unfold:bool ->
?strip:bool -> Netstream.in_obj_stream -> (string * string) list
This function expects that the current position of the passed in_obj_stream is the first byte of the header. The function scans the header and returns it. After that, the stream position is after the header and the terminating empty line (i.e. at the beginning of the message body).

The options downcase, unfold, and strip have the same meaning as in scan_header.


To read the mail message "file.txt":

 let ch = Netchannels.input_channel (open_in "file.txt") in
 let stream = Netstream.input_stream ch in
 let header = read_header stream in
 stream#close_in()  (* no need to close ch *)

val write_header : ?soft_eol:string ->
?eol:string -> Netchannels.out_obj_channel -> (string * string) list -> unit
This function writes the header to the passed out_obj_channel. The empty line following the header is also written.

Exact output format:

These rules ensure that the printed header will be well-formed with two exceptions:

These two problems cannot be addressed without taking the syntax of the header fields into account. See below how to create proper header fields from s_token lists.

Parsing Structured Values

The following types and functions allow it to build scanners for structured mail and MIME values in a highly configurable way.

Structured Values

RFC 822 (together with some other RFCs) defines lexical rules how formal mail header values should be divided up into tokens. Formal mail headers are those headers that are formed according to some grammar, e.g. mail addresses or MIME types.

Some of the characters separate phrases of the value; these are the "special" characters. For example, '@' is normally a special character for mail addresses, because it separates the user name from the domain name (as in user@domain). RFC 822 defines a fixed set of special characters, but other RFCs use different sets. Because of this, the following functions allow it to configure the set of special characters.

Every sequence of characters may be embraced by double quotes, which means that the sequence is meant as literal data item; special characters are not recognized inside a quoted string. You may use the backslash to insert any character (including double quotes) verbatim into the quoted string (e.g. "He said: \"Give it to me!\""). The sequence of a backslash character and another character is called a quoted pair.

Structured values may contain comments. The beginning of a comment is indicated by '(', and the end by ')'. Comments may be nested. Comments may contain quoted pairs. A comment counts as if a space character were written instead of it.

Control characters are the ASCII characters 0 to 31, and 127. RFC 822 demands that mail headers are 7 bit ASCII strings. Because of this, this module also counts the characters 128 to 255 as control characters.

Domain literals are strings embraced by '[' and ']'; such literals may contain quoted pairs. Today, domain literals are used to specify IP addresses (rare), e.g. user@[].

Every character sequence not falling in one of the above categories is an atom (a sequence of non-special and non-control characters). When recognized, atoms may be encoded in a character set different than US-ASCII; such atoms are called encoded words (see RFC 2047).

Scanning Using the Extended Interface

In order to scan a string containing a structured value, you must first create a mime_scanner using the function create_mime_scanner. The scanner contains the reference to the scanned string, and a specification how the string is to be scanned. The specification consists of the lists specials and scan_options.

The character list specials specifies the set of special characters. These are the characters that are not regarded as part of atoms, because they work as delimiters that separate atoms (like @ in the above example). In addition to this, when '"', '(', and '[' are seen as regular characters not delimiting quoted string, comments, and domain literals, respectively, these characters must also be added to specials. In detail, these rules apply:

If recognized, quoted strings are returned as QString s, where s is the string without the embracing quotes, and with already decoded quoted pairs.

Control characters c are returned as Control c.

If recognized, comments may either be returned as spaces (in the case you are not interested in the contents of comments), or as Comment tokens. The contents of comments are not further scanned; you must start a subscanner to analyze comments as structured values.

If recognized, domain literals are returned as DomainLiteral s, where s is the literal without brackets, and with decoded quoted pairs.

Atoms are returned as Atom s where s is a longest sequence of atomic characters (all characters which are neither special nor control characters nor delimiters for substructures). If the option Recognize_encoded_words is on, atoms which look like encoded words are returned as EncodedWord tokens. (Important note: Neither '?' nor '=' must be special in order to enable this functionality.)

After the mime_scanner has been created, you can scan the tokens by invoking scan_token which returns one token at a time, or by invoking scan_token_list which returns all following tokens.

There are two token types: s_token is the base type and is intended to be used for pattern matching. s_extended_token is a wrapper that additionally contains information where the token occurs.

Scanning Using the Simple Interface

Instead of creating a mime_scanner and calling the scan functions, you may also invoke scan_structured_value. This function returns the list of tokens directly; however, it is restricted to s_token.


type s_token =
| Atom of string
| EncodedWord of ((string * string) * string * string) (*Args: ((charset,lang),encoding,encoded_word)*)
| QString of string
| Control of char
| Special of char
| DomainLiteral of string
| Comment
| End
A token may be one of:

type s_option =
| No_backslash_escaping (*Do not handle backslashes in quoted string and comments as escape characters; backslashes are handled as normal characters. For example: The wrong qstring "C:\dir\file" will be returned as QString "C:\dir\file" when this option is in effect, and not as QString "C:dirfile" as by default. -- This is a common error in many MIME implementations.*)
| Return_comments (*Comments are returned as token Comment (unless '(' is included in the list of special characters, in which case comments are not recognized at all). You may get the exact location of the comment by applying get_pos and get_length to the extended token.*)
| Recognize_encoded_words (*Enables that encoded words are recognized and returned as EncodedWord instead of Atom.*)
type s_extended_token 
An opaque type containing the information of s_token plus:
val get_token : s_extended_token -> s_token
Return the s_token within the s_extended_token
val get_decoded_word : s_extended_token -> string
val get_charset : s_extended_token -> string
Return the decoded word (the contents of the word after decoding the "Q" or "B" representation), and the character set of the decoded word (uppercase).

These functions not only work for EncodedWord. The function get_decoded_word returns for the other kinds of token:

The function get_charset returns "US-ASCII" for them.
val get_language : s_extended_token -> string
Returns the language if the token is an EncodedWord, and "" for all other tokens.
val get_pos : s_extended_token -> int
Return the byte position where the token starts in the string (the first byte has position 0)
val get_line : s_extended_token -> int
Return the line number where the token starts (numbering begins usually with 1)
val get_column : s_extended_token -> int
Return the column of the line where the token starts (first column is number 0)
val get_length : s_extended_token -> int
Return the length of the token in bytes
val separates_adjacent_encoded_words : s_extended_token -> bool
True iff the current token is white space (i.e. Special ' ', Special '\t', Special '\r' or Special '\n') and the last non-white space token was EncodedWord and the next non-white space token will be EncodedWord.

The background of this function is that white space between encoded words does not have a meaning, and must be ignored by any application interpreting encoded words.

type mime_scanner 
The opaque type of a scanner for structured values
val create_mime_scanner : specials:char list ->
scan_options:s_option list ->
?pos:int -> ?line:int -> ?column:int -> string -> mime_scanner
Creates a new mime_scanner scanning the passed string.

specials : The list of characters recognized as special characters.
scan_options : The list of global options modifying the behaviour of the scanner
pos : The position of the byte where the scanner starts in the passed string. Defaults to 0.
line : The line number of this first byte. Defaults to 1.
column : The column number of this first byte. Default to 0.

Note for create_mime_scanner:

The optional parameters pos, line, column are intentionally placed after scan_options and before the string argument, so you can specify scanners by partially applying arguments to create_mime_scanner which are not yet connected with a particular string:

 let my_scanner_spec = create_mime_scanner my_specials my_options in
 let my_scanner = my_scanner_spec my_string in 

val get_pos_of_scanner : mime_scanner -> int
val get_line_of_scanner : mime_scanner -> int
val get_column_of_scanner : mime_scanner -> int
Return the current position, line, and column of a mime_scanner. The primary purpose of these functions is to simplify switching from one mime_scanner to another within a string:

 let scanner1 = create_mime_scanner ... s in
 ... now scanning some tokens from s using scanner1 ...
 let scanner2 = create_mime_scanner ... 
                  ?pos:(get_pos_of_scanner scanner1)
                  ?line:(get_line_of_scanner scanner1)
                  ?column:(get_column_of_scanner scanner1)
                  s in
 ... scanning more tokens from s using scanner2 ... 

Restriction: These functions are not available if the option Recognize_encoded_words is on. The reason is that this option enables look-ahead scanning; please use the location of the last scanned token instead.

Note: To improve the performance of switching, it is recommended to create scanner specs in advance (see the example my_scanner_spec above).

val scan_token : mime_scanner -> s_extended_token * s_token
Returns the next token, or End if there is no more token. The token is returned both as extended and as normal token.
val scan_token_list : mime_scanner ->
(s_extended_token * s_token) list
Returns all following tokens as a list (excluding End)
val scan_structured_value : string -> char list -> s_option list -> s_token list
This function is included for backwards compatibility, and for all cases not requiring extended tokens.

It scans the passed string according to the list of special characters and the list of options, and returns the list of all tokens.

val specials_rfc822 : char list
val specials_rfc2045 : char list
The sets of special characters defined by the RFCs 822 and 2045.

Parsing Certain Forms of Structured Values

val scan_encoded_text_value : string -> s_extended_token list
Scans a "text" value. The returned token list contains only Special, Atom and EncodedWord tokens. Spaces, TABs, CRs, LFs are returned (as Special) unless they occur between adjacent encoded words in which case they are suppressed. The characters '(', '[', and '"' are also returned as Special tokens, and are not interpreted as delimiters.

For instance, this function can be used to scan the "Subject" field of mail messages.

val scan_value_with_parameters : string -> s_option list -> string * (string * string) list
let name, params = scan_value_with_parameters s options: Scans values with annotations like name ; p1=v1 ; p2=v2 ; ... For example, MIME types like "text/plain;charset=ISO-8859-1" can be parsed.

The values may or may not be quoted. The characters ";", "=", and even "," are only accepted as part of values when they are quoted. On sytax errors, the function fails.

RFC 2231: This function supports some features of this RFC: Continued parameter values are concatenated. For example:

 Content-Type: message/external-body; access-type=URL;

This is returned as:

   [ ("access-type", "URL");
     ("URL", "") ]

However, encoded parameter values are not handled specially. The parameter title*=us-ascii'en-us'This%20is%20%2A%2A%2Afun%2A%2A%2A would be returned as ("title*", "us-ascii'en-us'This%20is%20%2A%2A%2Afun%2A%2A%2A"). Use scan_values_with_parameters_ep instead (see below).

Raises Failure on syntax errors.

type s_param 
The type of encoded parameters (RFC 2231)
val param_value : s_param -> string
val param_charset : s_param -> string
val param_language : s_param -> string
Return the decoded value of the parameter, the charset (uppercase), and the language. If the charset is not available, "" will be returned. If the language is not available, "" will be returned.
val mk_param : ?charset:string -> ?language:string -> string -> s_param
Creates a parameter from a value (in decoded form). The parameter may have a charset and a language.
val print_s_param : Format.formatter -> s_param -> unit
Prints a parameter to the formatter (as toploop printer)
val scan_value_with_parameters_ep : string ->
s_option list -> string * (string * s_param) list
let name, params = scan_value_with_parameters_ep s options: This version of the scanner copes with encoded parameters according to RFC 2231. Note: "ep" means "encoded parameters".

Example: doc.html;title*=us-ascii'en-us'This%20is%20%2A%2A%2Afun%2A%2A%2A

The parameter title would be returned as:

Raises Failure on syntax errors.
val scan_mime_type : string -> s_option list -> string * (string * string) list
let name, params = scan_mime_type s options: Scans MIME types like text/plain; charset=iso-8859-1 The name of the type and the names of the parameters are converted to lower case.

Raises Failure on syntax errors.

val scan_mime_type_ep : string ->
s_option list -> string * (string * s_param) list
let name, params = scan_mime_type_ep s options: This version copes with RFC-2231-encoded parameters.

Raises Failure on syntax errors.

val split_mime_type : string -> string * string
let (main_type, sub_type) = split_mime_type content_type: Splits the MIME type into main and sub type, for example split_mime_type "text/plain" = ("text", "plain") . The returned strings are always lowercase.

Raises Failure on syntax errors.

Printing Structured Values

exception Line_too_long
Raised when the hard limit of the line length is exceeded
val write_value : ?maxlen1:int ->
?maxlen:int ->
?hardmaxlen1:int ->
?hardmaxlen:int ->
?fold_qstring:bool ->
?fold_literal:bool ->
?unused:int Pervasives.ref ->
?hardunused:int Pervasives.ref ->
Netchannels.out_obj_channel -> s_token list -> unit
Writes the list of s_token to the out_obj_channel. The value is optionally folded into several lines while writing, but this is off by default. To enable folding, pass both maxlen1 and maxlen: The maxlen1 parameter specifies the length of the first line to write, the maxlen parameter specifies the length of the other lines.

If enabled, folding tries to ensure that the value is written in several lines that are not longer as specified by maxlen1 and maxlen. The value is split into lines by inserting "folding space" at certain locations (which is usually a linefeed followed by a space character, see below). The following table specifies between which tokens folding may happen:

 1st   \   2nd | Atom | QString | DLiteral | EncWord | Special | Spec ' '|
          Atom | FS   |  FS     |   FS     |   FS    |    -    |    F    |
       QString | FS   |  FS     |   FS     |   FS    |    -    |    F    |
 DomainLiteral | FS   |  FS     |   FS     |   FS    |    -    |    F    |
   EncodedWord | FS   |  FS     |   FS     |   FS    |    -    |    F    |
       Special | -    |  -      |   -      |   -     |    -    |    F    |
   Special ' ' | -    |  -      |   -      |   -     |    -    |    -    |

The table shows between which two types of tokens a space or a folding space is inserted:

Folding space is "\n ", i.e. only LF, not CRLF is used as end-of-line character. The function write_header will convert these LF to CRLF if needed.

Special '\t' is handled like Special ' '. Control characters are just printed, without folding. Comments, however, are substituted by either space or folding space. The token End is ignored.

Furthermore, folding may also happen within tokens:

It is possible to get the actual number of characters back that can still be printed into the last line without making the line too long. Pass an int ref as unused to get this value (it may be negative!). Pass an int ref as hardunused to get the number of characters that may be printed until the hard limit is exceeded.

The function normally does not fail when a line becomes too long, i.e. it exceeds maxlen1 or maxlen. However, it is possible to specify a hard maximum length (hardmaxlen1 and hardmaxlen). If these are exceeded, the function will raise Line_too_long.

For electronic mail, a maxlen of 78 and a hardmaxlen of 998 is recommended.

Known Problems:

Further Tips:
val param_tokens : ?maxlen:int -> (string * s_param) list -> s_token list
Formats a parameter list. For example, [ "a", "b"; "c", "d" ] is transformed to the token sequence corresponding to ; a=b; c=d. If maxlen is specified, it is ensured that the individual parameter (e.g. "a=b;") is not longer than maxlen-1, such that it will fit into a line with maximum length maxlen. By default, no maximum length is guaranteed. If maxlen is passed, or if a parameter specifies a character set or language, the encoding of RFC 2231 will be applied. If these conditions are not met, the parameters will be encoded traditionally.
val split_uri : string -> s_token list
Splits a long URI according to the algorithm of RFC 2017. The input string must only contain 7 bit characters, and must be, if necessary, already be URL-encoded.

Scanning MIME Messages

val scan_multipart_body : string ->
start_pos:int ->
end_pos:int -> boundary:string -> ((string * string) list * string) list
let [params1, value1; params2, value2; ...] = scan_multipart_body s start_pos end_pos boundary:

Scans the string s that is the body of a multipart message. The multipart message begins at position start_pos in s, and end_pos is the position of the character following the message. In boundary the boundary string must be passed (this is the "boundary" parameter of the multipart MIME type, e.g. multipart/mixed;boundary="some string" ).

The return value is the list of the parts, where each part is returned as pair (params, value). The left component params is the list of name/value pairs of the header of the part. The right component is the raw content of the part, i.e. if the part is encoded ("content-transfer-encoding"), the content is returned in the encoded representation. The caller is responsible for decoding the content.

The material before the first boundary and after the last boundary is not returned.

Multipart Messages

The MIME standard defines a way to group several message parts to a larger message (for E-Mails this technique is known as "attaching" files to messages); these are the so-called multipart messages. Such messages are recognized by the major type string "multipart", e.g. multipart/mixed or multipart/form-data. Multipart types MUST have a boundary parameter because boundaries are essential for the representation.

Multipart messages have a format like (where "_" denotes empty lines):

 Content-type: multipart/xyz; boundary="abc"
 Body begins here ("prologue")
 ...Header part 1...
 ...Body part 1...
 ...Header part 2...
 ...Body part 2

The parts are separated by boundary lines which begin with "--" and the string passed as boundary parameter. (Note that there may follow arbitrary text on boundary lines after "--abc".) The boundary is chosen such that it does not occur as prefix of any line of the inner parts of the message.

The parts are again MIME messages, with header and body. Note that it is explicitely allowed that the parts are even multipart messages.

The texts before the first boundary and after the last boundary are ignored.

Note that multipart messages as a whole MUST NOT be encoded. Only the PARTS of the messages may be encoded (if they are not multipart messages themselves).

Please read RFC 2046 if want to know the gory details of this brain-dead format.

val scan_multipart_body_and_decode : string ->
start_pos:int ->
end_pos:int -> boundary:string -> ((string * string) list * string) list
Same as scan_multipart_body, but decodes the bodies of the parts if they are encoded using the methods "base64" or "quoted printable". Fails, if an unknown encoding is used.
val scan_multipart_body_from_netstream : Netstream.in_obj_stream ->
boundary:string ->
create:((string * string) list -> 'a) ->
add:('a -> Netstream.in_obj_stream -> int -> int -> unit) ->
stop:('a -> unit) -> unit
scan_multipart_body_from_netstream s boundary create add stop:

Reads the MIME message from the netstream s block by block. The parts are delimited by the boundary.

Once a new part is detected and begins, the function create is called with the MIME header as argument. The result p of this function may be of any type.

For every chunk of the part that is being read, the function add is invoked: add p s k n.

Here, p is the value returned by the create invocation for the current part. s is the netstream. The current window of s contains the read chunk completely; the chunk begins at position k of the window (relative to the beginning of the window) and has a length of n bytes.

When the part has been fully read, the function stop is called with p as argument.

That means, for every part the following is executed:

Important Precondition: Exceptions:
val read_multipart_body : (Netstream.in_obj_stream -> 'a) ->
string -> Netstream.in_obj_stream -> 'a list
This is the "next generation" multipart message parser. It is called as follows:

let parts = read_multipart_body f boundary s

As precondition, the current position of the stream s must be at the beginning of the message body. The string boundary must be the message boundary (without "--"). The function f is called for every message part, and the resulting list parts is the concatentation of the values returned by f.

The stream passed to f is a substream of s that begins at the first byte of the header of the message part. The function f can read data from the substream as necessary. The substream terminates at the end of the message part. This means that f can simply read the data of the substream from the beginning to the end. It is not necessary that f reads the substream until EOF, however.

After all parts have been read, the trailing material of stream s is skipped until EOF of s is reached.

Helpers for MIME Messages

val create_boundary : ?random:string list -> ?nr:int -> unit -> string
Creates a boundary string that can be used to separate multipart messages. The string is 63 characters long and has the following "features":