[manual index][section index]

NAME

format - structured data interchange

SYNOPSIS

include "sys.m";
include "bufio.m";
include "sexprs.m";
include "format.m";
format := load Format Format->PATH;

Fmtspec: adt {
	name:   string;
	fields: cyclic array of Fmtspec;
};
Fmt: adt {
	kind:   int;
	fields: cyclic array of Fmt;
};
Fmtval: adt {
	text:   fn(v: self Fmtval): string;
	val:    ref Sexprs->Sexp;
	recs:   cyclic array of array of Fmtval;
};
Fmtfile: adt {
	spec:   array of Fmtspec;
	descr:  array of byte;

	new:    fn(spec: array of Fmtspec): Fmtfile;
	open:   fn(f: self Fmtfile, name: string):
               ref Bufio->Iobuf;
	read:   fn(f: self Fmtfile, iob: ref Bufio->Iobuf):
               (array of Fmtval, string);
};
init:     fn();
spec2se:  fn(spec: array of Fmtspec): list of ref Sexprs->Sexp;
spec2fmt: fn(spec: array of Fmtspec): array of Fmt;
se2fmt:   fn(spec: array of Fmtspec, se: ref Sexprs->Sexp):
             (array of Fmt, string);
rec2val:  fn(spec: array of Fmtspec, rec: ref Sexprs->Sexp):
             (array of Fmtval, string);

DESCRIPTION

Format provides support for programs that wish to marshal and unmarshal structured data. It is designed to enable a client to request that the structure data is provided by the server in a particular format, and for the server to be able to check that it is capable of providing that format.

A record consists of a set of fields, each represented by one element in an sexprs(2) list. The content of a field can be a simple value, or it can hold a list containing any number of sub-records, each holding a set of fields, recursively defined as above.

The Fmtspec type defines the format for a field in a record. Name gives the name of the field, and fields gives the structure of the fields in any sub-records it contains (for a field with a simple value, this will be nil). Thus an array of Fmtspec values can define the structure of all the fields in a record. Here is an example structure specification:

Name, Address, Phone: con iota;
Number, Kind: con iota;
spec := array[] of {
	Address => Fmtspec("address", nil),
	Name => Fmtspec("name", nil),
	Phone => Fmtspec("phone", array[] of {
		Kind => Fmtspec("kind", nil),
		Number => Fmtspec("number", nil),
	}),
};

By placing each field in the structure specification at a known index, a link is made from the symbolic constants in the program to the textual field names.

A structure specification may also be represented by a list of S-expressions, where each member of the list names a field in the structure. If a member is itself a list, it specifies a field containing sub-records: its first member gives the name of the field, and subsequent members specify the fields in its sub-records. For example, the above specification could be written as the S-expression:

(name address (phone number kind))

The Fmt type also defines a record structure, but with respect to an existing Fmtspec structure specification. An Fmt value represents a field, and kind holds the index of that field in the original structure specification.

Se2fmt converts from an S-expression list, se (a structure specification), to a set of Fmt values. The specification must be a subset of spec; i.e. each field in se must exist in spec. Se2fmt returns a tuple (f, err). If the specification is bad, f will be nil, and err describes the error. Otherwise, each member of f gives a field specified by se. For example, given the above structure definition, after executing:

	se := Sexp.parse("(address (phone number))").t0;
	(f, err) := se2fmt(spec, se);

f[0].kind will hold the symbolic constant Address, and f[1].fields[0].kind will hold Number.

Spec2se converts from a structure representation to its S-expression equivalent. Spec2fmt converts it to an array of Fmt structures mirroring it (equivalent to, but more efficient than, se2fmt(spec2se(spec)).t0)

Data representation
The above specifications do not define a format for the representation of the actual data. For convenience however, Format defines its own S-expression-based data format. In this form, the fields in a record are represented by items in an S-expression list. A field containing sub-records is represented as a (possibly empty) list containing the sub-records; otherwise the value of a field may be an arbitrary S-expression.

For example, a record corresponding to the structure specification

(name address (phone kind number))

might look like:

("Jonny Dawes" "24 Crag Lane"
	((home "+44 7924 43535") (office "034 433 645")))

Rec2val cracks such a record, rec, into its component values, checking that it is structurally compatible with the specification, spec, and returning a tuple (fields, err). If it failed, fields will be nil, and err describes the error. Otherwise each member of fields, say v, holds the value of its equivalent field in spec. For fields without sub-records, v.val holds the field's value; otherwise v.recs holds an array with one member for each sub-record, each holding an array of fields defined recursively as above.

Some file servers provide files to which a format specification may be written, and which provide a sequence of records in that format when read. The Fmtfile type provides support for such files. It provides the following operations:


Fmtfile.new(spec)
returns a new Fmtfile value that can be used to open files and read records conforming to the given spec.
f.open(name)
opens such a file, writes the format specification to it, and returns an Iobuf (see bufio(2)) ready for reading the file.
f.read
reads a record from the file; its return is the same as that from rec2val.

SOURCE

/appl/lib/format.b

SEE ALSO

sexprs(2), bufio(2), sexprs(6)

FORMAT(2 ) Rev:  Tue Mar 31 02:42:39 GMT 2015