.STEP File Format – Part 21 syntax

STEP product model

ISO 10303-21, “Part 21”, defines the clear text encoding of a STEP file: how data that is described by an EXPRESS schema is written as text. It does not define the schema, so a file is only as meaningful as the schema it names (see Schemas and core entities). The standard is sold by ISO; this page restates the rules from the third edition’s final text as STEP Tools publishes it (document ISO TC 184/SC4/WG11 N304, dated 2015-07-30) in its own words. The earlier editions are covered where the text says how they differ.

File structure

An exchange structure is a sequential file of text. In order:

  1. ISO-10303-21;
  2. The header section, HEADER; … ENDSEC; (required)
  3. An optional anchor section, ANCHOR; … ENDSEC;
  4. An optional reference section, REFERENCE; … ENDSEC;
  5. Any number of data sections, DATA; or DATA(…); … ENDSEC;
  6. END-ISO-10303-21;
  7. Any number of signature sections, SIGNATURE; … ENDSEC;, after the closing token

The data section is optional in the third edition, so a file can consist only of forwarding references. A file may be compressed in a ZIP archive (see below).

ISO-10303-21;
HEADER;
FILE_DESCRIPTION(('example'),'2;1');
FILE_NAME('part.stp','2026-01-01T12:00:00',('author'),('org'),'preproc','system','');
FILE_SCHEMA(('CONFIG_CONTROL_DESIGN'));
ENDSEC;
DATA;
#1=CARTESIAN_POINT('',(0.,0.,0.));
#2=DIRECTION('',(0.,0.,1.));
#3=DIRECTION('',(1.,0.,0.));
#4=AXIS2_PLACEMENT_3D('',#1,#2,#3);
ENDSEC;
END-ISO-10303-21;
A minimal file. The entity names are those of the schema named in FILE_SCHEMA, here the AP203 schema read from the STEPcode repository.

Characters and token separators

  • The alphabet is the Unicode code points U+0020 to U+007E and U+0080 to U+10FFFF, written in UTF-8. Octets outside the encoded ranges, including line feeds, carriage returns, tabs and form feeds, are ignored when processing the file, so a line break may appear anywhere, even inside a token. (Editions 1 and 2 used only the ASCII range, with ISO 8859 characters reached through string directives.)
  • Only the space separates tokens, apart from comments and the print directives \N\ (newline) and \F\ (form feed). Any number of separators may appear wherever one may.
  • A comment starts with /* and ends with */. They do not nest.
  • The print directives may appear wherever a separator may, and inside strings and binaries (where they add nothing to the value), but not inside a URI, the anchor section or the reference section.

Tokens

TokenFormExamples and rules
KeywordCapital letters, digits and low line, starting with a letterCARTESIAN_POINT. A leading ! (at most once, first) makes a user-defined keyword, whose meaning is agreed between the partners (!MYCURVE)
IntegerOptional sign and digits, base 1016, +12, -349, 012 (12). Not 26 54, 32.0 or + 12
RealOptional sign, one or more digits, a full stop, zero or more digits, then optionally E with an optional sign and digits1.5, 2., 0.E25, -32.178E+02. Not .5, 1, 1E05, 3.E or 1,000.00. No precision is conveyed
StringApostrophe, characters, apostropheAn apostrophe inside is doubled, a reverse solidus is doubled, '' is the empty string. Directives are listed below. The stored string, including its apostrophes, is at most 32 769 octets
Entity instance name# and digits, not all zero#12; leading zeros are insignificant (#001 is #1). Forward and backward references are allowed
Value instance name@ and digits, not all zeroDefined only in the reference section. Its number may not equal an entity instance name’s number, so a URI fragment of digits is unambiguous
Constant entity name#, a capital letter, then capitals and digits#INCH: refers to an entity constant defined in the schema (the first schema if there are several). May be used on the right-hand side only
Constant value name@, a capital letter, then capitals and digits@PI: refers to a value constant in the schema. Right-hand side only
EnumerationFull stop, capital letter, capitals and digits, full stop.STEEL.. BOOLEAN uses .T. and .F.; LOGICAL adds .U. (unknown)
BinaryDouble quotes around hexadecimal digits, the first digit being 0 to 3See below
Anchor name< URI fragment >At least one character must not be a digit. A UUID is allowed
Resource< URI ><picture.jpg>, <http://example.com/a.stp#shape>
Tag nameLetter, then letters and digitsUsed in anchor tags
$Special tokenThe attribute has no value (null)
*Special tokenThe attribute is a redeclared derived attribute and its value follows from the schema’s rules

Punctuation is the semicolon, parentheses, comma and solidus. A list is ( … ) with parameters separated by commas, can be empty and nests to any depth. A typed parameter is a keyword followed by a parameter in parentheses; it appears where the schema uses a SELECT type, for example #1=STEEL_BAR(FLOATINGNUMBER(77.0),MEASURED_MASS(13.25));.

String control directives

Characters outside U+0020 to U+007E may be written directly in UTF-8 or, for compatibility with the older editions, with a directive. A string may contain no raw control characters, so U+0000 to U+001F and U+007F must use \X\.

DirectiveMeaning
\S\cThe character whose code is that of c plus 128, in the ISO 8859 part chosen by the last \P directive in the string (part 1 if none). '\S\D' is a capital A with diaeresis, (U+00C4, code 0x44 + 0x80)
\PL\Selects ISO 8859 part 1 to 9 for the following \S\ directives in this string only: A is part 1, B part 2, and so on up to I
\X2\ hhhh … \X0\Each group of four hexadecimal digits is one UCS code point (16-bit)
\X4\ hhhhhhhh … \X0\Each group of eight hexadecimal digits is one code point (32-bit; the first two digits are always zero)
\X\hhOne code point U+0000 to U+00FF. Required for U+0000 to U+001F and U+007F
\N\, \F\Print control only: newline and form feed, which are not part of the string’s value
 
'Don''t'              Don't
'\X2\03C0\X0\'         π (U+03C0)
'\X4\001F600\X0\'      U+1F600
'see \X\A7 4.1'        see § 4.1
'line one\X\0Aline two'  line one, a line feed, line two
String examples.

Binary encoding

A binary value of p bits is left-padded with n zero bits (0 to 3) to a multiple of four, and written as one hexadecimal digit giving n followed by the hexadecimal digits of the padded bits, in double quotes. So the empty value is "0", a single 0 bit is "30", a single 1 bit is "31", 111011 is "23B" and 100100101010 is "092A".

Data sections

  • With a single data section, DATA; is enough, and FILE_SCHEMA names exactly one schema, which governs it. With several, each is DATA('name',('SCHEMA'));: a unique section name, and a list of exactly one string naming a schema from FILE_SCHEMA.
  • Each entity instance is #n=KEYWORD(attributes);, with a unique name in the whole file. Order does not matter, and a name may be used before it is defined, even in another data section.
  • The attributes appear in the order of the EXPRESS declaration: for a simple instance the entity’s own explicit attributes; inherited and derived parts are covered below.
  • A user-defined instance uses a ! keyword and an agreed layout (#12=!MYCURVE(0.,0.,0.,1.,$,$,$);). The standard recommends a separate schema instead.

How EXPRESS types are written

EXPRESSWritten as
INTEGER, REAL, NUMBERAn integer, or a real for REAL and NUMBER; precision and string width do not change the encoding
STRING, BINARYA string, a binary
BOOLEAN, LOGICAL, ENUMERATIONAn enumeration: .T. and .F., with .U. for LOGICAL
LIST, ARRAY, SET, BAGA list. A multidimensional array is a list of lists with the last index varying fastest. Empty is (). An array element or optional attribute with no value is $
SELECTA typed parameter naming the chosen defined type, applied recursively for nested selects
Defined TYPEAs its underlying type
Entity-valued attributeThe instance name of the referenced instance (#2=XXX(#1,#3);)
OPTIONAL attribute with no value$, never an empty list
Derived, redeclared as DERIVE* at that position
INVERSE, FUNCTION, PROCEDURE, RULE, WHERE rules, REMARKS, SCHEMANot written. A reference to an EXPRESS CONSTANT is a constant name

Entity type names are written upper case, using the schema’s published short names if it defines them.

Subtypes and complex instances

An instance that belongs to a subtype is complex. One of two mappings applies, decided per instance. Find the entity types in the instance’s evaluated set that have no subtype in that set (the leaf types).

  • Internal mapping, when there is exactly one leaf type: an ordinary record whose keyword is the leaf type and whose parameters are the inherited attributes first (the supertypes in the order of the SUBTYPE OF clause, the highest supertype first, repeated supertypes only once), then the leaf’s own.
  • External mapping, when there are several leaf types: a parenthesised sequence of simple records, one per entity type of the instance, in ascending order of type name (compared as upper-case letters and digits by octet value), each carrying only that type’s own explicit attributes. Example: #5=(LENGTH_UNIT()NAMED_UNIT(*)SI_UNIT(.MILLI.,.METRE.));.

Anchor, reference and signature sections

  • Anchor section. Each entry gives an external name to an instance or value: <name> = item {tag:item};. The item may be $, an integer, real, string, enumeration or binary, an instance name, a resource, or a (nested) list of these. Names are unique in the file and may be UUIDs. A tag, {preparation:<WELD_DC.XML>}, attaches extra information that is outside the information model. The section comes after the header and before any reference, data or signature section.
  • Reference section. Entries are #10 = <uri>; or @20 = <uri>;, binding an instance or value name to something in another file. The resource may be a whole file, a fragment that is an anchor (including a UUID), a fragment in the same file, a file in another format, or a fragment of digits, which means an instance name from an older-edition file. Each name appears once and is not also defined in a data section.
  • Signature section. SIGNATURE;, Base64 text and ENDSEC;. The content is a Cryptographic Message Syntax (RFC 5652) signature with external content, covering everything before the SIGNATURE; token including earlier signatures; only characters of the basic alphabet go into the digest.
ANCHOR;
<82ff3c50-3610-11e5-a2cb-0800200c9a66> = #10;
<bb2ac46e-3610-11e5-a151-feff819cdc9f> = (1.1, 2.1, 3.1);
<1231ea63-d573-4d50-81f3-8bb0e9c1cbf5> = #10 {ratio:196.73};
ENDSEC;
REFERENCE;
#10 = <http://example.com/product.stp#shape>;
@70 = <http://example.com/product.stp#value>;
ENDSEC;
Anchor and reference entries.

ZIP archives and directories

A file may be stored in a ZIP archive written with PKZIP 2.04g compression (no encryption, Unicode file names or Deflate64). The archive’s root, which must exist for it to conform, is a file named ISO-10303.p21. Other files are subsidiaries: further exchange structures, further archives or ancillary data such as pictures. Relative addresses resolve inside the archive and may not point outside it, and only the root can be referenced from outside, so subsidiaries are exposed through anchors in the root. A directory containing ISO-10303.p21 is treated the same way.

Not on this page

  • Annexes of the standard on tape and diskette transport, the PICS proforma, multiple-schema population methods (annex E) and the ECMAScript binding for the anchor section (annex F).
  • The grammar in Wirth syntax notation, which is the standard’s own table; the tokens table above is its content.
  • Part 28 XML encoding and Part 26 binary encoding of ISO 10303, which are other encodings of the same schemas.

Sources