.TH DATADEC L
.SH NAME
datadec \- ANSI C data declaration module constructor
.SH SYNOPSIS
datadec
.RB [\- vfno ]
.RB [\- s FUNCTIONNAME...]
.I basename
.RB infile
.PP
OR
.PP
datadec
.RB [\- m ]
.RB infile

.SH DESCRIPTION
.B Datadec
takes an input file containing a series of Haskell style recursive (or
inductive) datatype declarations - with optional hints on printing and freeing,
and builds an definition/implementation pair of ANSI C files \-
.I "basename.h"
and
.I "basename.c"
containing data declarations,
constructor functions, deconstructor functions and printing functions.

.PP
The two files produced together form a module that implements the relevent
data types.

.SH "OPTIONS"
.TP 8
.B "\-v"
enter verbose mode.
.I "Datadec"
now displays the data types that it parses, along with various almost
certainly useless bits of information about optimization.
.TP
.B "\-f"
Enable experimental generation of free_TYPE() functions.
See the later section on free functions.
.TP
.B "\-n"
do not perform various optimizations.
.TP
.B "\-o"
perform optimizations (the default).
.TP
.B "\-m"
Do **NOT** produce the normal module.  Instead, produce
.I "meta-data"
on stdout.  This lists, for each type and shape, the typename,
the shapename, and a comma-separated list of the shape parameter
types.
.TP
.B "\-s FUNCTIONNAME"
Add
.I FUNCTIONNAME
to the set of functions to be suppressed, i.e. not generated in the
output .c file.
Currently this is only implemented for print_TYPE functions,
but it would be trivial to implement for constructors, deconstructors
and free functions too.
The usual reason for using this suppression feature is because you
have provided a manually constructed optimized print function in the
GLOBAL section of the datadec input file.

.SH "AN EXAMPLE"
.PP
The simplest use is to prepare an input file, such as
.I "data.in,"
which might (for example) contain:
.nf
TYPE {
        intlist =  nil or cons( int first, intlist next );
        illist  =  nil or cons( intlist first, illist next );
        idtree  =  leaf( string id )
                or node( idtree left, idtree right );
}
.fi
To generate C code implementing these types, invoke:
.nf
     datadec eek data.in
.fi
which generates
.I "eek.h"
and
.I "eek.c"

.fi
If you wanted to suppress print_idtree and print_illist from eek.c
(while leaving their prototypes in eek.h), you would run:
.nf
     datadec -s print_idtree -s print_illist eek data.in
.fi

.SH THE DATA DECLARATION LANGUAGE

The language accepted by datadec is split into two components:
the "outer language" is patterned after
the GMD compiler tools
.B "LALR"
and
.B "REX"
(similar to Yacc and Lex)
and allows you to specify four sections (only the last is compulsory):

.PP
.nf
.B "[ EXPORT { free_format_text } ]"
.br
.B "[ GLOBAL { free_format_text } ]"
.br
.B "[ BEGIN { free_format_text } ]"
.br
.B "TYPE { types }"
.fi

.PP
The contents of the
.I "export"
section are placed in the header file (the .h).
Commonly, you may wish to add extern function declarations, public types and
external variable declarations
which must be
placed at the top of the header file, and also define some additional
procedures using the automatically generated types which must be placed after
the type declarations!
To achieve this, you should place a `@@' in the export section -- the text up
to that point is placed at the top of the header file, whereas the text
after it is placed at the bottom of the header file -- after all the types
have been defined.

.PP
Similarly, the contents of the 
.I "global"
section are placed in the C file,
again with `@@' being used to split the global section into "top of file" and
"bottom of file" pieces.
Note that the "top of file" piece is placed ABOVE the '#include "thismodule.h"
and thus CANNOT use the inductive data types themselves, however this makes it
the perfect place to add #include's to allow the data types to use other
types as fields in the shapes.  For example, should one of your inductive
data types use a "set h" parameter in a shape, #include "set.h" might need
to go into GLOBAL above the @@.

.PP
Similarly, the contents of the 
.I "begin"
section are placed in an initialization procedure, which the user of the
constructed module must remember to call at an appropriate juncture (eg.
immediately when main starts).

.PP
The
.I "types"
section contains the type declarations themselves - the inner language.

.PP
The "inner language" - that of specifying the actual types section -
is closely modelled on Haskell, Miranda or Hope, with printing rules added.
Here is the grammar:

.PP
.nf
types	= list(type)
.br
type 	= type_name '=' shape list( 'or' shape ) ';'
.br
shape	= constructor_name [ '(' params ') ] [ print ]
.br
params	= param list( ',' param)
.br
param	= ['-'] type_name param_name
.br
print	= list(element)
.br
element	= number | string_literal
.fi

.PP
Note that each data type is terminated by a semicolon,
and that (within one data type) each shape is separated from the next by 'or'
(just like the '|' in Haskell).
If a particular shape has parameters, they are separated from each other
by commas.
Each type name is simply an identifier.

.PP
.I "Datadec"
also generates routines to write each type to an open FILE *.
The method of printing each shape is governed by the presence or absence
of a print rule.  If no print rule is given, the constructor name is printed,
then an open bracket,
then each parameter is written out using the appropriate print routine,
with commas separating the parameters, then a close bracket.

.PP
If a print rule is given, each print element
(these are syntactically separated by whitespace)
is used to generate the write routine as follows:
A literal string will simply be printed
(well, '\\n' is turned into a newline!),
whereas a number (eg. 4) means that the
4th parameter is printed (invoking the print function for that routine).

.PP
For example, we could augment the
.I "idtree"
type from the example given above with print rules:

.nf
TYPE {
idtree  =  leaf( string id )                    "leaf(" 1 ")"
        or node( idtree left, idtree right )    "node( l=" 1 ", r=" 2 " )";
}
.fi

.PP
Now, an idtree constructed as
.nf
node( leaf( "hello" ), node( leaf( "there" ), leaf( "how are you" ) ) )
.fi
would print as:
.nf
node( l=leaf("hello"), r=node(l=leaf("there"),r=leaf("how are you")))
.fi

.SH SEE ALSO
.nf
LALR, REX, Haskell Language Definition.
.fi

.SH FREE FUNCTIONS (EXPERIMENTAL)

May 2014: After saying for 20 years that
"I must implement the missing free_TYPE functions sometime",
I have now experimentally added support for this. Currently
you have to enable the generation of these via '-f',
.I "Datadec"
will then generate a series of free_TYPE( TYPE t ) functions,
essentially these will perform a post-order tree traversal,
recursively freeing every non-basic parameter in each shape.
(Basic types, not needing freeing, are things like int, long,
BOOL etc).

.PP
However, in C you often share pointers (most obviously to readonly
string literals, but whole data structures can be shared),
so the user may or may not want to free these.  To handle this,
the above grammar has been extended to allow optional '-' hints
(meaning "never free this") on parameters within shapes.

.PP
So, for instance, given the type:

.nf
TYPE {
strlist  =  nil
        or cons( string h, strlist t )
}
.fi

the corresponding free_strlist() function will attempt to
.nf
free_string( head_of_list );
.fi

.PP
(assuming that in the GLOBAL section you will then provide
a function or macro to implement free_string(), as in:

.nf
#define free_string(s) free(s)
.fi

.PP
But if you prepend a '-' to the "string h" part of the type, i.e.:

.nf
TYPE {
strlist  =  nil
        or cons( -string h, strlist t )
}
.fi

.PP
then no call to free_string( head_of_list ) will be made.

.PP
However, even with this mechanism, it is still incredibly easy to
share pointers when building data structures, such that you
.I sometimes
want to free a parameter, and sometimes don't.
I am still thinking about this.
For now, I recommend that you compile all datadec generated
modules (with free functions in) against a memory checking
system like memlib, or use a tool like valgrind's memcheck
to validate your memory allocation and deallocation.

.SH MISSING FEATURES
.PP
A cool extra would be a sprint_TYPE function to print into a string.
This should be a trivial modification on the code that prints to a file,
of course we don't know how long the generated string will need to be.

.SH BUGS
Some single letter typenames (eg. "f" or "p") could clash with internal
parameter names in the print routines, leading to syntax errors when you
compile the files generated by datadec.
.PP
And, finally, one day I'll have to write the Perl, C++ and (perhaps) Java versions :-)

.SH "AUTHOR"
Duncan C. White, D.White@imperial.ac.uk.