ack/util/mcgg/iburg.1
2016-09-20 20:37:16 +02:00

286 lines
7 KiB
Groff

.TH IBURG 1 "local \- 1/26/93"
.\" $Id$
.SH NAME
iburg \- code generator generator
.SH SYNOPSIS
.B iburg
[
.I option
]...
[ [
.I input
]
.I output
]
.br
.SH DESCRIPTION
.PP
.I iburg
reads BURG specification from
.I input
and writes a pattern-matching code generator to
.IR output .
If
.I input
is `\-' or is omitted,
.I iburg
reads the standard input;
If
.I output
is `\-' or is omitted,
.I iburg
writes to the standard output.
.PP
.I iburg
accepts BURG specifications that conform to the following EBNF grammar.
Terminals are enclosed in single quotes, all other symbols are nonterminals,
{X} denotes zero or more instances of X, and [X] denotes an optional X.
.PP
.nf
.RS
.ft CW
spec: { dcl } `%%' { rule } [ `%%' ]
dcl: `%start' nonterm
`%term' { identifier `=' integer }
rule: nonterm `:' tree `=' integer [ cost ] `;'
cost: `(' integer ')'
tree: term `(' tree `,' tree `)'
term `(' tree `)'
term
nonterm
.RE
.fi
.PP
Specifications are structurally similar to
.IR yacc 's.
Text between
`\f(CW%{\fP'
and
`\f(CW%}\fP'
is called the configuration section; there may be several such segments.
All are concatenated and copied verbatim into the head of the generated
parser, which is called burm.
Text after the second
`\f(CW%%\fP',
if any, is also copied verbatim into
.IR burm ,
at the end.
.PP
Specifications consist of declarations, a
`\f(CW%%\fP'
separator, and rules.
Input is line-oriented; each declaration and rule must appear on a separate line,
and declarations must begin in column 1.
Declarations declare terminals \(em the operators in subject
trees \(em and associate a unique, positive external symbol
number with each one.
Non-terminals are declared by their presence
on the left side of rules. The
\f(CW%start\fP
declaration optionally declares a non-terminal as the start symbol.
In the grammar above,
\f(CWterm\fP
and
\f(CWnonterm\fP
denote identifiers that are terminals and non-terminals, respectively.
.PP
Rules define tree patterns in a fully parenthesized prefix
form. Every non-terminal denotes a tree.
Each operator has a fixed
arity, which is inferred from the rules in which it is used.
A chain rule is a rule whose pattern is another non-terminal.
If no start symbol is declared, the non-terminal defined by the first rule is used.
.PP
Each rule has a unique, positive external rule number, which
comes after the pattern and is preceded by a
`\f(CW=\fP'.
External rule numbers are used to report the
matching rule to a user-supplied semantic action routine.
Rules end with an optional non-negative, integer cost; omitted costs
default to zero.
.PP
The display below shows a fragment of a BURG specification.
This example uses upper-case for terminals and lower-case for non-terminals.
.PP
.nf
.ft CW
%{
enum { ADDI=309, ADDRLP=295, ASGNI=53,
CNSTI=21, CVCI=85, I0I=661, INDIRC=67 };
typedef struct tree {
int op;
struct tree *kids[2];
int val;
struct { int state; } x;
} *NODEPTR_TYPE, *Tree;
#define LEFT_CHILD(p) ((p)->kids[0])
#define RIGHT_CHILD(p) ((p)->kids[1])
#define PANIC printf
#define STATE_LABEL(p) ((p)->x.state)
int OP_LABEL(NODEPTR_TYPE p) {
switch (p->op) {
case CNSTI: if (p->val == 0) return 661 /* I0I */;
default: return p->op;
}
}
%}
%term ADDI=309 ADDRLP=295 ASGNI=53
%term CNSTI=21 CVCI=85 I0I=661 INDIRC=67
%%
stmt: ASGNI(disp,reg) = 4 (1);
stmt: reg = 5;
reg: ADDI(reg,rc) = 6 (1);
reg: CVCI(INDIRC(disp)) = 7 (1);
reg: I0I = 8;
reg: disp = 9 (1);
disp: ADDI(reg,con) = 10;
disp: ADDRLP = 11;
rc: con = 12;
rc: reg = 13;
con: CNSTI = 14;
con: I0I = 15;
%%
.fi
.PP
The configuration section configures
\f(CWburm\fP
for the trees being parsed and the client's environment.
As shown, this section must define
\f(CWNODEPTR_TYPE\fP
to be a visible typedef symbol for a pointer to a
node in the subject tree.
\f(CWburm\fP
invokes
\f(CWOP_LABEL(p)\fP,
\f(CWLEFT\_CHILD(p)\fP, and
\f(CWRIGHT\_CHILD(p)\fP
to read the operator and children from the node pointed to by \f(CWp\fP.
It invokes
\f(CWPANIC\fP
when it detects an error.
If the configuration section defines these operations as macros, they are implemented in-line;
otherwise, they must be implemented as functions.
.PP
By default,
\f(CWburm\fP
computes and stores a single integral state in each node of the subject tree.
The configuration section must define a macro
\f(CWSTATE_LABEL(p)\fP
to access the state field of the node pointed to
by \f(CWp\fP. It must be large enough to hold a pointer, and
a macro is required because it is used as an lvalue.
The configuration section may define the macro
\f(CWSTATE_TYPE\fP
to specify a different type for state values; if
\f(CWSTATE_TYPE\fP
is not defined, int is used.
.PP
The configuration section may also define
\f(CWALLOC(n)\fP
for allocating
\f(CWn\fP
bytes.
If
\f(CWALLOC\fP
is not defined,
.IR malloc (3)
is used.
.SH OPTIONS
.TP
.BI \-p \ prefix
.br
.ns
.TP
.BI \-p prefix
Use
.I prefix
as the disambiquating prefix for exported names and visible fields.
The default is `\f(CWburm\fP'.
.TP
.BI \-maxcost= ddd
Use the integral value
.I ddd
as the maximum cost.
The default is \f(CWSHRT_MAX\fR
when ints are bigger than shorts, and
\f(CWSHRT_MAX\fP/2 when ints and shorts are the same size.
.TP
.B \-I
Emit code for the following data values and functions.
.sp
.nf
.ft CW
char burm_arity[];
char *burm_opname[];
char *burm_ntname[];
char *burm_string[];
short burm_cost[][4];
int burm_op_label(NODEPTR_TYPE p);
NODEPTR_TYPE burm_child(NODEPTR_TYPE p, int index);
STATE_TYPE burm_state_label(NODEPTR_TYPE p);
.sp
.fi
.ft R
\f(CWburm_arity\fP and
\f(CWburm_opname\fP
are indexed by external symbol numbers and gives their arities.
\f(CWburm_ntname\fP
is indexed by a non-terminal number and gives its string name.
\f(CWburm_string\fP
and
\f(CWburm_cost\fP
are indexed by an external rule number and give the string
representation and cost of each rule.
The functions encapsulate the similarly named macros.
.TP
.B \-T
Arrange for
.sp
.nf
.ft CW
void burm_trace(NODEPTR_TYPE p, int eruleno,
int cost, int bestcost);
.sp
.fi
.ft R
to be called at each successful match.
\f(CWp\fP
identifies the node and
\f(CWeruleno\fP
identifies the matching rule;
\f(CWeruleno\fP
is an index into \f(CWburm_string\fP.
\f(CWcost\fP
is the cost of the match and
\f(CWbestcost\fP
is the cost of the best previous match. The current match
wins only if
\f(CWcost\fP
is less than \f(CWbestcost\fP.
SHRT_MAX represents the infinite cost of no previous match.
\f(CWburm_trace\fP must be declared in the configuration section.
.SH "SEE ALSO"
C. W. Fraser, R. R. Henry and T. A. Proebsting,
`BURG \(em Fast optimal instruction selection and tree parsing,'
.I
SIGPLAN Notices
.BR 27 ,
4 (Apr. 1992), 68-76.
.PP
C. W. Fraser, D. R. Hanson and T. A. Proebsting,
`Engineering a simple, efficient code generator generator,'
.I
ACM Letters on Programming Languages and Systems
.BR 1 ,
3 (Sep. 1992), 213-226.
.br
.SH BUGS
Mail bug reports along with the shortest input
that exposes them to drh@drhanson.net.