freebsd-skq/contrib/groff/refer/label.y
1996-09-07 16:18:32 +00:00

1178 lines
28 KiB
Plaintext

/* -*- C++ -*-
Copyright (C) 1989, 1990, 1991, 1992 Free Software Foundation, Inc.
Written by James Clark (jjc@jclark.com)
This file is part of groff.
groff is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 2, or (at your option) any later
version.
groff is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received a copy of the GNU General Public License along
with groff; see the file COPYING. If not, write to the Free Software
Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
%{
#include "refer.h"
#include "refid.h"
#include "ref.h"
#include "token.h"
int yylex();
void yyerror(const char *);
int yyparse();
static const char *format_serial(char c, int n);
struct label_info {
int start;
int length;
int count;
int total;
label_info(const string &);
};
label_info *lookup_label(const string &label);
struct expression {
enum {
// Does the tentative label depend on the reference?
CONTAINS_VARIABLE = 01,
CONTAINS_STAR = 02,
CONTAINS_FORMAT = 04,
CONTAINS_AT = 010
};
virtual ~expression() { }
virtual void evaluate(int, const reference &, string &,
substring_position &) = 0;
virtual unsigned analyze() { return 0; }
};
class at_expr : public expression {
public:
at_expr() { }
void evaluate(int, const reference &, string &, substring_position &);
unsigned analyze() { return CONTAINS_VARIABLE|CONTAINS_AT; }
};
class format_expr : public expression {
char type;
int width;
int first_number;
public:
format_expr(char c, int w = 0, int f = 1)
: type(c), width(w), first_number(f) { }
void evaluate(int, const reference &, string &, substring_position &);
unsigned analyze() { return CONTAINS_FORMAT; }
};
class field_expr : public expression {
int number;
char name;
public:
field_expr(char nm, int num) : name(nm), number(num) { }
void evaluate(int, const reference &, string &, substring_position &);
unsigned analyze() { return CONTAINS_VARIABLE; }
};
class literal_expr : public expression {
string s;
public:
literal_expr(const char *ptr, int len) : s(ptr, len) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class unary_expr : public expression {
protected:
expression *expr;
public:
unary_expr(expression *e) : expr(e) { }
~unary_expr() { delete expr; }
void evaluate(int, const reference &, string &, substring_position &) = 0;
unsigned analyze() { return expr ? expr->analyze() : 0; }
};
// This caches the analysis of an expression.
class analyzed_expr : public unary_expr {
unsigned flags;
public:
analyzed_expr(expression *);
void evaluate(int, const reference &, string &, substring_position &);
unsigned analyze() { return flags; }
};
class star_expr : public unary_expr {
public:
star_expr(expression *e) : unary_expr(e) { }
void evaluate(int, const reference &, string &, substring_position &);
unsigned analyze() {
return ((expr ? (expr->analyze() & ~CONTAINS_VARIABLE) : 0)
| CONTAINS_STAR);
}
};
typedef void map_func(const char *, const char *, string &);
class map_expr : public unary_expr {
map_func *func;
public:
map_expr(expression *e, map_func *f) : unary_expr(e), func(f) { }
void evaluate(int, const reference &, string &, substring_position &);
};
typedef const char *extractor_func(const char *, const char *, const char **);
class extractor_expr : public unary_expr {
int part;
extractor_func *func;
public:
enum { BEFORE = +1, MATCH = 0, AFTER = -1 };
extractor_expr(expression *e, extractor_func *f, int pt)
: unary_expr(e), func(f), part(pt) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class truncate_expr : public unary_expr {
int n;
public:
truncate_expr(expression *e, int i) : n(i), unary_expr(e) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class separator_expr : public unary_expr {
public:
separator_expr(expression *e) : unary_expr(e) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class binary_expr : public expression {
protected:
expression *expr1;
expression *expr2;
public:
binary_expr(expression *e1, expression *e2) : expr1(e1), expr2(e2) { }
~binary_expr() { delete expr1; delete expr2; }
void evaluate(int, const reference &, string &, substring_position &) = 0;
unsigned analyze() {
return (expr1 ? expr1->analyze() : 0) | (expr2 ? expr2->analyze() : 0);
}
};
class alternative_expr : public binary_expr {
public:
alternative_expr(expression *e1, expression *e2) : binary_expr(e1, e2) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class list_expr : public binary_expr {
public:
list_expr(expression *e1, expression *e2) : binary_expr(e1, e2) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class substitute_expr : public binary_expr {
public:
substitute_expr(expression *e1, expression *e2) : binary_expr(e1, e2) { }
void evaluate(int, const reference &, string &, substring_position &);
};
class ternary_expr : public expression {
protected:
expression *expr1;
expression *expr2;
expression *expr3;
public:
ternary_expr(expression *e1, expression *e2, expression *e3)
: expr1(e1), expr2(e2), expr3(e3) { }
~ternary_expr() { delete expr1; delete expr2; delete expr3; }
void evaluate(int, const reference &, string &, substring_position &) = 0;
unsigned analyze() {
return ((expr1 ? expr1->analyze() : 0)
| (expr2 ? expr2->analyze() : 0)
| (expr3 ? expr3->analyze() : 0));
}
};
class conditional_expr : public ternary_expr {
public:
conditional_expr(expression *e1, expression *e2, expression *e3)
: ternary_expr(e1, e2, e3) { }
void evaluate(int, const reference &, string &, substring_position &);
};
static expression *parsed_label = 0;
static expression *parsed_date_label = 0;
static expression *parsed_short_label = 0;
static expression *parse_result;
string literals;
%}
%union {
int num;
expression *expr;
struct { int ndigits; int val; } dig;
struct { int start; int len; } str;
}
/* uppercase or lowercase letter */
%token <num> TOKEN_LETTER
/* literal characters */
%token <str> TOKEN_LITERAL
/* digit */
%token <num> TOKEN_DIGIT
%type <expr> conditional
%type <expr> alternative
%type <expr> list
%type <expr> string
%type <expr> substitute
%type <expr> optional_conditional
%type <num> number
%type <dig> digits
%type <num> optional_number
%type <num> flag
%%
expr:
optional_conditional
{ parse_result = ($1 ? new analyzed_expr($1) : 0); }
;
conditional:
alternative
{ $$ = $1; }
| alternative '?' optional_conditional ':' conditional
{ $$ = new conditional_expr($1, $3, $5); }
;
optional_conditional:
/* empty */
{ $$ = 0; }
| conditional
{ $$ = $1; }
;
alternative:
list
{ $$ = $1; }
| alternative '|' list
{ $$ = new alternative_expr($1, $3); }
| alternative '&' list
{ $$ = new conditional_expr($1, $3, 0); }
;
list:
substitute
{ $$ = $1; }
| list substitute
{ $$ = new list_expr($1, $2); }
;
substitute:
string
{ $$ = $1; }
| substitute '~' string
{ $$ = new substitute_expr($1, $3); }
;
string:
'@'
{ $$ = new at_expr; }
| TOKEN_LITERAL
{
$$ = new literal_expr(literals.contents() + $1.start,
$1.len);
}
| TOKEN_LETTER
{ $$ = new field_expr($1, 0); }
| TOKEN_LETTER number
{ $$ = new field_expr($1, $2 - 1); }
| '%' TOKEN_LETTER
{
switch ($2) {
case 'I':
case 'i':
case 'A':
case 'a':
$$ = new format_expr($2);
break;
default:
command_error("unrecognized format `%1'", char($2));
$$ = new format_expr('a');
break;
}
}
| '%' digits
{
$$ = new format_expr('0', $2.ndigits, $2.val);
}
| string '.' flag TOKEN_LETTER optional_number
{
switch ($4) {
case 'l':
$$ = new map_expr($1, lowercase);
break;
case 'u':
$$ = new map_expr($1, uppercase);
break;
case 'c':
$$ = new map_expr($1, capitalize);
break;
case 'r':
$$ = new map_expr($1, reverse_name);
break;
case 'a':
$$ = new map_expr($1, abbreviate_name);
break;
case 'y':
$$ = new extractor_expr($1, find_year, $3);
break;
case 'n':
$$ = new extractor_expr($1, find_last_name, $3);
break;
default:
$$ = $1;
command_error("unknown function `%1'", char($4));
break;
}
}
| string '+' number
{ $$ = new truncate_expr($1, $3); }
| string '-' number
{ $$ = new truncate_expr($1, -$3); }
| string '*'
{ $$ = new star_expr($1); }
| '(' optional_conditional ')'
{ $$ = $2; }
| '<' optional_conditional '>'
{ $$ = new separator_expr($2); }
;
optional_number:
/* empty */
{ $$ = -1; }
| number
{ $$ = $1; }
;
number:
TOKEN_DIGIT
{ $$ = $1; }
| number TOKEN_DIGIT
{ $$ = $1*10 + $2; }
;
digits:
TOKEN_DIGIT
{ $$.ndigits = 1; $$.val = $1; }
| digits TOKEN_DIGIT
{ $$.ndigits = $1.ndigits + 1; $$.val = $1.val*10 + $2; }
;
flag:
/* empty */
{ $$ = 0; }
| '+'
{ $$ = 1; }
| '-'
{ $$ = -1; }
;
%%
/* bison defines const to be empty unless __STDC__ is defined, which it
isn't under cfront */
#ifdef const
#undef const
#endif
const char *spec_ptr;
const char *spec_end;
const char *spec_cur;
int yylex()
{
while (spec_ptr < spec_end && csspace(*spec_ptr))
spec_ptr++;
spec_cur = spec_ptr;
if (spec_ptr >= spec_end)
return 0;
unsigned char c = *spec_ptr++;
if (csalpha(c)) {
yylval.num = c;
return TOKEN_LETTER;
}
if (csdigit(c)) {
yylval.num = c - '0';
return TOKEN_DIGIT;
}
if (c == '\'') {
yylval.str.start = literals.length();
for (; spec_ptr < spec_end; spec_ptr++) {
if (*spec_ptr == '\'') {
if (++spec_ptr < spec_end && *spec_ptr == '\'')
literals += '\'';
else {
yylval.str.len = literals.length() - yylval.str.start;
return TOKEN_LITERAL;
}
}
else
literals += *spec_ptr;
}
yylval.str.len = literals.length() - yylval.str.start;
return TOKEN_LITERAL;
}
return c;
}
int set_label_spec(const char *label_spec)
{
spec_cur = spec_ptr = label_spec;
spec_end = strchr(label_spec, '\0');
literals.clear();
if (yyparse())
return 0;
delete parsed_label;
parsed_label = parse_result;
return 1;
}
int set_date_label_spec(const char *label_spec)
{
spec_cur = spec_ptr = label_spec;
spec_end = strchr(label_spec, '\0');
literals.clear();
if (yyparse())
return 0;
delete parsed_date_label;
parsed_date_label = parse_result;
return 1;
}
int set_short_label_spec(const char *label_spec)
{
spec_cur = spec_ptr = label_spec;
spec_end = strchr(label_spec, '\0');
literals.clear();
if (yyparse())
return 0;
delete parsed_short_label;
parsed_short_label = parse_result;
return 1;
}
void yyerror(const char *message)
{
if (spec_cur < spec_end)
command_error("label specification %1 before `%2'", message, spec_cur);
else
command_error("label specification %1 at end of string",
message, spec_cur);
}
void at_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &)
{
if (tentative)
ref.canonicalize_authors(result);
else {
const char *end, *start = ref.get_authors(&end);
if (start)
result.append(start, end - start);
}
}
void format_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &)
{
if (tentative)
return;
const label_info *lp = ref.get_label_ptr();
int num = lp == 0 ? ref.get_number() : lp->count;
if (type != '0')
result += format_serial(type, num + 1);
else {
const char *ptr = itoa(num + first_number);
int pad = width - strlen(ptr);
while (--pad >= 0)
result += '0';
result += ptr;
}
}
static const char *format_serial(char c, int n)
{
assert(n > 0);
static char buf[128]; // more than enough.
switch (c) {
case 'i':
case 'I':
{
char *p = buf;
// troff uses z and w to represent 10000 and 5000 in Roman
// numerals; I can find no historical basis for this usage
const char *s = c == 'i' ? "zwmdclxvi" : "ZWMDCLXVI";
if (n >= 40000)
return itoa(n);
while (n >= 10000) {
*p++ = s[0];
n -= 10000;
}
for (int i = 1000; i > 0; i /= 10, s += 2) {
int m = n/i;
n -= m*i;
switch (m) {
case 3:
*p++ = s[2];
/* falls through */
case 2:
*p++ = s[2];
/* falls through */
case 1:
*p++ = s[2];
break;
case 4:
*p++ = s[2];
*p++ = s[1];
break;
case 8:
*p++ = s[1];
*p++ = s[2];
*p++ = s[2];
*p++ = s[2];
break;
case 7:
*p++ = s[1];
*p++ = s[2];
*p++ = s[2];
break;
case 6:
*p++ = s[1];
*p++ = s[2];
break;
case 5:
*p++ = s[1];
break;
case 9:
*p++ = s[2];
*p++ = s[0];
}
}
*p = 0;
break;
}
case 'a':
case 'A':
{
char *p = buf;
// this is derived from troff/reg.c
while (n > 0) {
int d = n % 26;
if (d == 0)
d = 26;
n -= d;
n /= 26;
*p++ = c + d - 1; // ASCII dependent
}
*p-- = 0;
// Reverse it.
char *q = buf;
while (q < p) {
char temp = *q;
*q = *p;
*p = temp;
--p;
++q;
}
break;
}
default:
assert(0);
}
return buf;
}
void field_expr::evaluate(int, const reference &ref,
string &result, substring_position &)
{
const char *end;
const char *start = ref.get_field(name, &end);
if (start) {
start = nth_field(number, start, &end);
if (start)
result.append(start, end - start);
}
}
void literal_expr::evaluate(int, const reference &,
string &result, substring_position &)
{
result += s;
}
analyzed_expr::analyzed_expr(expression *e)
: unary_expr(e), flags(e ? e->analyze() : 0)
{
}
void analyzed_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
if (expr)
expr->evaluate(tentative, ref, result, pos);
}
void star_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
const label_info *lp = ref.get_label_ptr();
if (!tentative
&& (lp == 0 || lp->total > 1)
&& expr)
expr->evaluate(tentative, ref, result, pos);
}
void separator_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
int start_length = result.length();
int is_first = pos.start < 0;
if (expr)
expr->evaluate(tentative, ref, result, pos);
if (is_first) {
pos.start = start_length;
pos.length = result.length() - start_length;
}
}
void map_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &)
{
if (expr) {
string temp;
substring_position temp_pos;
expr->evaluate(tentative, ref, temp, temp_pos);
(*func)(temp.contents(), temp.contents() + temp.length(), result);
}
}
void extractor_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &)
{
if (expr) {
string temp;
substring_position temp_pos;
expr->evaluate(tentative, ref, temp, temp_pos);
const char *end, *start = (*func)(temp.contents(),
temp.contents() + temp.length(),
&end);
switch (part) {
case BEFORE:
if (start)
result.append(temp.contents(), start - temp.contents());
else
result += temp;
break;
case MATCH:
if (start)
result.append(start, end - start);
break;
case AFTER:
if (start)
result.append(end, temp.contents() + temp.length() - end);
break;
default:
assert(0);
}
}
}
static void first_part(int len, const char *ptr, const char *end,
string &result)
{
for (;;) {
const char *token_start = ptr;
if (!get_token(&ptr, end))
break;
const token_info *ti = lookup_token(token_start, ptr);
int counts = ti->sortify_non_empty(token_start, ptr);
if (counts && --len < 0)
break;
if (counts || ti->is_accent())
result.append(token_start, ptr - token_start);
}
}
static void last_part(int len, const char *ptr, const char *end,
string &result)
{
const char *start = ptr;
int count = 0;
for (;;) {
const char *token_start = ptr;
if (!get_token(&ptr, end))
break;
const token_info *ti = lookup_token(token_start, ptr);
if (ti->sortify_non_empty(token_start, ptr))
count++;
}
ptr = start;
int skip = count - len;
if (skip > 0) {
for (;;) {
const char *token_start = ptr;
if (!get_token(&ptr, end))
assert(0);
const token_info *ti = lookup_token(token_start, ptr);
if (ti->sortify_non_empty(token_start, ptr) && --skip < 0) {
ptr = token_start;
break;
}
}
}
first_part(len, ptr, end, result);
}
void truncate_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &)
{
if (expr) {
string temp;
substring_position temp_pos;
expr->evaluate(tentative, ref, temp, temp_pos);
const char *start = temp.contents();
const char *end = start + temp.length();
if (n > 0)
first_part(n, start, end, result);
else if (n < 0)
last_part(-n, start, end, result);
}
}
void alternative_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
int start_length = result.length();
if (expr1)
expr1->evaluate(tentative, ref, result, pos);
if (result.length() == start_length && expr2)
expr2->evaluate(tentative, ref, result, pos);
}
void list_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
if (expr1)
expr1->evaluate(tentative, ref, result, pos);
if (expr2)
expr2->evaluate(tentative, ref, result, pos);
}
void substitute_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
int start_length = result.length();
if (expr1)
expr1->evaluate(tentative, ref, result, pos);
if (result.length() > start_length && result[result.length() - 1] == '-') {
// ought to see if pos covers the -
result.set_length(result.length() - 1);
if (expr2)
expr2->evaluate(tentative, ref, result, pos);
}
}
void conditional_expr::evaluate(int tentative, const reference &ref,
string &result, substring_position &pos)
{
string temp;
substring_position temp_pos;
if (expr1)
expr1->evaluate(tentative, ref, temp, temp_pos);
if (temp.length() > 0) {
if (expr2)
expr2->evaluate(tentative, ref, result, pos);
}
else {
if (expr3)
expr3->evaluate(tentative, ref, result, pos);
}
}
void reference::pre_compute_label()
{
if (parsed_label != 0
&& (parsed_label->analyze() & expression::CONTAINS_VARIABLE)) {
label.clear();
substring_position temp_pos;
parsed_label->evaluate(1, *this, label, temp_pos);
label_ptr = lookup_label(label);
}
}
void reference::compute_label()
{
label.clear();
if (parsed_label)
parsed_label->evaluate(0, *this, label, separator_pos);
if (short_label_flag && parsed_short_label)
parsed_short_label->evaluate(0, *this, short_label, short_separator_pos);
if (date_as_label) {
string new_date;
if (parsed_date_label) {
substring_position temp_pos;
parsed_date_label->evaluate(0, *this, new_date, temp_pos);
}
set_date(new_date);
}
if (label_ptr)
label_ptr->count += 1;
}
void reference::immediate_compute_label()
{
if (label_ptr)
label_ptr->total = 2; // force use of disambiguator
compute_label();
}
int reference::merge_labels(reference **v, int n, label_type type,
string &result)
{
if (abbreviate_label_ranges)
return merge_labels_by_number(v, n, type, result);
else
return merge_labels_by_parts(v, n, type, result);
}
int reference::merge_labels_by_number(reference **v, int n, label_type type,
string &result)
{
if (n <= 1)
return 0;
int num = get_number();
// Only merge three or more labels.
if (v[0]->get_number() != num + 1
|| v[1]->get_number() != num + 2)
return 0;
int i;
for (i = 2; i < n; i++)
if (v[i]->get_number() != num + i + 1)
break;
result = get_label(type);
result += label_range_indicator;
result += v[i - 1]->get_label(type);
return i;
}
const substring_position &reference::get_separator_pos(label_type type) const
{
if (type == SHORT_LABEL && short_label_flag)
return short_separator_pos;
else
return separator_pos;
}
const string &reference::get_label(label_type type) const
{
if (type == SHORT_LABEL && short_label_flag)
return short_label;
else
return label;
}
int reference::merge_labels_by_parts(reference **v, int n, label_type type,
string &result)
{
if (n <= 0)
return 0;
const string &lb = get_label(type);
const substring_position &sp = get_separator_pos(type);
if (sp.start < 0
|| sp.start != v[0]->get_separator_pos(type).start
|| memcmp(lb.contents(), v[0]->get_label(type).contents(),
sp.start) != 0)
return 0;
result = lb;
int i = 0;
do {
result += separate_label_second_parts;
const substring_position &s = v[i]->get_separator_pos(type);
int sep_end_pos = s.start + s.length;
result.append(v[i]->get_label(type).contents() + sep_end_pos,
v[i]->get_label(type).length() - sep_end_pos);
} while (++i < n
&& sp.start == v[i]->get_separator_pos(type).start
&& memcmp(lb.contents(), v[i]->get_label(type).contents(),
sp.start) == 0);
return i;
}
string label_pool;
label_info::label_info(const string &s)
: count(0), total(1), length(s.length()), start(label_pool.length())
{
label_pool += s;
}
static label_info **label_table = 0;
static int label_table_size = 0;
static int label_table_used = 0;
label_info *lookup_label(const string &label)
{
if (label_table == 0) {
label_table = new label_info *[17];
label_table_size = 17;
for (int i = 0; i < 17; i++)
label_table[i] = 0;
}
unsigned h = hash_string(label.contents(), label.length()) % label_table_size;
label_info **ptr;
for (ptr = label_table + h;
*ptr != 0;
(ptr == label_table)
? (ptr = label_table + label_table_size - 1)
: ptr--)
if ((*ptr)->length == label.length()
&& memcmp(label_pool.contents() + (*ptr)->start, label.contents(),
label.length()) == 0) {
(*ptr)->total += 1;
return *ptr;
}
label_info *result = *ptr = new label_info(label);
if (++label_table_used * 2 > label_table_size) {
// Rehash the table.
label_info **old_table = label_table;
int old_size = label_table_size;
label_table_size = next_size(label_table_size);
label_table = new label_info *[label_table_size];
int i;
for (i = 0; i < label_table_size; i++)
label_table[i] = 0;
for (i = 0; i < old_size; i++)
if (old_table[i]) {
unsigned h = hash_string(label_pool.contents() + old_table[i]->start,
old_table[i]->length);
label_info **p;
for (p = label_table + (h % label_table_size);
*p != 0;
(p == label_table)
? (p = label_table + label_table_size - 1)
: --p)
;
*p = old_table[i];
}
a_delete old_table;
}
return result;
}
void clear_labels()
{
for (int i = 0; i < label_table_size; i++) {
delete label_table[i];
label_table[i] = 0;
}
label_table_used = 0;
label_pool.clear();
}
static void consider_authors(reference **start, reference **end, int i);
void compute_labels(reference **v, int n)
{
if (parsed_label
&& (parsed_label->analyze() & expression::CONTAINS_AT)
&& sort_fields.length() >= 2
&& sort_fields[0] == 'A'
&& sort_fields[1] == '+')
consider_authors(v, v + n, 0);
for (int i = 0; i < n; i++)
v[i]->compute_label();
}
/* A reference with a list of authors <A0,A1,...,AN> _needs_ author i
where 0 <= i <= N if there exists a reference with a list of authors
<B0,B1,...,BM> such that <A0,A1,...,AN> != <B0,B1,...,BM> and M >= i
and Aj = Bj for 0 <= j < i. In this case if we can't say ``A0,
A1,...,A(i-1) et al'' because this would match both <A0,A1,...,AN> and
<B0,B1,...,BM>. If a reference needs author i we only have to call
need_author(j) for some j >= i such that the reference also needs
author j. */
/* This function handles 2 tasks:
determine which authors are needed (cannot be elided with et al.);
determine which authors can have only last names in the labels.
References >= start and < end have the same first i author names.
Also they're sorted by A+. */
static void consider_authors(reference **start, reference **end, int i)
{
if (start >= end)
return;
reference **p = start;
if (i >= (*p)->get_nauthors()) {
for (++p; p < end && i >= (*p)->get_nauthors(); p++)
;
if (p < end && i > 0) {
// If we have an author list <A B C> and an author list <A B C D>,
// then both lists need C.
for (reference **q = start; q < end; q++)
(*q)->need_author(i - 1);
}
start = p;
}
while (p < end) {
reference **last_name_start = p;
reference **name_start = p;
for (++p;
p < end && i < (*p)->get_nauthors()
&& same_author_last_name(**last_name_start, **p, i);
p++) {
if (!same_author_name(**name_start, **p, i)) {
consider_authors(name_start, p, i + 1);
name_start = p;
}
}
consider_authors(name_start, p, i + 1);
if (last_name_start == name_start) {
for (reference **q = last_name_start; q < p; q++)
(*q)->set_last_name_unambiguous(i);
}
// If we have an author list <A B C D> and <A B C E>, then the lists
// need author D and E respectively.
if (name_start > start || p < end) {
for (reference **q = last_name_start; q < p; q++)
(*q)->need_author(i);
}
}
}
int same_author_last_name(const reference &r1, const reference &r2, int n)
{
const char *ae1;
const char *as1 = r1.get_sort_field(0, n, 0, &ae1);
assert(as1 != 0);
const char *ae2;
const char *as2 = r2.get_sort_field(0, n, 0, &ae2);
assert(as2 != 0);
return ae1 - as1 == ae2 - as2 && memcmp(as1, as2, ae1 - as1) == 0;
}
int same_author_name(const reference &r1, const reference &r2, int n)
{
const char *ae1;
const char *as1 = r1.get_sort_field(0, n, -1, &ae1);
assert(as1 != 0);
const char *ae2;
const char *as2 = r2.get_sort_field(0, n, -1, &ae2);
assert(as2 != 0);
return ae1 - as1 == ae2 - as2 && memcmp(as1, as2, ae1 - as1) == 0;
}
void int_set::set(int i)
{
assert(i >= 0);
int bytei = i >> 3;
if (bytei >= v.length()) {
int old_length = v.length();
v.set_length(bytei + 1);
for (int j = old_length; j <= bytei; j++)
v[j] = 0;
}
v[bytei] |= 1 << (i & 7);
}
int int_set::get(int i) const
{
assert(i >= 0);
int bytei = i >> 3;
return bytei >= v.length() ? 0 : (v[bytei] & (1 << (i & 7))) != 0;
}
void reference::set_last_name_unambiguous(int i)
{
last_name_unambiguous.set(i);
}
void reference::need_author(int n)
{
if (n > last_needed_author)
last_needed_author = n;
}
const char *reference::get_authors(const char **end) const
{
if (!computed_authors) {
((reference *)this)->computed_authors = 1;
string &result = ((reference *)this)->authors;
int na = get_nauthors();
result.clear();
for (int i = 0; i < na; i++) {
if (last_name_unambiguous.get(i)) {
const char *e, *start = get_author_last_name(i, &e);
assert(start != 0);
result.append(start, e - start);
}
else {
const char *e, *start = get_author(i, &e);
assert(start != 0);
result.append(start, e - start);
}
if (i == last_needed_author
&& et_al.length() > 0
&& et_al_min_elide > 0
&& last_needed_author + et_al_min_elide < na
&& na >= et_al_min_total) {
result += et_al;
break;
}
if (i < na - 1) {
if (na == 2)
result += join_authors_exactly_two;
else if (i < na - 2)
result += join_authors_default;
else
result += join_authors_last_two;
}
}
}
const char *start = authors.contents();
*end = start + authors.length();
return start;
}
int reference::get_nauthors() const
{
if (nauthors < 0) {
const char *dummy;
int na;
for (na = 0; get_author(na, &dummy) != 0; na++)
;
((reference *)this)->nauthors = na;
}
return nauthors;
}