/* Copyright (C) 1989-2020 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 3 of the License, 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 this program. If not, see . */
#include "troff.h"
#include "dictionary.h"
#include "hvunits.h"
#include "stringclass.h"
#include "mtsm.h"
#include "env.h"
#include "request.h"
#include "node.h"
#include "token.h"
#include "div.h"
#include "reg.h"
#include "font.h"
#include "charinfo.h"
#include "macropath.h"
#include "input.h"
#include
symbol default_family("T");
enum { ADJUST_LEFT = 0,
ADJUST_BOTH = 1,
ADJUST_CENTER = 3,
ADJUST_RIGHT = 5,
ADJUST_MAX = 5
};
enum {
// Not all combinations are valid; see hyphenate_request() below.
HYPHEN_NONE = 0,
HYPHEN_DEFAULT = 1,
HYPHEN_NOT_LAST_LINE = 2,
HYPHEN_NOT_LAST_CHARS = 4,
HYPHEN_NOT_FIRST_CHARS = 8,
HYPHEN_LAST_CHAR = 16,
HYPHEN_FIRST_CHAR = 32,
HYPHEN_MAX = 63
};
struct env_list_node {
environment *env;
env_list_node *next;
env_list_node(environment *e, env_list_node *p) : env(e), next(p) {}
};
env_list_node *env_stack;
dictionary env_dictionary(10);
environment *curenv;
static int next_line_number = 0;
extern int suppress_push;
extern statem *get_diversion_state();
charinfo *field_delimiter_char;
charinfo *padding_indicator_char;
int translate_space_to_dummy = 0;
class pending_output_line {
node *nd;
int no_fill;
int was_centered;
vunits vs;
vunits post_vs;
hunits width;
#ifdef WIDOW_CONTROL
int last_line; // Is it the last line of the paragraph?
#endif /* WIDOW_CONTROL */
public:
pending_output_line *next;
pending_output_line(node *, int, vunits, vunits, hunits, int,
pending_output_line * = 0);
~pending_output_line();
int output();
#ifdef WIDOW_CONTROL
friend void environment::mark_last_line();
friend void environment::output(node *, int, vunits, vunits, hunits, int);
#endif /* WIDOW_CONTROL */
};
pending_output_line::pending_output_line(node *n, int nf, vunits v, vunits pv,
hunits w, int ce,
pending_output_line *p)
: nd(n), no_fill(nf), was_centered(ce), vs(v), post_vs(pv), width(w),
#ifdef WIDOW_CONTROL
last_line(0),
#endif /* WIDOW_CONTROL */
next(p)
{
}
pending_output_line::~pending_output_line()
{
delete_node_list(nd);
}
int pending_output_line::output()
{
if (trap_sprung_flag)
return 0;
#ifdef WIDOW_CONTROL
if (next && next->last_line && !no_fill) {
curdiv->need(vs + post_vs + vunits(vresolution));
if (trap_sprung_flag) {
next->last_line = 0; // Try to avoid infinite loops.
return 0;
}
}
#endif
curenv->construct_format_state(nd, was_centered, !no_fill);
curdiv->output(nd, no_fill, vs, post_vs, width);
nd = 0;
return 1;
}
void environment::output(node *nd, int no_fill_flag,
vunits vs, vunits post_vs,
hunits width, int was_centered)
{
#ifdef WIDOW_CONTROL
while (pending_lines) {
if (widow_control && !pending_lines->no_fill && !pending_lines->next)
break;
if (!pending_lines->output())
break;
pending_output_line *tem = pending_lines;
pending_lines = pending_lines->next;
delete tem;
}
#else /* WIDOW_CONTROL */
output_pending_lines();
#endif /* WIDOW_CONTROL */
if (!trap_sprung_flag && !pending_lines
#ifdef WIDOW_CONTROL
&& (!widow_control || no_fill_flag)
#endif /* WIDOW_CONTROL */
) {
curenv->construct_format_state(nd, was_centered, !no_fill_flag);
curdiv->output(nd, no_fill_flag, vs, post_vs, width);
} else {
pending_output_line **p;
for (p = &pending_lines; *p; p = &(*p)->next)
;
*p = new pending_output_line(nd, no_fill_flag, vs, post_vs, width,
was_centered);
}
}
// a line from .tl goes at the head of the queue
void environment::output_title(node *nd, int no_fill_flag,
vunits vs, vunits post_vs,
hunits width)
{
if (!trap_sprung_flag)
curdiv->output(nd, no_fill_flag, vs, post_vs, width);
else
pending_lines = new pending_output_line(nd, no_fill_flag, vs, post_vs,
width, 0, pending_lines);
}
void environment::output_pending_lines()
{
while (pending_lines && pending_lines->output()) {
pending_output_line *tem = pending_lines;
pending_lines = pending_lines->next;
delete tem;
}
}
#ifdef WIDOW_CONTROL
void environment::mark_last_line()
{
if (!widow_control || !pending_lines)
return;
pending_output_line *p;
for (p = pending_lines; p->next; p = p->next)
;
if (!p->no_fill)
p->last_line = 1;
}
void widow_control_request()
{
int n;
if (has_arg() && get_integer(&n))
curenv->widow_control = n != 0;
else
curenv->widow_control = 1;
skip_line();
}
#endif /* WIDOW_CONTROL */
/* font_size functions */
size_range *font_size::size_table = 0;
int font_size::nranges = 0;
extern "C" {
int compare_ranges(const void *p1, const void *p2)
{
return ((size_range *)p1)->min - ((size_range *)p2)->min;
}
}
void font_size::init_size_table(int *sizes)
{
nranges = 0;
while (sizes[nranges*2] != 0)
nranges++;
assert(nranges > 0);
size_table = new size_range[nranges];
for (int i = 0; i < nranges; i++) {
size_table[i].min = sizes[i*2];
size_table[i].max = sizes[i*2 + 1];
}
qsort(size_table, nranges, sizeof(size_range), compare_ranges);
}
font_size::font_size(int sp)
{
for (int i = 0; i < nranges; i++) {
if (sp < size_table[i].min) {
if (i > 0 && size_table[i].min - sp >= sp - size_table[i - 1].max)
p = size_table[i - 1].max;
else
p = size_table[i].min;
return;
}
if (sp <= size_table[i].max) {
p = sp;
return;
}
}
p = size_table[nranges - 1].max;
}
int font_size::to_units()
{
return scale(p, units_per_inch, sizescale*72);
}
// we can't do this in a static constructor because various dictionaries
// have to get initialized first
static symbol default_environment_name("0");
void init_environments()
{
curenv = new environment(default_environment_name);
(void)env_dictionary.lookup(default_environment_name, curenv);
}
void tab_character()
{
curenv->tab_char = get_optional_char();
skip_line();
}
void leader_character()
{
curenv->leader_char = get_optional_char();
skip_line();
}
void environment::add_char(charinfo *ci)
{
int s;
node *gc_np = 0;
if (interrupted)
;
// don't allow fields in dummy environments
else if (ci == field_delimiter_char && !dummy) {
if (current_field)
wrap_up_field();
else
start_field();
}
else if (current_field && ci == padding_indicator_char)
add_padding();
else if (current_tab) {
if (tab_contents == 0)
tab_contents = new line_start_node;
if (ci != hyphen_indicator_char)
tab_contents = tab_contents->add_char(ci, this, &tab_width, &s, &gc_np);
else
tab_contents = tab_contents->add_discretionary_hyphen();
}
else {
if (line == 0)
start_line();
#if 0
fprintf(stderr, "current line is\n");
line->debug_node_list();
#endif
if (ci != hyphen_indicator_char)
line = line->add_char(ci, this, &width_total, &space_total, &gc_np);
else
line = line->add_discretionary_hyphen();
}
#if 0
fprintf(stderr, "now after we have added character the line is\n");
line->debug_node_list();
#endif
if ((!suppress_push) && gc_np) {
if (gc_np && (gc_np->state == 0)) {
gc_np->state = construct_state(0);
gc_np->push_state = get_diversion_state();
}
else if (line && (line->state == 0)) {
line->state = construct_state(0);
line->push_state = get_diversion_state();
}
}
#if 0
fprintf(stderr, "now we have possibly added the state the line is\n");
line->debug_node_list();
#endif
}
node *environment::make_char_node(charinfo *ci)
{
return make_node(ci, this);
}
void environment::add_node(node *n)
{
if (n == 0)
return;
if (!suppress_push) {
if (n->is_special && n->state == NULL)
n->state = construct_state(0);
n->push_state = get_diversion_state();
}
if (current_tab || current_field)
n->freeze_space();
if (interrupted) {
delete n;
}
else if (current_tab) {
n->next = tab_contents;
tab_contents = n;
tab_width += n->width();
}
else {
if (line == 0) {
if (discarding && n->discardable()) {
// XXX possibly: input_line_start -= n->width();
delete n;
return;
}
start_line();
}
width_total += n->width();
space_total += n->nspaces();
n->next = line;
line = n;
construct_new_line_state(line);
}
}
void environment::add_hyphen_indicator()
{
if (current_tab || interrupted || current_field
|| hyphen_indicator_char != 0)
return;
if (line == 0)
start_line();
line = line->add_discretionary_hyphen();
}
int environment::get_hyphenation_flags()
{
return hyphenation_flags;
}
int environment::get_hyphen_line_max()
{
return hyphen_line_max;
}
int environment::get_hyphen_line_count()
{
return hyphen_line_count;
}
int environment::get_center_lines()
{
return center_lines;
}
int environment::get_right_justify_lines()
{
return right_justify_lines;
}
int environment::get_no_number_count()
{
return no_number_count;
}
void environment::add_italic_correction()
{
if (current_tab) {
if (tab_contents)
tab_contents = tab_contents->add_italic_correction(&tab_width);
}
else if (line)
line = line->add_italic_correction(&width_total);
}
void environment::space_newline()
{
assert(!current_tab && !current_field);
if (interrupted)
return;
hunits x = H0;
hunits sw = env_space_width(this);
hunits ssw = env_sentence_space_width(this);
if (!translate_space_to_dummy) {
x = sw;
if (node_list_ends_sentence(line) == 1)
x += ssw;
}
width_list *w = new width_list(sw, ssw);
if (node_list_ends_sentence(line) == 1)
w->next = new width_list(sw, ssw);
if (line != 0 && line->merge_space(x, sw, ssw)) {
width_total += x;
return;
}
add_node(new word_space_node(x, get_fill_color(), w));
possibly_break_line(0, spread_flag);
spread_flag = 0;
}
void environment::space()
{
space(env_space_width(this), env_sentence_space_width(this));
}
void environment::space(hunits space_width, hunits sentence_space_width)
{
if (interrupted)
return;
if (current_field && padding_indicator_char == 0) {
add_padding();
return;
}
hunits x = translate_space_to_dummy ? H0 : space_width;
node *p = current_tab ? tab_contents : line;
hunits *tp = current_tab ? &tab_width : &width_total;
if (p && p->nspaces() == 1 && p->width() == x
&& node_list_ends_sentence(p->next) == 1) {
hunits xx = translate_space_to_dummy ? H0 : sentence_space_width;
if (p->merge_space(xx, space_width, sentence_space_width)) {
*tp += xx;
return;
}
}
if (p && p->merge_space(x, space_width, sentence_space_width)) {
*tp += x;
return;
}
add_node(new word_space_node(x,
get_fill_color(),
new width_list(space_width,
sentence_space_width)));
possibly_break_line(0, spread_flag);
spread_flag = 0;
}
static node *do_underline_special(bool do_underline_spaces)
{
macro m;
m.append_str("x u ");
m.append(do_underline_spaces ? '1' : '0');
return new special_node(m, 1);
}
bool environment::set_font(symbol nm)
{
if (interrupted) {
warning(WARN_FONT, "ignoring font selection on interrupted line");
return true; // "no operation" is successful
}
if (nm == symbol("P") || nm.is_empty()) {
if (family->make_definite(prev_fontno) < 0)
return false;
int tem = fontno;
fontno = prev_fontno;
prev_fontno = tem;
}
else {
prev_fontno = fontno;
int n = symbol_fontno(nm);
if (n < 0) {
n = next_available_font_position();
if (!mount_font(n, nm))
return false;
}
if (family->make_definite(n) < 0)
return false;
fontno = n;
}
if (underline_spaces && fontno != prev_fontno) {
if (fontno == get_underline_fontno())
add_node(do_underline_special(true));
if (prev_fontno == get_underline_fontno())
add_node(do_underline_special(false));
}
return true;
}
bool environment::set_font(int n)
{
if (interrupted)
return false;
if (is_good_fontno(n)) {
prev_fontno = fontno;
fontno = n;
}
else {
warning(WARN_FONT, "no font mounted at position %1", n);
return false;
}
return true;
}
void environment::set_family(symbol fam)
{
if (interrupted)
return;
if (fam.is_null() || fam.is_empty()) {
int previous_mounting_position = prev_family->make_definite(fontno);
assert(previous_mounting_position >= 0);
if (previous_mounting_position < 0)
return;
font_family *tem = family;
family = prev_family;
prev_family = tem;
}
else {
font_family *f = lookup_family(fam);
// If the family isn't already in the dictionary, looking it up will
// create an entry for it. That doesn't mean that it will be
// resolvable to a real font when combined with a style name.
assert((f != 0 /* nullptr */) &&
(0 != "font family dictionary lookup"));
if (0 /* nullptr */ == f)
return;
if (f->make_definite(fontno) < 0) {
error("no font family named '%1' exists", fam.contents());
return;
}
prev_family = family;
family = f;
}
}
void environment::set_size(int n)
{
if (interrupted)
return;
if (n == 0) {
font_size temp = prev_size;
prev_size = size;
size = temp;
int temp2 = prev_requested_size;
prev_requested_size = requested_size;
requested_size = temp2;
}
else {
prev_size = size;
size = font_size(n);
prev_requested_size = requested_size;
requested_size = n;
}
}
void environment::set_char_height(int n)
{
if (interrupted)
return;
if (n == requested_size || n <= 0)
char_height = 0;
else
char_height = n;
}
void environment::set_char_slant(int n)
{
if (interrupted)
return;
char_slant = n;
}
color *environment::get_prev_glyph_color()
{
return prev_glyph_color;
}
color *environment::get_glyph_color()
{
return glyph_color;
}
color *environment::get_prev_fill_color()
{
return prev_fill_color;
}
color *environment::get_fill_color()
{
return fill_color;
}
void environment::set_glyph_color(color *c)
{
if (interrupted)
return;
curenv->prev_glyph_color = curenv->glyph_color;
curenv->glyph_color = c;
}
void environment::set_fill_color(color *c)
{
if (interrupted)
return;
curenv->prev_fill_color = curenv->fill_color;
curenv->fill_color = c;
}
environment::environment(symbol nm)
: dummy(0),
prev_line_length((units_per_inch*13)/2),
line_length((units_per_inch*13)/2),
prev_title_length((units_per_inch*13)/2),
title_length((units_per_inch*13)/2),
prev_size(sizescale*10),
size(sizescale*10),
requested_size(sizescale*10),
prev_requested_size(sizescale*10),
char_height(0),
char_slant(0),
space_size(12),
sentence_space_size(12),
adjust_mode(ADJUST_BOTH),
fill(1),
interrupted(0),
prev_line_interrupted(0),
center_lines(0),
right_justify_lines(0),
prev_vertical_spacing(points_to_units(12)),
vertical_spacing(points_to_units(12)),
prev_post_vertical_spacing(0),
post_vertical_spacing(0),
prev_line_spacing(1),
line_spacing(1),
prev_indent(0),
indent(0),
temporary_indent(0),
have_temporary_indent(0),
underline_lines(0),
underline_spaces(0),
input_trap_count(0),
continued_input_trap(0),
line(0),
prev_text_length(0),
width_total(0),
space_total(0),
input_line_start(0),
line_tabs(0),
current_tab(TAB_NONE),
leader_node(0),
tab_char(0),
leader_char(charset_table['.']),
current_field(0),
discarding(0),
spread_flag(0),
margin_character_flags(0),
margin_character_node(0),
margin_character_distance(points_to_units(10)),
numbering_nodes(0),
number_text_separation(1),
line_number_indent(0),
line_number_multiple(1),
no_number_count(0),
hyphenation_flags(1),
hyphen_line_count(0),
hyphen_line_max(-1),
hyphenation_space(H0),
hyphenation_margin(H0),
composite(0),
pending_lines(0),
#ifdef WIDOW_CONTROL
widow_control(0),
#endif /* WIDOW_CONTROL */
glyph_color(&default_color),
prev_glyph_color(&default_color),
fill_color(&default_color),
prev_fill_color(&default_color),
seen_space(0),
seen_eol(0),
suppress_next_eol(0),
seen_break(0),
tabs(units_per_inch/2, TAB_LEFT),
name(nm),
control_char('.'),
no_break_control_char('\''),
hyphen_indicator_char(0)
{
prev_family = family = lookup_family(default_family);
prev_fontno = fontno = 1;
if (!is_good_fontno(1))
fatal("font mounted at position 1 is not valid");
if (family->make_definite(1) < 0)
fatal("invalid default font family '%1'",
default_family.contents());
prev_fontno = fontno;
}
environment::environment(const environment *e)
: dummy(1),
prev_line_length(e->prev_line_length),
line_length(e->line_length),
prev_title_length(e->prev_title_length),
title_length(e->title_length),
prev_size(e->prev_size),
size(e->size),
requested_size(e->requested_size),
prev_requested_size(e->prev_requested_size),
char_height(e->char_height),
char_slant(e->char_slant),
prev_fontno(e->prev_fontno),
fontno(e->fontno),
prev_family(e->prev_family),
family(e->family),
space_size(e->space_size),
sentence_space_size(e->sentence_space_size),
adjust_mode(e->adjust_mode),
fill(e->fill),
interrupted(0),
prev_line_interrupted(0),
center_lines(0),
right_justify_lines(0),
prev_vertical_spacing(e->prev_vertical_spacing),
vertical_spacing(e->vertical_spacing),
prev_post_vertical_spacing(e->prev_post_vertical_spacing),
post_vertical_spacing(e->post_vertical_spacing),
prev_line_spacing(e->prev_line_spacing),
line_spacing(e->line_spacing),
prev_indent(e->prev_indent),
indent(e->indent),
temporary_indent(0),
have_temporary_indent(0),
underline_lines(0),
underline_spaces(0),
input_trap_count(0),
continued_input_trap(0),
line(0),
prev_text_length(e->prev_text_length),
width_total(0),
space_total(0),
input_line_start(0),
line_tabs(e->line_tabs),
current_tab(TAB_NONE),
leader_node(0),
tab_char(e->tab_char),
leader_char(e->leader_char),
current_field(0),
discarding(0),
spread_flag(0),
margin_character_flags(e->margin_character_flags),
margin_character_node(e->margin_character_node),
margin_character_distance(e->margin_character_distance),
numbering_nodes(0),
number_text_separation(e->number_text_separation),
line_number_indent(e->line_number_indent),
line_number_multiple(e->line_number_multiple),
no_number_count(e->no_number_count),
hyphenation_flags(e->hyphenation_flags),
hyphen_line_count(0),
hyphen_line_max(e->hyphen_line_max),
hyphenation_space(e->hyphenation_space),
hyphenation_margin(e->hyphenation_margin),
composite(0),
pending_lines(0),
#ifdef WIDOW_CONTROL
widow_control(e->widow_control),
#endif /* WIDOW_CONTROL */
glyph_color(e->glyph_color),
prev_glyph_color(e->prev_glyph_color),
fill_color(e->fill_color),
prev_fill_color(e->prev_fill_color),
seen_space(e->seen_space),
seen_eol(e->seen_eol),
suppress_next_eol(e->suppress_next_eol),
seen_break(e->seen_break),
tabs(e->tabs),
name(e->name), // so that, e.g., '.if "\n[.ev]"0"' works
control_char(e->control_char),
no_break_control_char(e->no_break_control_char),
hyphen_indicator_char(e->hyphen_indicator_char)
{
}
void environment::copy(const environment *e)
{
prev_line_length = e->prev_line_length;
line_length = e->line_length;
prev_title_length = e->prev_title_length;
title_length = e->title_length;
prev_size = e->prev_size;
size = e->size;
prev_requested_size = e->prev_requested_size;
requested_size = e->requested_size;
char_height = e->char_height;
char_slant = e->char_slant;
space_size = e->space_size;
sentence_space_size = e->sentence_space_size;
adjust_mode = e->adjust_mode;
fill = e->fill;
interrupted = 0;
prev_line_interrupted = 0;
center_lines = 0;
right_justify_lines = 0;
prev_vertical_spacing = e->prev_vertical_spacing;
vertical_spacing = e->vertical_spacing;
prev_post_vertical_spacing = e->prev_post_vertical_spacing,
post_vertical_spacing = e->post_vertical_spacing,
prev_line_spacing = e->prev_line_spacing;
line_spacing = e->line_spacing;
prev_indent = e->prev_indent;
indent = e->indent;
have_temporary_indent = 0;
temporary_indent = 0;
underline_lines = 0;
underline_spaces = 0;
input_trap_count = 0;
continued_input_trap = 0;
prev_text_length = e->prev_text_length;
width_total = 0;
space_total = 0;
input_line_start = 0;
control_char = e->control_char;
no_break_control_char = e->no_break_control_char;
hyphen_indicator_char = e->hyphen_indicator_char;
spread_flag = 0;
line = 0;
pending_lines = 0;
discarding = 0;
tabs = e->tabs;
line_tabs = e->line_tabs;
current_tab = TAB_NONE;
current_field = 0;
margin_character_flags = e->margin_character_flags;
if (e->margin_character_node)
margin_character_node = e->margin_character_node->copy();
margin_character_distance = e->margin_character_distance;
numbering_nodes = 0;
number_text_separation = e->number_text_separation;
line_number_multiple = e->line_number_multiple;
line_number_indent = e->line_number_indent;
no_number_count = e->no_number_count;
tab_char = e->tab_char;
leader_char = e->leader_char;
hyphenation_flags = e->hyphenation_flags;
fontno = e->fontno;
prev_fontno = e->prev_fontno;
dummy = e->dummy;
family = e->family;
prev_family = e->prev_family;
leader_node = 0;
#ifdef WIDOW_CONTROL
widow_control = e->widow_control;
#endif /* WIDOW_CONTROL */
hyphen_line_max = e->hyphen_line_max;
hyphen_line_count = 0;
hyphenation_space = e->hyphenation_space;
hyphenation_margin = e->hyphenation_margin;
composite = 0;
glyph_color= e->glyph_color;
prev_glyph_color = e->prev_glyph_color;
fill_color = e->fill_color;
prev_fill_color = e->prev_fill_color;
}
environment::~environment()
{
delete leader_node;
delete_node_list(line);
delete_node_list(numbering_nodes);
}
hunits environment::get_input_line_position()
{
hunits n;
if (line == 0)
n = -input_line_start;
else
n = width_total - input_line_start;
if (current_tab)
n += tab_width;
return n;
}
void environment::set_input_line_position(hunits n)
{
input_line_start = line == 0 ? -n : width_total - n;
if (current_tab)
input_line_start += tab_width;
}
hunits environment::get_line_length()
{
return line_length;
}
hunits environment::get_saved_line_length()
{
if (line)
return target_text_length + saved_indent;
else
return line_length;
}
vunits environment::get_vertical_spacing()
{
return vertical_spacing;
}
vunits environment::get_post_vertical_spacing()
{
return post_vertical_spacing;
}
int environment::get_line_spacing()
{
return line_spacing;
}
vunits environment::total_post_vertical_spacing()
{
vunits tem(post_vertical_spacing);
if (line_spacing > 1)
tem += (line_spacing - 1)*vertical_spacing;
return tem;
}
int environment::get_bold()
{
return get_bold_fontno(fontno);
}
hunits environment::get_digit_width()
{
return env_digit_width(this);
}
int environment::get_adjust_mode()
{
return adjust_mode;
}
int environment::get_fill()
{
return fill;
}
hunits environment::get_indent()
{
return indent;
}
hunits environment::get_saved_indent()
{
if (line)
return saved_indent;
else if (have_temporary_indent)
return temporary_indent;
else
return indent;
}
hunits environment::get_temporary_indent()
{
return temporary_indent;
}
hunits environment::get_title_length()
{
return title_length;
}
node *environment::get_prev_char()
{
for (node *n = current_tab ? tab_contents : line; n; n = n->next) {
node *last = n->last_char_node();
if (last)
return last;
}
return 0;
}
hunits environment::get_prev_char_width()
{
node *last = get_prev_char();
if (!last)
return H0;
return last->width();
}
hunits environment::get_prev_char_skew()
{
node *last = get_prev_char();
if (!last)
return H0;
return last->skew();
}
vunits environment::get_prev_char_height()
{
node *last = get_prev_char();
if (!last)
return V0;
vunits min, max;
last->vertical_extent(&min, &max);
return -min;
}
vunits environment::get_prev_char_depth()
{
node *last = get_prev_char();
if (!last)
return V0;
vunits min, max;
last->vertical_extent(&min, &max);
return max;
}
hunits environment::get_text_length()
{
hunits n = line == 0 ? H0 : width_total;
if (current_tab)
n += tab_width;
return n;
}
hunits environment::get_prev_text_length()
{
return prev_text_length;
}
static int sb_reg_contents = 0;
static int st_reg_contents = 0;
static int ct_reg_contents = 0;
static int rsb_reg_contents = 0;
static int rst_reg_contents = 0;
static int skw_reg_contents = 0;
static int ssc_reg_contents = 0;
void environment::width_registers()
{
// this is used to implement \w; it sets the st, sb, ct registers
vunits min = 0, max = 0, cur = 0;
int character_type = 0;
ssc_reg_contents = line ? line->subscript_correction().to_units() : 0;
skw_reg_contents = line ? line->skew().to_units() : 0;
line = reverse_node_list(line);
vunits real_min = V0;
vunits real_max = V0;
vunits v1, v2;
for (node *tem = line; tem; tem = tem->next) {
tem->vertical_extent(&v1, &v2);
v1 += cur;
if (v1 < real_min)
real_min = v1;
v2 += cur;
if (v2 > real_max)
real_max = v2;
if ((cur += tem->vertical_width()) < min)
min = cur;
else if (cur > max)
max = cur;
character_type |= tem->character_type();
}
line = reverse_node_list(line);
st_reg_contents = -min.to_units();
sb_reg_contents = -max.to_units();
rst_reg_contents = -real_min.to_units();
rsb_reg_contents = -real_max.to_units();
ct_reg_contents = character_type;
}
node *environment::extract_output_line()
{
if (current_tab)
wrap_up_tab();
node *n = line;
line = 0;
return n;
}
/* environment related requests */
void environment_switch()
{
if (curenv->is_dummy()) {
error("cannot switch out of dummy environment");
}
else {
symbol nm = get_long_name();
if (nm.is_null()) {
if (env_stack == 0)
error("environment stack underflow");
else {
int seen_space = curenv->seen_space;
int seen_eol = curenv->seen_eol;
int suppress_next_eol = curenv->suppress_next_eol;
curenv = env_stack->env;
curenv->seen_space = seen_space;
curenv->seen_eol = seen_eol;
curenv->suppress_next_eol = suppress_next_eol;
env_list_node *tem = env_stack;
env_stack = env_stack->next;
delete tem;
}
}
else {
environment *e = (environment *)env_dictionary.lookup(nm);
if (!e) {
e = new environment(nm);
(void)env_dictionary.lookup(nm, e);
}
env_stack = new env_list_node(curenv, env_stack);
curenv = e;
}
}
skip_line();
}
void environment_copy()
{
environment *e=0;
tok.skip();
symbol nm = get_long_name();
if (nm.is_null()) {
error("no environment specified to copy from");
}
else {
e = (environment *)env_dictionary.lookup(nm);
if (e)
curenv->copy(e);
else
error("cannot copy from nonexistent environment '%1'",
nm.contents());
}
skip_line();
}
void fill_color_change()
{
symbol s = get_name();
if (s.is_null())
curenv->set_fill_color(curenv->get_prev_fill_color());
else
do_fill_color(s);
skip_line();
}
void glyph_color_change()
{
symbol s = get_name();
if (s.is_null())
curenv->set_glyph_color(curenv->get_prev_glyph_color());
else
do_glyph_color(s);
skip_line();
}
static symbol P_symbol("P");
void font_change()
{
symbol s = get_name();
bool is_number = true;
if (s.is_null() || s == P_symbol) {
s = P_symbol;
is_number = false;
}
else {
for (const char *p = s.contents(); p != 0 && *p != 0; p++)
if (!csdigit(*p)) {
is_number = false;
break;
}
}
// environment::set_font warns if a bogus mounting position is
// requested. We must warn here if a bogus font name is selected.
if (is_number)
(void) curenv->set_font(atoi(s.contents()));
else
if (!curenv->set_font(s))
warning(WARN_FONT, "cannot select font '%1'", s.contents());
skip_line();
}
void family_change()
{
symbol s = get_name();
curenv->set_family(s);
skip_line();
}
void point_size()
{
int n;
if (has_arg() && get_number(&n, 'z', curenv->get_requested_point_size())) {
if (n <= 0)
n = 1;
curenv->set_size(n);
}
else
curenv->set_size(0);
skip_line();
}
void override_sizes()
{
int n = 16;
int *sizes = new int[n];
int i = 0;
char *buf = read_string();
if (!buf)
return;
char *p = strtok(buf, " \t");
for (;;) {
if (!p)
break;
int lower, upper;
switch (sscanf(p, "%d-%d", &lower, &upper)) {
case 1:
upper = lower;
// fall through
case 2:
if (lower <= upper && lower >= 0)
break;
// fall through
default:
warning(WARN_RANGE, "bad size range '%1'", p);
return;
}
if (i + 2 > n) {
int *old_sizes = sizes;
sizes = new int[n*2];
memcpy(sizes, old_sizes, n*sizeof(int));
n *= 2;
delete[] old_sizes;
}
sizes[i++] = lower;
if (lower == 0)
break;
sizes[i++] = upper;
p = strtok(0, " \t");
}
font_size::init_size_table(sizes);
}
void space_size()
{
int n;
if (get_integer(&n)) {
if (n < 0)
warning(WARN_RANGE, "negative word space size ignored: '%1'", n);
else
curenv->space_size = n;
if (has_arg() && get_integer(&n))
if (n < 0)
warning(WARN_RANGE, "negative sentence space size ignored: "
"'%1'", n);
else
curenv->sentence_space_size = n;
else
curenv->sentence_space_size = curenv->space_size;
}
skip_line();
}
void fill()
{
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break();
curenv->fill = 1;
tok.next();
}
void no_fill()
{
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break();
curenv->fill = 0;
curenv->suppress_next_eol = 1;
tok.next();
}
void center()
{
int n;
if (!has_arg() || !get_integer(&n))
n = 1;
else if (n < 0)
n = 0;
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break();
curenv->right_justify_lines = 0;
curenv->center_lines = n;
curdiv->modified_tag.incl(MTSM_CE);
tok.next();
}
void right_justify()
{
int n;
if (!has_arg() || !get_integer(&n))
n = 1;
else if (n < 0)
n = 0;
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break();
curenv->center_lines = 0;
curenv->right_justify_lines = n;
curdiv->modified_tag.incl(MTSM_RJ);
tok.next();
}
void line_length()
{
hunits temp;
hunits minimum_length = font::hor;
if (has_arg() && get_hunits(&temp, 'm', curenv->line_length)) {
if (temp < minimum_length) {
warning(WARN_RANGE, "invalid line length %1u rounded to device"
" horizontal motion quantum", temp.to_units());
temp = minimum_length;
}
}
else
temp = curenv->prev_line_length;
curenv->prev_line_length = curenv->line_length;
curenv->line_length = temp;
curdiv->modified_tag.incl(MTSM_LL);
skip_line();
}
void title_length()
{
hunits temp;
hunits minimum_length = font::hor;
if (has_arg() && get_hunits(&temp, 'm', curenv->title_length)) {
if (temp < minimum_length) {
warning(WARN_RANGE, "invalid title length %1u rounded to device"
" horizontal motion quantum", temp.to_units());
temp = minimum_length;
}
}
else
temp = curenv->prev_title_length;
curenv->prev_title_length = curenv->title_length;
curenv->title_length = temp;
skip_line();
}
void vertical_spacing()
{
vunits temp;
if (has_arg() && get_vunits(&temp, 'p', curenv->vertical_spacing)) {
if (temp < V0) {
warning(WARN_RANGE, "vertical spacing must not be negative");
temp = vresolution;
}
}
else
temp = curenv->prev_vertical_spacing;
curenv->prev_vertical_spacing = curenv->vertical_spacing;
curenv->vertical_spacing = temp;
skip_line();
}
void post_vertical_spacing()
{
vunits temp;
if (has_arg() && get_vunits(&temp, 'p', curenv->post_vertical_spacing)) {
if (temp < V0) {
warning(WARN_RANGE,
"post vertical spacing must be greater than or equal to 0");
temp = V0;
}
}
else
temp = curenv->prev_post_vertical_spacing;
curenv->prev_post_vertical_spacing = curenv->post_vertical_spacing;
curenv->post_vertical_spacing = temp;
skip_line();
}
void line_spacing()
{
int temp;
if (has_arg() && get_integer(&temp)) {
if (temp < 1) {
warning(WARN_RANGE, "value %1 out of range: interpreted as 1", temp);
temp = 1;
}
}
else
temp = curenv->prev_line_spacing;
curenv->prev_line_spacing = curenv->line_spacing;
curenv->line_spacing = temp;
skip_line();
}
void indent()
{
hunits temp;
if (has_arg() && get_hunits(&temp, 'm', curenv->indent)) {
if (temp < H0) {
warning(WARN_RANGE, "indent cannot be negative");
temp = H0;
}
}
else
temp = curenv->prev_indent;
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break();
curenv->have_temporary_indent = 0;
curenv->prev_indent = curenv->indent;
curenv->indent = temp;
curdiv->modified_tag.incl(MTSM_IN);
tok.next();
}
void temporary_indent()
{
int err = 0;
hunits temp;
if (!get_hunits(&temp, 'm', curenv->get_indent()))
err = 1;
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break();
if (temp < H0) {
warning(WARN_RANGE, "total indent cannot be negative");
temp = H0;
}
if (!err) {
curenv->temporary_indent = temp;
curenv->have_temporary_indent = 1;
curdiv->modified_tag.incl(MTSM_TI);
}
tok.next();
}
void do_underline(int underline_spaces)
{
int n;
if (!has_arg() || !get_integer(&n))
n = 1;
if (n <= 0) {
if (curenv->underline_lines > 0) {
curenv->prev_fontno = curenv->fontno;
curenv->fontno = curenv->pre_underline_fontno;
if (underline_spaces) {
curenv->underline_spaces = 0;
curenv->add_node(do_underline_special(false));
}
}
curenv->underline_lines = 0;
}
else {
curenv->underline_lines = n;
curenv->pre_underline_fontno = curenv->fontno;
curenv->fontno = get_underline_fontno();
if (underline_spaces) {
curenv->underline_spaces = 1;
curenv->add_node(do_underline_special(true));
}
}
skip_line();
}
void continuous_underline()
{
do_underline(1);
}
void underline()
{
do_underline(0);
}
void control_char()
{
curenv->control_char = '.';
if (has_arg()) {
if (tok.ch() == 0)
error("bad control character");
else
curenv->control_char = tok.ch();
}
skip_line();
}
void no_break_control_char()
{
curenv->no_break_control_char = '\'';
if (has_arg()) {
if (tok.ch() == 0)
error("bad control character");
else
curenv->no_break_control_char = tok.ch();
}
skip_line();
}
void margin_character()
{
while (tok.is_space())
tok.next();
charinfo *ci = tok.get_char();
if (ci) {
// Call tok.next() only after making the node so that
// .mc \s+9\(br\s0 works.
node *nd = curenv->make_char_node(ci);
tok.next();
if (nd) {
delete curenv->margin_character_node;
curenv->margin_character_node = nd;
curenv->margin_character_flags = MARGIN_CHARACTER_ON
| MARGIN_CHARACTER_NEXT;
hunits d;
if (has_arg() && get_hunits(&d, 'm'))
curenv->margin_character_distance = d;
}
}
else {
check_missing_character();
curenv->margin_character_flags &= ~MARGIN_CHARACTER_ON;
if (curenv->margin_character_flags == 0) {
delete curenv->margin_character_node;
curenv->margin_character_node = 0;
}
}
skip_line();
}
void number_lines()
{
delete_node_list(curenv->numbering_nodes);
curenv->numbering_nodes = 0;
if (has_arg()) {
node *nd = 0;
for (int i = '9'; i >= '0'; i--) {
node *tem = make_node(charset_table[i], curenv);
if (!tem) {
skip_line();
return;
}
tem->next = nd;
nd = tem;
}
curenv->numbering_nodes = nd;
curenv->line_number_digit_width = env_digit_width(curenv);
int n;
if (!tok.usable_as_delimiter()) {
if (get_integer(&n, next_line_number)) {
next_line_number = n;
if (next_line_number < 0) {
warning(WARN_RANGE, "negative line number");
next_line_number = 0;
}
}
}
else
while (!tok.is_space() && !tok.is_newline() && !tok.is_eof())
tok.next();
if (has_arg()) {
if (!tok.usable_as_delimiter()) {
if (get_integer(&n)) {
if (n <= 0) {
warning(WARN_RANGE, "negative or zero line number multiple");
}
else
curenv->line_number_multiple = n;
}
}
else
while (!tok.is_space() && !tok.is_newline() && !tok.is_eof())
tok.next();
if (has_arg()) {
if (!tok.usable_as_delimiter()) {
if (get_integer(&n))
curenv->number_text_separation = n;
}
else
while (!tok.is_space() && !tok.is_newline() && !tok.is_eof())
tok.next();
if (has_arg() && !tok.usable_as_delimiter() && get_integer(&n))
curenv->line_number_indent = n;
}
}
}
skip_line();
}
void no_number()
{
int n;
if (has_arg() && get_integer(&n))
curenv->no_number_count = n > 0 ? n : 0;
else
curenv->no_number_count = 1;
skip_line();
}
void no_hyphenate()
{
curenv->hyphenation_flags = 0;
skip_line();
}
void hyphenate_request()
{
int n;
if (has_arg() && get_integer(&n)) {
if (n < HYPHEN_NONE) {
warning(WARN_RANGE, "negative hyphenation flags ignored: %1", n);
} else if (n > HYPHEN_MAX) {
warning(WARN_RANGE, "unknown hyphenation flags ignored (maximum "
"%1): %2", HYPHEN_MAX, n);
} else if (((n & HYPHEN_DEFAULT) && (n & ~HYPHEN_DEFAULT))
|| ((n & HYPHEN_FIRST_CHAR) && (n & HYPHEN_NOT_FIRST_CHARS))
|| ((n & HYPHEN_LAST_CHAR) && (n & HYPHEN_NOT_LAST_CHARS)))
warning(WARN_SYNTAX, "contradictory hyphenation flags ignored: "
"%1", n);
else
curenv->hyphenation_flags = n;
}
else
curenv->hyphenation_flags = 1;
skip_line();
}
void hyphen_char()
{
curenv->hyphen_indicator_char = get_optional_char();
skip_line();
}
void hyphen_line_max_request()
{
int n;
if (has_arg() && get_integer(&n))
curenv->hyphen_line_max = n;
else
curenv->hyphen_line_max = -1;
skip_line();
}
void environment::interrupt()
{
if (!dummy) {
add_node(new transparent_dummy_node);
interrupted = 1;
}
}
void environment::newline()
{
int was_centered = 0;
if (underline_lines > 0) {
if (--underline_lines == 0) {
prev_fontno = fontno;
fontno = pre_underline_fontno;
if (underline_spaces) {
underline_spaces = 0;
add_node(do_underline_special(false));
}
}
}
if (current_field)
wrap_up_field();
if (current_tab)
wrap_up_tab();
// strip trailing spaces
while (line != 0 && line->discardable()) {
width_total -= line->width();
space_total -= line->nspaces();
node *tem = line;
line = line->next;
delete tem;
}
node *to_be_output = 0;
hunits to_be_output_width;
prev_line_interrupted = 0;
if (dummy)
space_newline();
else if (interrupted) {
interrupted = 0;
// see environment::final_break
prev_line_interrupted = is_exit_underway ? 2 : 1;
}
else if (center_lines > 0) {
--center_lines;
hunits x = target_text_length - width_total;
if (x > H0)
saved_indent += x/2;
to_be_output = line;
was_centered = 1;
to_be_output_width = width_total;
line = 0;
}
else if (right_justify_lines > 0) {
--right_justify_lines;
hunits x = target_text_length - width_total;
if (x > H0)
saved_indent += x;
to_be_output = line;
to_be_output_width = width_total;
line = 0;
}
else if (fill)
space_newline();
else {
to_be_output = line;
to_be_output_width = width_total;
line = 0;
}
input_line_start = line == 0 ? H0 : width_total;
if (to_be_output) {
if (is_html && !fill) {
curdiv->modified_tag.incl(MTSM_EOL);
if (suppress_next_eol)
suppress_next_eol = 0;
else
seen_eol = 1;
}
output_line(to_be_output, to_be_output_width, was_centered);
hyphen_line_count = 0;
}
if (input_trap_count > 0) {
if (!(continued_input_trap && prev_line_interrupted))
if (--input_trap_count == 0)
spring_trap(input_trap);
}
}
void environment::output_line(node *n, hunits width, int was_centered)
{
prev_text_length = width;
if (margin_character_flags) {
hunits d = line_length + margin_character_distance - saved_indent - width;
if (d > 0) {
n = new hmotion_node(d, get_fill_color(), n);
width += d;
}
margin_character_flags &= ~MARGIN_CHARACTER_NEXT;
node *tem;
if (!margin_character_flags) {
tem = margin_character_node;
margin_character_node = 0;
}
else
tem = margin_character_node->copy();
tem->next = n;
n = tem;
width += tem->width();
}
node *nn = 0;
while (n != 0) {
node *tem = n->next;
n->next = nn;
nn = n;
n = tem;
}
if (!saved_indent.is_zero())
nn = new hmotion_node(saved_indent, get_fill_color(), nn);
width += saved_indent;
if (no_number_count > 0)
--no_number_count;
else if (numbering_nodes) {
hunits w = (line_number_digit_width
*(3+line_number_indent+number_text_separation));
if (next_line_number % line_number_multiple != 0)
nn = new hmotion_node(w, get_fill_color(), nn);
else {
hunits x = w;
nn = new hmotion_node(number_text_separation * line_number_digit_width,
get_fill_color(), nn);
x -= number_text_separation*line_number_digit_width;
char buf[30];
sprintf(buf, "%3d", next_line_number);
for (char *p = strchr(buf, '\0') - 1; p >= buf && *p != ' '; --p) {
node *gn = numbering_nodes;
for (int count = *p - '0'; count > 0; count--)
gn = gn->next;
gn = gn->copy();
x -= gn->width();
gn->next = nn;
nn = gn;
}
nn = new hmotion_node(x, get_fill_color(), nn);
}
width += w;
++next_line_number;
}
output(nn, !fill, vertical_spacing, total_post_vertical_spacing(), width,
was_centered);
}
void environment::start_line()
{
assert(line == 0);
discarding = 0;
line = new line_start_node;
if (have_temporary_indent) {
saved_indent = temporary_indent;
have_temporary_indent = 0;
}
else
saved_indent = indent;
target_text_length = line_length - saved_indent;
width_total = H0;
space_total = 0;
}
hunits environment::get_hyphenation_space()
{
return hyphenation_space;
}
void hyphenation_space_request()
{
hunits n;
if (get_hunits(&n, 'm')) {
if (n < H0) {
warning(WARN_RANGE, "hyphenation space cannot be negative");
n = H0;
}
curenv->hyphenation_space = n;
}
skip_line();
}
hunits environment::get_hyphenation_margin()
{
return hyphenation_margin;
}
void hyphenation_margin_request()
{
hunits n;
if (get_hunits(&n, 'm')) {
if (n < H0) {
warning(WARN_RANGE, "hyphenation margin cannot be negative");
n = H0;
}
curenv->hyphenation_margin = n;
}
skip_line();
}
breakpoint *environment::choose_breakpoint()
{
hunits x = width_total;
int s = space_total;
node *n = line;
breakpoint *best_bp = 0; // the best breakpoint so far
int best_bp_fits = 0;
while (n != 0) {
x -= n->width();
s -= n->nspaces();
breakpoint *bp = n->get_breakpoints(x, s);
while (bp != 0) {
if (bp->width <= target_text_length) {
if (!bp->hyphenated) {
breakpoint *tem = bp->next;
bp->next = 0;
while (tem != 0) {
breakpoint *tem1 = tem;
tem = tem->next;
delete tem1;
}
if (best_bp_fits
// Decide whether to use the hyphenated breakpoint.
&& (hyphen_line_max < 0
// Only choose the hyphenated breakpoint if it would not
// exceed the maximum number of consecutive hyphenated
// lines.
|| hyphen_line_count + 1 <= hyphen_line_max)
&& !(adjust_mode == ADJUST_BOTH
// Don't choose the hyphenated breakpoint if the line
// can be justified by adding no more than
// hyphenation_space to any word space.
? (bp->nspaces > 0
&& (((target_text_length - bp->width
+ (bp->nspaces - 1)*hresolution)/bp->nspaces)
<= hyphenation_space))
// Don't choose the hyphenated breakpoint if the line
// is no more than hyphenation_margin short.
: target_text_length - bp->width <= hyphenation_margin)) {
delete bp;
return best_bp;
}
if (best_bp)
delete best_bp;
return bp;
}
else {
if ((adjust_mode == ADJUST_BOTH
? hyphenation_space == H0
: hyphenation_margin == H0)
&& (hyphen_line_max < 0
|| hyphen_line_count + 1 <= hyphen_line_max)) {
// No need to consider a non-hyphenated breakpoint.
if (best_bp)
delete best_bp;
breakpoint *tem = bp->next;
bp->next = 0;
while (tem != 0) {
breakpoint *tem1 = tem;
tem = tem->next;
delete tem1;
}
return bp;
}
// It fits but it's hyphenated.
if (!best_bp_fits) {
if (best_bp)
delete best_bp;
best_bp = bp;
bp = bp->next;
best_bp_fits = 1;
}
else {
breakpoint *tem = bp;
bp = bp->next;
delete tem;
}
}
}
else {
if (best_bp)
delete best_bp;
best_bp = bp;
bp = bp->next;
}
}
n = n->next;
}
if (best_bp) {
if (!best_bp_fits)
output_warning(WARN_BREAK, "cannot break line");
return best_bp;
}
return 0;
}
void environment::hyphenate_line(int start_here)
{
assert(line != 0);
hyphenation_type prev_type = line->get_hyphenation_type();
node **startp;
if (start_here)
startp = &line;
else
for (startp = &line->next; *startp != 0; startp = &(*startp)->next) {
hyphenation_type this_type = (*startp)->get_hyphenation_type();
if (prev_type == HYPHEN_BOUNDARY && this_type == HYPHEN_MIDDLE)
break;
prev_type = this_type;
}
if (*startp == 0)
return;
node *tem = *startp;
do {
tem = tem->next;
} while (tem != 0 && tem->get_hyphenation_type() == HYPHEN_MIDDLE);
int inhibit = (tem != 0 && tem->get_hyphenation_type() == HYPHEN_INHIBIT);
node *end = tem;
hyphen_list *sl = 0;
tem = *startp;
node *forward = 0;
int i = 0;
while (tem != end) {
sl = tem->get_hyphen_list(sl, &i);
node *tem1 = tem;
tem = tem->next;
tem1->next = forward;
forward = tem1;
}
if (!inhibit) {
// this is for characters like hyphen and emdash
int prev_code = 0;
for (hyphen_list *h = sl; h; h = h->next) {
h->breakable = (prev_code != 0
&& h->next != 0
&& h->next->hyphenation_code != 0);
prev_code = h->hyphenation_code;
}
}
if (hyphenation_flags != 0
&& !inhibit
// this may not be right if we have extra space on this line
&& !((hyphenation_flags & HYPHEN_NOT_LAST_LINE)
&& (curdiv->distance_to_next_trap()
<= vertical_spacing + total_post_vertical_spacing()))
&& i >= (4
- (hyphenation_flags & HYPHEN_FIRST_CHAR ? 1 : 0)
- (hyphenation_flags & HYPHEN_LAST_CHAR ? 1 : 0)
+ (hyphenation_flags & HYPHEN_NOT_FIRST_CHARS ? 1 : 0)
+ (hyphenation_flags & HYPHEN_NOT_LAST_CHARS ? 1 : 0)))
hyphenate(sl, hyphenation_flags);
while (forward != 0) {
node *tem1 = forward;
forward = forward->next;
tem1->next = 0;
tem = tem1->add_self(tem, &sl);
}
*startp = tem;
}
static node *node_list_reverse(node *n)
{
node *res = 0;
while (n) {
node *tem = n;
n = n->next;
tem->next = res;
res = tem;
}
return res;
}
static void distribute_space(node *n, int nspaces, hunits desired_space,
bool force_reverse_node_list = false)
{
if (desired_space.is_zero() || nspaces == 0)
return;
// Positive desired space is the typical case. Negative desired space
// is possible if we have overrun an unbreakable line. But we should
// not get here if there are no adjustable space nodes to adjust.
assert(nspaces > 0);
// Space cannot always be distributed evenly among all of the space
// nodes in the node list: there are limits to device resolution. We
// add space until we run out, which might happen before the end of
// the line. To achieve uniform typographical grayness and avoid
// rivers, we switch the end from which space is initially distributed
// with each line requiring it, unless compelled to reverse it. The
// node list's natural ordering is in the direction of text flow, so
// we distribute space initially from the left, unlike AT&T troff.
static bool do_reverse_node_list = false;
if (force_reverse_node_list || do_reverse_node_list)
n = node_list_reverse(n);
if (!force_reverse_node_list && spread_limit >= 0
&& desired_space.to_units() > 0) {
hunits em = curenv->get_size();
double Ems = (double)desired_space.to_units() / nspaces
/ (em.is_zero() ? hresolution : em.to_units());
if (Ems > spread_limit)
output_warning(WARN_BREAK, "spreading %1m per space", Ems);
}
for (node *tem = n; tem; tem = tem->next)
tem->spread_space(&nspaces, &desired_space);
if (force_reverse_node_list || do_reverse_node_list)
(void)node_list_reverse(n);
if (!force_reverse_node_list)
do_reverse_node_list = !do_reverse_node_list;
}
void environment::possibly_break_line(int start_here, int forced)
{
int was_centered = center_lines > 0;
if (!fill || current_tab || current_field || dummy)
return;
while (line != 0
&& (forced
// When a macro follows a paragraph in fill mode, the
// current line should not be empty.
|| (width_total - line->width()) > target_text_length)) {
hyphenate_line(start_here);
breakpoint *bp = choose_breakpoint();
if (bp == 0)
// we'll find one eventually
return;
node *pre, *post;
node **ndp = &line;
while (*ndp != bp->nd)
ndp = &(*ndp)->next;
bp->nd->split(bp->index, &pre, &post);
*ndp = post;
hunits extra_space_width = H0;
switch(adjust_mode) {
case ADJUST_BOTH:
if (bp->nspaces != 0)
extra_space_width = target_text_length - bp->width;
else if (bp->width > 0 && target_text_length > 0
&& target_text_length > bp->width)
output_warning(WARN_BREAK, "cannot adjust line");
break;
case ADJUST_CENTER:
saved_indent += (target_text_length - bp->width)/2;
was_centered = 1;
break;
case ADJUST_RIGHT:
saved_indent += target_text_length - bp->width;
break;
}
distribute_space(pre, bp->nspaces, extra_space_width);
hunits output_width = bp->width + extra_space_width;
// This should become an assert() when we can get reliable width
// data from CJK glyphs. See Savannah #44018.
if (output_width <= 0) {
double output_width_in_ems = output_width.to_units();
output_warning(WARN_BREAK, "line has non-positive width %1m",
output_width_in_ems);
return;
}
input_line_start -= output_width;
if (bp->hyphenated)
hyphen_line_count++;
else
hyphen_line_count = 0;
delete bp;
space_total = 0;
width_total = 0;
node *first_non_discardable = 0;
node *tem;
for (tem = line; tem != 0; tem = tem->next)
if (!tem->discardable())
first_non_discardable = tem;
node *to_be_discarded;
if (first_non_discardable) {
to_be_discarded = first_non_discardable->next;
first_non_discardable->next = 0;
for (tem = line; tem != 0; tem = tem->next) {
width_total += tem->width();
space_total += tem->nspaces();
}
discarding = 0;
}
else {
discarding = 1;
to_be_discarded = line;
line = 0;
}
// Do output_line() here so that line will be 0 iff the
// the environment will be empty.
output_line(pre, output_width, was_centered);
while (to_be_discarded != 0) {
tem = to_be_discarded;
to_be_discarded = to_be_discarded->next;
input_line_start -= tem->width();
delete tem;
}
if (line != 0) {
if (have_temporary_indent) {
saved_indent = temporary_indent;
have_temporary_indent = 0;
}
else
saved_indent = indent;
target_text_length = line_length - saved_indent;
}
}
}
/*
Do the break at the end of input after the end macro (if any).
Unix troff behaves as follows: if the last line is
foo bar\c
it will output foo on the current page, and bar on the next page;
if the last line is
foo\c
or
foo bar
everything will be output on the current page. This behaviour must be
considered a bug.
The problem is that some macro packages rely on this. For example,
the ATK macros have an end macro that emits \c if it needs to print a
table of contents but doesn't do a 'bp in the end macro; instead the
'bp is done in the bottom of page trap. This works with Unix troff,
provided that the current environment is not empty at the end of the
input file.
The following will make macro packages that do that sort of thing work
even if the current environment is empty at the end of the input file.
If the last input line used \c and this line occurred in the end macro,
then we'll force everything out on the current page, but we'll make
sure that the environment isn't empty so that we won't exit at the
bottom of this page.
*/
void environment::final_break()
{
if (prev_line_interrupted == 2) {
do_break();
add_node(new transparent_dummy_node);
}
else
do_break();
}
node *environment::make_tag(const char *nm, int i)
{
if (is_html) {
/*
* need to emit tag for post-grohtml
* but we check to see whether we can emit specials
*/
if (curdiv == topdiv && topdiv->before_first_page)
topdiv->begin_page();
macro m;
m.append_str("devtag:");
for (const char *p = nm; *p; p++)
if (!is_invalid_input_char((unsigned char)*p))
m.append(*p);
m.append(' ');
m.append_int(i);
return new special_node(m);
}
return 0;
}
void environment::dump_troff_state()
{
#define SPACES " "
fprintf(stderr, SPACES "register 'in' = %d\n", curenv->indent.to_units());
if (curenv->have_temporary_indent)
fprintf(stderr, SPACES "register 'ti' = %d\n",
curenv->temporary_indent.to_units());
fprintf(stderr, SPACES "centered lines 'ce' = %d\n", curenv->center_lines);
fprintf(stderr, SPACES "register 'll' = %d\n",
curenv->line_length.to_units());
fprintf(stderr, SPACES "fill 'fi=1/nf=0' = %d\n", curenv->fill);
fprintf(stderr, SPACES "page offset 'po' = %d\n",
topdiv->get_page_offset().to_units());
fprintf(stderr, SPACES "seen_break = %d\n", curenv->seen_break);
fprintf(stderr, SPACES "seen_space = %d\n", curenv->seen_space);
fflush(stderr);
#undef SPACES
}
statem *environment::construct_state(int only_eol)
{
if (is_html) {
statem *s = new statem();
if (!only_eol) {
s->add_tag(MTSM_IN, indent);
s->add_tag(MTSM_LL, line_length);
s->add_tag(MTSM_PO, topdiv->get_page_offset().to_units());
s->add_tag(MTSM_RJ, right_justify_lines);
if (have_temporary_indent)
s->add_tag(MTSM_TI, temporary_indent);
s->add_tag_ta();
if (seen_break)
s->add_tag(MTSM_BR);
if (seen_space != 0)
s->add_tag(MTSM_SP, seen_space);
seen_break = 0;
seen_space = 0;
}
if (seen_eol) {
s->add_tag(MTSM_EOL);
s->add_tag(MTSM_CE, center_lines);
}
seen_eol = 0;
return s;
}
else
return NULL;
}
void environment::construct_format_state(node *n, int was_centered,
int filling)
{
if (is_html) {
// find first glyph node which has a state.
while (n != 0 && n->state == 0)
n = n->next;
if (n == 0 || (n->state == 0))
return;
if (seen_space != 0)
n->state->add_tag(MTSM_SP, seen_space);
if (seen_eol && topdiv == curdiv)
n->state->add_tag(MTSM_EOL);
seen_space = 0;
seen_eol = 0;
if (was_centered)
n->state->add_tag(MTSM_CE, center_lines+1);
else
n->state->add_tag_if_unknown(MTSM_CE, 0);
n->state->add_tag_if_unknown(MTSM_FI, filling);
n = n->next;
while (n != 0) {
if (n->state != 0) {
n->state->sub_tag_ce();
n->state->add_tag_if_unknown(MTSM_FI, filling);
}
n = n->next;
}
}
}
void environment::construct_new_line_state(node *n)
{
if (is_html) {
// find first glyph node which has a state.
while (n != 0 && n->state == 0)
n = n->next;
if (n == 0 || n->state == 0)
return;
if (seen_space != 0)
n->state->add_tag(MTSM_SP, seen_space);
if (seen_eol && topdiv == curdiv)
n->state->add_tag(MTSM_EOL);
seen_space = 0;
seen_eol = 0;
}
}
extern int global_diverted_space;
void environment::do_break(int do_spread)
{
int was_centered = 0;
if (curdiv == topdiv && topdiv->before_first_page) {
topdiv->begin_page();
return;
}
if (current_tab)
wrap_up_tab();
if (line) {
// this is so that hyphenation works
if (line->nspaces() == 0) {
line = new space_node(H0, get_fill_color(), line);
space_total++;
}
possibly_break_line(0, do_spread);
}
while (line != 0 && line->discardable()) {
width_total -= line->width();
space_total -= line->nspaces();
node *tem = line;
line = line->next;
delete tem;
}
discarding = 0;
input_line_start = H0;
if (line != 0) {
if (fill) {
switch (adjust_mode) {
case ADJUST_CENTER:
saved_indent += (target_text_length - width_total)/2;
was_centered = 1;
break;
case ADJUST_RIGHT:
saved_indent += target_text_length - width_total;
break;
}
}
node *tem = line;
line = 0;
output_line(tem, width_total, was_centered);
hyphen_line_count = 0;
}
prev_line_interrupted = 0;
#ifdef WIDOW_CONTROL
mark_last_line();
output_pending_lines();
#endif /* WIDOW_CONTROL */
if (!global_diverted_space) {
curdiv->modified_tag.incl(MTSM_BR);
seen_break = 1;
}
}
int environment::is_empty()
{
return !current_tab && line == 0 && pending_lines == 0;
}
void do_break_request(int spread)
{
while (!tok.is_newline() && !tok.is_eof())
tok.next();
if (break_flag)
curenv->do_break(spread);
tok.next();
}
void break_request()
{
do_break_request(0);
}
void break_spread_request()
{
do_break_request(1);
}
void title()
{
if (curdiv == topdiv && topdiv->before_first_page) {
handle_initial_title();
return;
}
node *part[3];
hunits part_width[3];
part[0] = part[1] = part[2] = 0;
environment env(curenv);
environment *oldenv = curenv;
curenv = &env;
read_title_parts(part, part_width);
curenv = oldenv;
curenv->size = env.size;
curenv->prev_size = env.prev_size;
curenv->requested_size = env.requested_size;
curenv->prev_requested_size = env.prev_requested_size;
curenv->char_height = env.char_height;
curenv->char_slant = env.char_slant;
curenv->fontno = env.fontno;
curenv->prev_fontno = env.prev_fontno;
curenv->glyph_color = env.glyph_color;
curenv->prev_glyph_color = env.prev_glyph_color;
curenv->fill_color = env.fill_color;
curenv->prev_fill_color = env.prev_fill_color;
node *n = 0;
node *p = part[2];
while (p != 0) {
node *tem = p;
p = p->next;
tem->next = n;
n = tem;
}
hunits length_title(curenv->title_length);
hunits f = length_title - part_width[1];
hunits f2 = f/2;
n = new hmotion_node(f2 - part_width[2], curenv->get_fill_color(), n);
p = part[1];
while (p != 0) {
node *tem = p;
p = p->next;
tem->next = n;
n = tem;
}
n = new hmotion_node(f - f2 - part_width[0], curenv->get_fill_color(), n);
p = part[0];
while (p != 0) {
node *tem = p;
p = p->next;
tem->next = n;
n = tem;
}
curenv->output_title(n, !curenv->fill, curenv->vertical_spacing,
curenv->total_post_vertical_spacing(), length_title);
curenv->hyphen_line_count = 0;
tok.next();
}
void adjust()
{
curenv->adjust_mode |= 1;
if (has_arg()) {
switch (tok.ch()) {
case 'l':
curenv->adjust_mode = ADJUST_LEFT;
break;
case 'r':
curenv->adjust_mode = ADJUST_RIGHT;
break;
case 'c':
curenv->adjust_mode = ADJUST_CENTER;
break;
case 'b':
case 'n':
curenv->adjust_mode = ADJUST_BOTH;
break;
default:
int n;
if (get_integer(&n)) {
if (n < 0)
warning(WARN_RANGE, "negative adjustment mode");
else if (n > ADJUST_MAX)
warning(WARN_RANGE, "out-of-range adjustment mode ignored: "
"%1", n);
else
curenv->adjust_mode = n;
}
}
}
skip_line();
}
void no_adjust()
{
curenv->adjust_mode &= ~1;
skip_line();
}
void do_input_trap(int continued)
{
curenv->input_trap_count = 0;
if (continued)
curenv->continued_input_trap = 1;
else
curenv->continued_input_trap = 0;
int n;
if (has_arg() && get_integer(&n)) {
if (n <= 0)
warning(WARN_RANGE,
"number of lines for input trap must be greater than zero");
else {
symbol s = get_name(true /* required */);
if (!s.is_null()) {
curenv->input_trap_count = n;
curenv->input_trap = s;
}
}
}
skip_line();
}
void input_trap()
{
do_input_trap(0);
}
void input_trap_continued()
{
do_input_trap(1);
}
/* tabs */
// must not be R or C or L or a legitimate part of a number expression
const char TAB_REPEAT_CHAR = 'T';
struct tab {
tab *next;
hunits pos;
tab_type type;
tab(hunits, tab_type);
enum { BLOCK = 1024 };
};
tab::tab(hunits x, tab_type t) : next(0), pos(x), type(t)
{
}
tab_stops::tab_stops(hunits distance, tab_type type)
: initial_list(0)
{
repeated_list = new tab(distance, type);
}
tab_stops::~tab_stops()
{
clear();
}
tab_type tab_stops::distance_to_next_tab(hunits curpos, hunits *distance)
{
hunits nextpos;
return distance_to_next_tab(curpos, distance, &nextpos);
}
tab_type tab_stops::distance_to_next_tab(hunits curpos,
hunits *distance,
hunits *nextpos)
{
hunits lastpos = 0;
tab *tem;
for (tem = initial_list; tem && tem->pos <= curpos; tem = tem->next)
lastpos = tem->pos;
if (tem) {
*distance = tem->pos - curpos;
*nextpos = tem->pos;
return tem->type;
}
if (repeated_list == 0)
return TAB_NONE;
hunits base = lastpos;
for (;;) {
for (tem = repeated_list; tem && tem->pos + base <= curpos;
tem = tem->next)
lastpos = tem->pos;
if (tem) {
*distance = tem->pos + base - curpos;
*nextpos = tem->pos + base;
return tem->type;
}
if (lastpos < 0)
lastpos = 0;
base += lastpos;
}
return TAB_NONE;
}
const char *tab_stops::to_string()
{
static char *buf = 0;
static int buf_size = 0;
// figure out a maximum on the amount of space we can need
int count = 0;
tab *p;
for (p = initial_list; p; p = p->next)
++count;
for (p = repeated_list; p; p = p->next)
++count;
// (10 for digits + 1 for u + 1 for 'C' or 'R') + 2 for ' &' + 1 for '\0'
int need = count*12 + 3;
if (buf == 0 || need > buf_size) {
if (buf)
delete[] buf;
buf_size = need;
buf = new char[buf_size];
}
char *ptr = buf;
for (p = initial_list; p; p = p->next) {
strcpy(ptr, i_to_a(p->pos.to_units()));
ptr = strchr(ptr, '\0');
*ptr++ = 'u';
*ptr = '\0';
switch (p->type) {
case TAB_LEFT:
break;
case TAB_RIGHT:
*ptr++ = 'R';
break;
case TAB_CENTER:
*ptr++ = 'C';
break;
case TAB_NONE:
default:
assert(0);
}
}
if (repeated_list)
*ptr++ = TAB_REPEAT_CHAR;
for (p = repeated_list; p; p = p->next) {
strcpy(ptr, i_to_a(p->pos.to_units()));
ptr = strchr(ptr, '\0');
*ptr++ = 'u';
*ptr = '\0';
switch (p->type) {
case TAB_LEFT:
break;
case TAB_RIGHT:
*ptr++ = 'R';
break;
case TAB_CENTER:
*ptr++ = 'C';
break;
case TAB_NONE:
default:
assert(0);
}
}
*ptr++ = '\0';
return buf;
}
tab_stops::tab_stops() : initial_list(0), repeated_list(0)
{
}
tab_stops::tab_stops(const tab_stops &ts)
: initial_list(0), repeated_list(0)
{
tab **p = &initial_list;
tab *t = ts.initial_list;
while (t) {
*p = new tab(t->pos, t->type);
t = t->next;
p = &(*p)->next;
}
p = &repeated_list;
t = ts.repeated_list;
while (t) {
*p = new tab(t->pos, t->type);
t = t->next;
p = &(*p)->next;
}
}
void tab_stops::clear()
{
while (initial_list) {
tab *tem = initial_list;
initial_list = initial_list->next;
delete tem;
}
while (repeated_list) {
tab *tem = repeated_list;
repeated_list = repeated_list->next;
delete tem;
}
}
void tab_stops::add_tab(hunits pos, tab_type type, int repeated)
{
tab **p;
for (p = repeated ? &repeated_list : &initial_list; *p; p = &(*p)->next)
;
*p = new tab(pos, type);
}
void tab_stops::operator=(const tab_stops &ts)
{
clear();
tab **p = &initial_list;
tab *t = ts.initial_list;
while (t) {
*p = new tab(t->pos, t->type);
t = t->next;
p = &(*p)->next;
}
p = &repeated_list;
t = ts.repeated_list;
while (t) {
*p = new tab(t->pos, t->type);
t = t->next;
p = &(*p)->next;
}
}
void set_tabs()
{
hunits pos;
hunits prev_pos = 0;
bool is_first_stop = true;
bool is_repeating_stop = false;
tab_stops tabs;
while (has_arg()) {
if (tok.ch() == TAB_REPEAT_CHAR) {
tok.next();
is_repeating_stop = true;
prev_pos = 0;
}
if (!get_hunits(&pos, 'm', prev_pos))
break;
tab_type type = TAB_LEFT;
if (tok.ch() == 'C') {
tok.next();
type = TAB_CENTER;
}
else if (tok.ch() == 'R') {
tok.next();
type = TAB_RIGHT;
}
else if (tok.ch() == 'L') {
tok.next();
}
if (pos <= prev_pos && ((!is_first_stop) || is_repeating_stop))
warning(WARN_RANGE,
"positions of tab stops must be strictly increasing");
else {
tabs.add_tab(pos, type, is_repeating_stop);
prev_pos = pos;
is_first_stop = false;
}
}
curenv->tabs = tabs;
curdiv->modified_tag.incl(MTSM_TA);
skip_line();
}
const char *environment::get_tabs()
{
return tabs.to_string();
}
tab_type environment::distance_to_next_tab(hunits *distance)
{
return line_tabs
? curenv->tabs.distance_to_next_tab(get_text_length(), distance)
: curenv->tabs.distance_to_next_tab(get_input_line_position(), distance);
}
tab_type environment::distance_to_next_tab(hunits *distance, hunits *leftpos)
{
return line_tabs
? curenv->tabs.distance_to_next_tab(get_text_length(), distance, leftpos)
: curenv->tabs.distance_to_next_tab(get_input_line_position(), distance,
leftpos);
}
void field_characters()
{
field_delimiter_char = get_optional_char();
if (field_delimiter_char)
padding_indicator_char = get_optional_char();
else
padding_indicator_char = 0;
skip_line();
}
void line_tabs_request()
{
int n;
if (has_arg() && get_integer(&n))
curenv->line_tabs = n != 0;
else
curenv->line_tabs = 1;
skip_line();
}
int environment::get_line_tabs()
{
return line_tabs;
}
void environment::wrap_up_tab()
{
if (!current_tab)
return;
if (line == 0)
start_line();
hunits tab_amount;
switch (current_tab) {
case TAB_RIGHT:
tab_amount = tab_distance - tab_width;
line = make_tab_node(tab_amount, line);
break;
case TAB_CENTER:
tab_amount = tab_distance - tab_width/2;
line = make_tab_node(tab_amount, line);
break;
case TAB_NONE:
case TAB_LEFT:
default:
assert(0);
}
width_total += tab_amount;
width_total += tab_width;
if (current_field) {
if (tab_precedes_field) {
pre_field_width += tab_amount;
tab_precedes_field = 0;
}
field_distance -= tab_amount;
field_spaces += tab_field_spaces;
}
if (tab_contents != 0) {
node *tem;
for (tem = tab_contents; tem->next != 0; tem = tem->next)
;
tem->next = line;
line = tab_contents;
}
tab_field_spaces = 0;
tab_contents = 0;
tab_width = H0;
tab_distance = H0;
current_tab = TAB_NONE;
}
node *environment::make_tab_node(hunits d, node *next)
{
if (leader_node != 0 && d < 0) {
error("motion generated by leader cannot be negative");
delete leader_node;
leader_node = 0;
}
if (!leader_node)
return new hmotion_node(d, 1, 0, get_fill_color(), next);
node *n = new hline_node(d, leader_node, next);
leader_node = 0;
return n;
}
void environment::handle_tab(int is_leader)
{
hunits d;
hunits absolute;
if (current_tab)
wrap_up_tab();
charinfo *ci = is_leader ? leader_char : tab_char;
delete leader_node;
leader_node = ci ? make_char_node(ci) : 0;
tab_type t = distance_to_next_tab(&d, &absolute);
switch (t) {
case TAB_NONE:
return;
case TAB_LEFT:
add_node(make_tag("tab L", absolute.to_units()));
add_node(make_tab_node(d));
return;
case TAB_RIGHT:
add_node(make_tag("tab R", absolute.to_units()));
break;
case TAB_CENTER:
add_node(make_tag("tab C", absolute.to_units()));
break;
default:
assert(0);
}
tab_width = 0;
tab_distance = d;
tab_contents = 0;
current_tab = t;
tab_field_spaces = 0;
}
void environment::start_field()
{
assert(!current_field);
hunits d;
if (distance_to_next_tab(&d) != TAB_NONE) {
pre_field_width = get_text_length();
field_distance = d;
current_field = 1;
field_spaces = 0;
tab_field_spaces = 0;
for (node *p = line; p; p = p->next)
if (p->nspaces()) {
p->freeze_space();
space_total--;
}
tab_precedes_field = current_tab != TAB_NONE;
}
else
error("zero field width");
}
void environment::wrap_up_field()
{
if (!current_tab && field_spaces == 0)
add_padding();
hunits padding = field_distance - (get_text_length() - pre_field_width);
if (current_tab && tab_field_spaces != 0) {
hunits tab_padding = scale(padding,
tab_field_spaces,
field_spaces + tab_field_spaces);
padding -= tab_padding;
distribute_space(tab_contents, tab_field_spaces, tab_padding,
true /* force reversal of node list */);
tab_field_spaces = 0;
tab_width += tab_padding;
}
if (field_spaces != 0) {
distribute_space(line, field_spaces, padding,
true /* force reversal of node list */);
width_total += padding;
if (current_tab) {
// the start of the tab has been moved to the right by padding, so
tab_distance -= padding;
if (tab_distance <= H0) {
// use the next tab stop instead
current_tab = tabs.distance_to_next_tab(get_input_line_position()
- tab_width,
&tab_distance);
if (current_tab == TAB_NONE || current_tab == TAB_LEFT) {
width_total += tab_width;
if (current_tab == TAB_LEFT) {
line = make_tab_node(tab_distance, line);
width_total += tab_distance;
current_tab = TAB_NONE;
}
if (tab_contents != 0) {
node *tem;
for (tem = tab_contents; tem->next != 0; tem = tem->next)
;
tem->next = line;
line = tab_contents;
tab_contents = 0;
}
tab_width = H0;
tab_distance = H0;
}
}
}
}
current_field = 0;
}
void environment::add_padding()
{
if (current_tab) {
tab_contents = new space_node(H0, get_fill_color(), tab_contents);
tab_field_spaces++;
}
else {
if (line == 0)
start_line();
line = new space_node(H0, get_fill_color(), line);
field_spaces++;
}
}
typedef int (environment::*INT_FUNCP)();
typedef vunits (environment::*VUNITS_FUNCP)();
typedef hunits (environment::*HUNITS_FUNCP)();
typedef const char *(environment::*STRING_FUNCP)();
class int_env_reg : public reg {
INT_FUNCP func;
public:
int_env_reg(INT_FUNCP);
const char *get_string();
bool get_value(units *val);
};
class vunits_env_reg : public reg {
VUNITS_FUNCP func;
public:
vunits_env_reg(VUNITS_FUNCP f);
const char *get_string();
bool get_value(units *val);
};
class hunits_env_reg : public reg {
HUNITS_FUNCP func;
public:
hunits_env_reg(HUNITS_FUNCP f);
const char *get_string();
bool get_value(units *val);
};
class string_env_reg : public reg {
STRING_FUNCP func;
public:
string_env_reg(STRING_FUNCP);
const char *get_string();
};
int_env_reg::int_env_reg(INT_FUNCP f) : func(f)
{
}
bool int_env_reg::get_value(units *val)
{
*val = (curenv->*func)();
return true;
}
const char *int_env_reg::get_string()
{
return i_to_a((curenv->*func)());
}
vunits_env_reg::vunits_env_reg(VUNITS_FUNCP f) : func(f)
{
}
bool vunits_env_reg::get_value(units *val)
{
*val = (curenv->*func)().to_units();
return true;
}
const char *vunits_env_reg::get_string()
{
return i_to_a((curenv->*func)().to_units());
}
hunits_env_reg::hunits_env_reg(HUNITS_FUNCP f) : func(f)
{
}
bool hunits_env_reg::get_value(units *val)
{
*val = (curenv->*func)().to_units();
return true;
}
const char *hunits_env_reg::get_string()
{
return i_to_a((curenv->*func)().to_units());
}
string_env_reg::string_env_reg(STRING_FUNCP f) : func(f)
{
}
const char *string_env_reg::get_string()
{
return (curenv->*func)();
}
class horizontal_place_reg : public general_reg {
public:
horizontal_place_reg();
bool get_value(units *);
void set_value(units);
};
horizontal_place_reg::horizontal_place_reg()
{
}
bool horizontal_place_reg::get_value(units *res)
{
*res = curenv->get_input_line_position().to_units();
return true;
}
void horizontal_place_reg::set_value(units n)
{
curenv->set_input_line_position(hunits(n));
}
int environment::get_zoom()
{
return env_get_zoom(this);
}
int environment::get_numbering_nodes()
{
return (curenv->numbering_nodes ? 1 : 0);
}
const char *environment::get_font_family_string()
{
return family->nm.contents();
}
const char *environment::get_glyph_color_string()
{
return glyph_color->nm.contents();
}
const char *environment::get_fill_color_string()
{
return fill_color->nm.contents();
}
const char *environment::get_font_name_string()
{
symbol f = get_font_name(fontno, this);
return f.contents();
}
const char *environment::get_style_name_string()
{
symbol f = get_style_name(fontno);
return f.contents();
}
const char *environment::get_name_string()
{
return name.contents();
}
// Convert a quantity in scaled points to ascii decimal fraction.
const char *sptoa(int sp)
{
assert(sp > 0);
assert(sizescale > 0);
if (sizescale == 1)
return i_to_a(sp);
if (sp % sizescale == 0)
return i_to_a(sp/sizescale);
// See if 1/sizescale is exactly representable as a decimal fraction,
// ie its only prime factors are 2 and 5.
int n = sizescale;
int power2 = 0;
while ((n & 1) == 0) {
n >>= 1;
power2++;
}
int power5 = 0;
while ((n % 5) == 0) {
n /= 5;
power5++;
}
if (n == 1) {
int decimal_point = power5 > power2 ? power5 : power2;
if (decimal_point <= 10) {
int factor = 1;
int t;
for (t = decimal_point - power2; --t >= 0;)
factor *= 2;
for (t = decimal_point - power5; --t >= 0;)
factor *= 5;
if (factor == 1 || sp <= INT_MAX/factor)
return if_to_a(sp*factor, decimal_point);
}
}
double s = double(sp)/double(sizescale);
double factor = 10.0;
double val = s;
int decimal_point = 0;
do {
double v = ceil(s*factor);
if (v > INT_MAX)
break;
val = v;
factor *= 10.0;
} while (++decimal_point < 10);
return if_to_a(int(val), decimal_point);
}
const char *environment::get_point_size_string()
{
return sptoa(curenv->get_point_size());
}
const char *environment::get_requested_point_size_string()
{
return sptoa(curenv->get_requested_point_size());
}
void environment::print_env()
{
// at the time of calling .pev, those values are always zero or
// meaningless:
//
// char_height, char_slant,
// interrupted
// current_tab, tab_width, tab_distance
// current_field, field_distance, pre_field_width, field_spaces,
// tab_field_spaces, tab_precedes_field
// composite
//
errprint(" previous line length: %1u\n", prev_line_length.to_units());
errprint(" line length: %1u\n", line_length.to_units());
errprint(" previous title length: %1u\n", prev_title_length.to_units());
errprint(" title length: %1u\n", title_length.to_units());
errprint(" previous size: %1p (%2s)\n",
prev_size.to_points(), prev_size.to_scaled_points());
errprint(" size: %1p (%2s)\n",
size.to_points(), size.to_scaled_points());
errprint(" previous requested size: %1s\n", prev_requested_size);
errprint(" requested size: %1s\n", requested_size);
errprint(" previous font number: %1\n", prev_fontno);
errprint(" font number: %1\n", fontno);
errprint(" previous family: '%1'\n", prev_family->nm.contents());
errprint(" family: '%1'\n", family->nm.contents());
errprint(" space size: %1/36 em\n", space_size);
errprint(" sentence space size: %1/36 em\n", sentence_space_size);
errprint(" previous line interrupted: %1\n",
prev_line_interrupted ? "yes" : "no");
errprint(" fill mode: %1\n", fill ? "on" : "off");
errprint(" adjust mode: %1\n",
adjust_mode == ADJUST_LEFT
? "left"
: adjust_mode == ADJUST_BOTH
? "both"
: adjust_mode == ADJUST_CENTER
? "center"
: "right");
if (center_lines)
errprint(" lines to center: %1\n", center_lines);
if (right_justify_lines)
errprint(" lines to right justify: %1\n", right_justify_lines);
errprint(" previous vertical spacing: %1u\n",
prev_vertical_spacing.to_units());
errprint(" vertical spacing: %1u\n", vertical_spacing.to_units());
errprint(" previous post-vertical spacing: %1u\n",
prev_post_vertical_spacing.to_units());
errprint(" post-vertical spacing: %1u\n",
post_vertical_spacing.to_units());
errprint(" previous line spacing: %1\n", prev_line_spacing);
errprint(" line spacing: %1\n", line_spacing);
errprint(" previous indentation: %1u\n", prev_indent.to_units());
errprint(" indentation: %1u\n", indent.to_units());
errprint(" temporary indentation: %1u\n", temporary_indent.to_units());
errprint(" have temporary indentation: %1\n",
have_temporary_indent ? "yes" : "no");
errprint(" currently used indentation: %1u\n", saved_indent.to_units());
errprint(" target text length: %1u\n", target_text_length.to_units());
if (underline_lines) {
errprint(" lines to underline: %1\n", underline_lines);
errprint(" font number before underlining: %1\n", pre_underline_fontno);
errprint(" underline spaces: %1\n", underline_spaces ? "yes" : "no");
}
if (input_trap.contents()) {
errprint(" input trap macro: '%1'\n", input_trap.contents());
errprint(" input trap line counter: %1\n", input_trap_count);
errprint(" continued input trap: %1\n",
continued_input_trap ? "yes" : "no");
}
errprint(" previous text length: %1u\n", prev_text_length.to_units());
errprint(" total width: %1u\n", width_total.to_units());
errprint(" total number of spaces: %1\n", space_total);
errprint(" input line start: %1u\n", input_line_start.to_units());
errprint(" line tabs: %1\n", line_tabs ? "yes" : "no");
errprint(" discarding: %1\n", discarding ? "yes" : "no");
errprint(" spread flag set: %1\n", spread_flag ? "yes" : "no"); // \p
if (margin_character_node) {
errprint(" margin character flags: %1\n",
margin_character_flags == MARGIN_CHARACTER_ON
? "on"
: margin_character_flags == MARGIN_CHARACTER_NEXT
? "next"
: margin_character_flags == (MARGIN_CHARACTER_ON
| MARGIN_CHARACTER_NEXT)
? "on, next"
: "none");
errprint(" margin character distance: %1u\n",
margin_character_distance.to_units());
}
if (numbering_nodes) {
errprint(" line number digit width: %1u\n",
line_number_digit_width.to_units());
errprint(" separation between number and text: %1 digit spaces\n",
number_text_separation);
errprint(" line number indentation: %1 digit spaces\n",
line_number_indent);
errprint(" print line numbers every %1line%1\n",
line_number_multiple > 1 ? i_to_a(line_number_multiple) : "",
line_number_multiple > 1 ? "s" : "");
errprint(" lines not to enumerate: %1\n", no_number_count);
}
string hf = hyphenation_flags ? "on" : "off";
if (hyphenation_flags & HYPHEN_NOT_LAST_LINE)
hf += ", not last line";
if (hyphenation_flags & HYPHEN_LAST_CHAR)
hf += ", last char";
if (hyphenation_flags & HYPHEN_NOT_LAST_CHARS)
hf += ", not last two chars";
if (hyphenation_flags & HYPHEN_FIRST_CHAR)
hf += ", first char";
if (hyphenation_flags & HYPHEN_NOT_FIRST_CHARS)
hf += ", not first two chars";
hf += '\0';
errprint(" hyphenation_flags: %1\n", hf.contents());
errprint(" number of consecutive hyphenated lines: %1\n",
hyphen_line_count);
errprint(" maximum number of consecutive hyphenated lines: %1\n",
hyphen_line_max);
errprint(" hyphenation space: %1u\n", hyphenation_space.to_units());
errprint(" hyphenation margin: %1u\n", hyphenation_margin.to_units());
#ifdef WIDOW_CONTROL
errprint(" widow control: %1\n", widow_control ? "yes" : "no");
#endif /* WIDOW_CONTROL */
}
void print_env()
{
errprint("Current Environment:\n");
curenv->print_env();
dictionary_iterator iter(env_dictionary);
symbol s;
environment *e;
while (iter.get(&s, (void **)&e)) {
assert(!s.is_null());
errprint("Environment %1:\n", s.contents());
if (e != curenv)
e->print_env();
else
errprint(" current\n");
}
fflush(stderr);
skip_line();
}
#define init_int_env_reg(name, func) \
register_dictionary.define(name, new int_env_reg(&environment::func))
#define init_vunits_env_reg(name, func) \
register_dictionary.define(name, new vunits_env_reg(&environment::func))
#define init_hunits_env_reg(name, func) \
register_dictionary.define(name, new hunits_env_reg(&environment::func))
#define init_string_env_reg(name, func) \
register_dictionary.define(name, new string_env_reg(&environment::func))
void init_env_requests()
{
init_request("ad", adjust);
init_request("br", break_request);
init_request("brp", break_spread_request);
init_request("c2", no_break_control_char);
init_request("cc", control_char);
init_request("ce", center);
init_request("cu", continuous_underline);
init_request("ev", environment_switch);
init_request("evc", environment_copy);
init_request("fam", family_change);
init_request("fc", field_characters);
init_request("fi", fill);
init_request("fcolor", fill_color_change);
init_request("ft", font_change);
init_request("gcolor", glyph_color_change);
init_request("hc", hyphen_char);
init_request("hlm", hyphen_line_max_request);
init_request("hy", hyphenate_request);
init_request("hym", hyphenation_margin_request);
init_request("hys", hyphenation_space_request);
init_request("in", indent);
init_request("it", input_trap);
init_request("itc", input_trap_continued);
init_request("lc", leader_character);
init_request("linetabs", line_tabs_request);
init_request("ll", line_length);
init_request("ls", line_spacing);
init_request("lt", title_length);
init_request("mc", margin_character);
init_request("na", no_adjust);
init_request("nf", no_fill);
init_request("nh", no_hyphenate);
init_request("nm", number_lines);
init_request("nn", no_number);
init_request("pev", print_env);
init_request("ps", point_size);
init_request("pvs", post_vertical_spacing);
init_request("rj", right_justify);
init_request("sizes", override_sizes);
init_request("ss", space_size);
init_request("ta", set_tabs);
init_request("ti", temporary_indent);
init_request("tc", tab_character);
init_request("tl", title);
init_request("ul", underline);
init_request("vs", vertical_spacing);
#ifdef WIDOW_CONTROL
init_request("wdc", widow_control_request);
#endif /* WIDOW_CONTROL */
init_int_env_reg(".b", get_bold);
init_vunits_env_reg(".cdp", get_prev_char_depth);
init_int_env_reg(".ce", get_center_lines);
init_vunits_env_reg(".cht", get_prev_char_height);
init_hunits_env_reg(".csk", get_prev_char_skew);
init_string_env_reg(".ev", get_name_string);
init_int_env_reg(".f", get_font);
init_string_env_reg(".fam", get_font_family_string);
init_string_env_reg(".fn", get_font_name_string);
init_int_env_reg(".height", get_char_height);
init_int_env_reg(".hlc", get_hyphen_line_count);
init_int_env_reg(".hlm", get_hyphen_line_max);
init_int_env_reg(".hy", get_hyphenation_flags);
init_hunits_env_reg(".hym", get_hyphenation_margin);
init_hunits_env_reg(".hys", get_hyphenation_space);
init_hunits_env_reg(".i", get_indent);
init_hunits_env_reg(".in", get_saved_indent);
init_int_env_reg(".int", get_prev_line_interrupted);
init_int_env_reg(".linetabs", get_line_tabs);
init_hunits_env_reg(".lt", get_title_length);
init_int_env_reg(".j", get_adjust_mode);
init_hunits_env_reg(".k", get_text_length);
init_int_env_reg(".L", get_line_spacing);
init_hunits_env_reg(".l", get_line_length);
init_hunits_env_reg(".ll", get_saved_line_length);
init_string_env_reg(".M", get_fill_color_string);
init_string_env_reg(".m", get_glyph_color_string);
init_hunits_env_reg(".n", get_prev_text_length);
init_int_env_reg(".nm", get_numbering_nodes);
init_int_env_reg(".nn", get_no_number_count);
init_int_env_reg(".ps", get_point_size);
init_int_env_reg(".psr", get_requested_point_size);
init_vunits_env_reg(".pvs", get_post_vertical_spacing);
init_int_env_reg(".rj", get_right_justify_lines);
init_string_env_reg(".s", get_point_size_string);
init_int_env_reg(".slant", get_char_slant);
init_int_env_reg(".ss", get_space_size);
init_int_env_reg(".sss", get_sentence_space_size);
init_string_env_reg(".sr", get_requested_point_size_string);
init_string_env_reg(".sty", get_style_name_string);
init_string_env_reg(".tabs", get_tabs);
init_int_env_reg(".u", get_fill);
init_vunits_env_reg(".v", get_vertical_spacing);
init_hunits_env_reg(".w", get_prev_char_width);
init_int_env_reg(".zoom", get_zoom);
register_dictionary.define("ct", new variable_reg(&ct_reg_contents));
register_dictionary.define("hp", new horizontal_place_reg);
register_dictionary.define("ln", new variable_reg(&next_line_number));
register_dictionary.define("rsb", new variable_reg(&rsb_reg_contents));
register_dictionary.define("rst", new variable_reg(&rst_reg_contents));
register_dictionary.define("sb", new variable_reg(&sb_reg_contents));
register_dictionary.define("skw", new variable_reg(&skw_reg_contents));
register_dictionary.define("ssc", new variable_reg(&ssc_reg_contents));
register_dictionary.define("st", new variable_reg(&st_reg_contents));
}
// Hyphenation - TeX's hyphenation algorithm with a less fancy implementation.
struct trie_node;
class trie {
trie_node *tp;
virtual void do_match(int len, void *val) = 0;
virtual void do_delete(void *) = 0;
void delete_trie_node(trie_node *);
public:
trie() : tp(0) {}
virtual ~trie(); // virtual to shut up g++
void insert(const char *, int, void *);
// find calls do_match for each match it finds
void find(const char *pat, int patlen);
void clear();
};
class hyphen_trie : private trie {
int *h;
void do_match(int i, void *v);
void do_delete(void *v);
void insert_pattern(const char *pat, int patlen, int *num);
void insert_hyphenation(dictionary *ex, const char *pat, int patlen);
int hpf_getc(FILE *f);
public:
hyphen_trie() {}
~hyphen_trie() {}
void hyphenate(const char *word, int len, int *hyphens);
void read_patterns_file(const char *name, int append, dictionary *ex);
};
struct hyphenation_language {
symbol name;
dictionary exceptions;
hyphen_trie patterns;
hyphenation_language(symbol nm) : name(nm), exceptions(501) {}
~hyphenation_language() { }
};
dictionary language_dictionary(5);
hyphenation_language *current_language = 0;
static void set_hyphenation_language()
{
symbol nm = get_name(true /* required */);
if (!nm.is_null()) {
current_language = (hyphenation_language *)language_dictionary.lookup(nm);
if (!current_language) {
current_language = new hyphenation_language(nm);
(void)language_dictionary.lookup(nm, (void *)current_language);
}
}
skip_line();
}
const int WORD_MAX = 256; // we use unsigned char for offsets in
// hyphenation exceptions
static void hyphen_word()
{
if (!current_language) {
error("no current hyphenation language");
skip_line();
return;
}
char buf[WORD_MAX + 1];
unsigned char pos[WORD_MAX + 2];
for (;;) {
tok.skip();
if (tok.is_newline() || tok.is_eof())
break;
int i = 0;
int npos = 0;
while (i < WORD_MAX && !tok.is_space() && !tok.is_newline()
&& !tok.is_eof()) {
charinfo *ci = tok.get_char(true /* required */);
if (ci == 0) {
skip_line();
return;
}
tok.next();
if (ci->get_ascii_code() == '-') {
if (i > 0 && (npos == 0 || pos[npos - 1] != i))
pos[npos++] = i;
}
else {
unsigned char c = ci->get_hyphenation_code();
if (c == 0)
break;
buf[i++] = c;
}
}
if (i > 0) {
pos[npos] = 0;
buf[i] = 0;
unsigned char *tem = new unsigned char[npos + 1];
memcpy(tem, pos, npos + 1);
tem = (unsigned char *)current_language->exceptions.lookup(symbol(buf),
tem);
if (tem)
delete[] tem;
}
}
skip_line();
}
struct trie_node {
char c;
trie_node *down;
trie_node *right;
void *val;
trie_node(char, trie_node *);
};
trie_node::trie_node(char ch, trie_node *p)
: c(ch), down(0), right(p), val(0)
{
}
trie::~trie()
{
clear();
}
void trie::clear()
{
delete_trie_node(tp);
tp = 0;
}
void trie::delete_trie_node(trie_node *p)
{
if (p) {
delete_trie_node(p->down);
delete_trie_node(p->right);
if (p->val)
do_delete(p->val);
delete p;
}
}
void trie::insert(const char *pat, int patlen, void *val)
{
trie_node **p = &tp;
assert(patlen > 0 && pat != 0);
for (;;) {
while (*p != 0 && (*p)->c < pat[0])
p = &((*p)->right);
if (*p == 0 || (*p)->c != pat[0])
*p = new trie_node(pat[0], *p);
if (--patlen == 0) {
(*p)->val = val;
break;
}
++pat;
p = &((*p)->down);
}
}
void trie::find(const char *pat, int patlen)
{
trie_node *p = tp;
for (int i = 0; p != 0 && i < patlen; i++) {
while (p != 0 && p->c < pat[i])
p = p->right;
if (p != 0 && p->c == pat[i]) {
if (p->val != 0)
do_match(i+1, p->val);
p = p->down;
}
else
break;
}
}
struct operation {
operation *next;
short distance;
short num;
operation(int, int, operation *);
};
operation::operation(int i, int j, operation *op)
: next(op), distance(j), num(i)
{
}
void hyphen_trie::insert_pattern(const char *pat, int patlen, int *num)
{
operation *op = 0;
for (int i = 0; i < patlen+1; i++)
if (num[i] != 0)
op = new operation(num[i], patlen - i, op);
insert(pat, patlen, op);
}
void hyphen_trie::insert_hyphenation(dictionary *ex, const char *pat,
int patlen)
{
char buf[WORD_MAX + 2];
unsigned char pos[WORD_MAX + 2];
int i = 0, j = 0;
int npos = 0;
while (j < patlen) {
unsigned char c = pat[j++];
if (c == '-') {
if (i > 0 && (npos == 0 || pos[npos - 1] != i))
pos[npos++] = i;
}
else if (c == ' ')
buf[i++] = ' ';
else
buf[i++] = hpf_code_table[c];
}
if (i > 0) {
pos[npos] = 0;
buf[i] = 0;
unsigned char *tem = new unsigned char[npos + 1];
memcpy(tem, pos, npos + 1);
tem = (unsigned char *)ex->lookup(symbol(buf), tem);
if (tem)
delete[] tem;
}
}
void hyphen_trie::hyphenate(const char *word, int len, int *hyphens)
{
int j;
for (j = 0; j < len + 1; j++)
hyphens[j] = 0;
for (j = 0; j < len - 1; j++) {
h = hyphens + j;
find(word + j, len - j);
}
}
inline int max(int m, int n)
{
return m > n ? m : n;
}
void hyphen_trie::do_match(int i, void *v)
{
operation *op = (operation *)v;
while (op != 0) {
h[i - op->distance] = max(h[i - op->distance], op->num);
op = op->next;
}
}
void hyphen_trie::do_delete(void *v)
{
operation *op = (operation *)v;
while (op) {
operation *tem = op;
op = tem->next;
delete tem;
}
}
/* We use very simple rules to parse TeX's hyphenation patterns.
. '%' starts a comment even if preceded by '\'.
. No support for digraphs and like '\$'.
. '^^xx' ('x' is 0-9 or a-f), and '^^x' (character code of 'x' in the
range 0-127) are recognized; other use of '^' causes an error.
. No macro expansion.
. We check for the expression '\patterns{...}' (possibly with
whitespace before and after the braces). Everything between the
braces is taken as hyphenation patterns. Consequently, '{' and '}'
are not allowed in patterns.
. Similarly, '\hyphenation{...}' gives a list of hyphenation
exceptions.
. '\endinput' is recognized also.
. For backwards compatibility, if '\patterns' is missing, the
whole file is treated as a list of hyphenation patterns (only
recognizing '%' as the start of a comment.
*/
int hyphen_trie::hpf_getc(FILE *f)
{
int c = getc(f);
int c1;
int cc = 0;
if (c != '^')
return c;
c = getc(f);
if (c != '^')
goto fail;
c = getc(f);
c1 = getc(f);
if (((c >= '0' && c <= '9') || (c >= 'a' && c <= 'f'))
&& ((c1 >= '0' && c1 <= '9') || (c1 >= 'a' && c1 <= 'f'))) {
if (c >= '0' && c <= '9')
c -= '0';
else
c = c - 'a' + 10;
if (c1 >= '0' && c1 <= '9')
c1 -= '0';
else
c1 = c1 - 'a' + 10;
cc = c * 16 + c1;
}
else {
ungetc(c1, f);
if (c >= 0 && c <= 63)
cc = c + 64;
else if (c >= 64 && c <= 127)
cc = c - 64;
else
goto fail;
}
return cc;
fail:
error("invalid ^, ^^x, or ^^xx character in hyphenation patterns file");
return c;
}
void hyphen_trie::read_patterns_file(const char *name, int append,
dictionary *ex)
{
if (!append)
clear();
char buf[WORD_MAX + 1];
for (int i = 0; i < WORD_MAX + 1; i++)
buf[i] = 0;
int num[WORD_MAX + 1];
errno = 0;
char *path = 0;
FILE *fp = mac_path->open_file(name, &path);
if (fp == 0) {
error("can't find hyphenation patterns file '%1'", name);
return;
}
int c = hpf_getc(fp);
int have_patterns = 0; // we've seen \patterns
int final_pattern = 0; // 1 if we have a trailing closing brace
int have_hyphenation = 0; // we've seen \hyphenation
int final_hyphenation = 0; // 1 if we have a trailing closing brace
int have_keyword = 0; // we've seen either \patterns or \hyphenation
int traditional = 0; // don't handle \patterns
for (;;) {
for (;;) {
if (c == '%') { // skip comments
do {
c = getc(fp);
} while (c != EOF && c != '\n');
}
if (c == EOF || !csspace(c))
break;
c = hpf_getc(fp);
}
if (c == EOF) {
if (have_keyword || traditional) // we are done
break;
else { // rescan file in 'traditional' mode
rewind(fp);
traditional = 1;
c = hpf_getc(fp);
continue;
}
}
int i = 0;
num[0] = 0;
if (!(c == '{' || c == '}')) { // skip braces at line start
do { // scan patterns
if (csdigit(c))
num[i] = c - '0';
else {
buf[i++] = c;
num[i] = 0;
}
c = hpf_getc(fp);
} while (i < WORD_MAX && c != EOF && !csspace(c)
&& c != '%' && c != '{' && c != '}');
}
if (!traditional) {
if (i >= 9 && !strncmp(buf + i - 9, "\\patterns", 9)) {
while (csspace(c))
c = hpf_getc(fp);
if (c == '{') {
if (have_patterns || have_hyphenation)
error("\\patterns is not allowed inside of %1 group",
have_patterns ? "\\patterns" : "\\hyphenation");
else {
have_patterns = 1;
have_keyword = 1;
}
c = hpf_getc(fp);
continue;
}
}
else if (i >= 12 && !strncmp(buf + i - 12, "\\hyphenation", 12)) {
while (csspace(c))
c = hpf_getc(fp);
if (c == '{') {
if (have_patterns || have_hyphenation)
error("\\hyphenation is not allowed inside of %1 group",
have_patterns ? "\\patterns" : "\\hyphenation");
else {
have_hyphenation = 1;
have_keyword = 1;
}
c = hpf_getc(fp);
continue;
}
}
else if (strstr(buf, "\\endinput")) {
if (have_patterns || have_hyphenation)
error("found \\endinput inside of %1 group",
have_patterns ? "\\patterns" : "\\hyphenation");
break;
}
else if (c == '}') {
if (have_patterns) {
have_patterns = 0;
if (i > 0)
final_pattern = 1;
}
else if (have_hyphenation) {
have_hyphenation = 0;
if (i > 0)
final_hyphenation = 1;
}
c = hpf_getc(fp);
}
else if (c == '{') {
if (have_patterns || have_hyphenation)
error("'{' is not allowed within %1 group",
have_patterns ? "\\patterns" : "\\hyphenation");
c = hpf_getc(fp); // skipped if not starting \patterns
// or \hyphenation
}
}
else {
if (c == '{' || c == '}')
c = hpf_getc(fp);
}
if (i > 0) {
if (have_patterns || final_pattern || traditional) {
for (int j = 0; j < i; j++)
buf[j] = hpf_code_table[(unsigned char)buf[j]];
insert_pattern(buf, i, num);
final_pattern = 0;
}
else if (have_hyphenation || final_hyphenation) {
// hyphenation exceptions in a pattern file are subject to `.hy'
// restrictions; we mark such entries with a trailing space
buf[i++] = ' ';
insert_hyphenation(ex, buf, i);
final_hyphenation = 0;
}
}
}
fclose(fp);
free(path);
return;
}
void hyphenate(hyphen_list *h, unsigned flags)
{
if (!current_language)
return;
while (h) {
while (h && h->hyphenation_code == 0)
h = h->next;
int len = 0;
char hbuf[WORD_MAX + 2];
char *buf = hbuf + 1;
hyphen_list *tem;
for (tem = h; tem && len < WORD_MAX; tem = tem->next) {
if (tem->hyphenation_code != 0)
buf[len++] = tem->hyphenation_code;
else
break;
}
hyphen_list *nexth = tem;
if (len >= 2) {
// check `.hw' entries
buf[len] = 0;
unsigned char *pos
= (unsigned char *)current_language->exceptions.lookup(buf);
if (pos != 0) {
int j = 0;
int i = 1;
for (tem = h; tem != 0; tem = tem->next, i++)
if (pos[j] == i) {
tem->hyphen = 1;
j++;
}
}
else {
// check `\hyphenation' entries from pattern files;
// such entries are marked with a trailing space
buf[len] = ' ';
buf[len + 1] = 0;
pos = (unsigned char *)current_language->exceptions.lookup(buf);
if (pos != 0) {
int j = 0;
int i = 1;
tem = h;
if (pos[j] == i) {
if (flags & HYPHEN_FIRST_CHAR)
tem->hyphen = 1;
j++;
}
tem = tem->next;
i++;
if (pos[j] == i) {
if (!(flags & HYPHEN_NOT_FIRST_CHARS))
tem->hyphen = 1;
j++;
}
tem = tem->next;
i++;
if (!(flags & HYPHEN_LAST_CHAR))
--len;
if (flags & HYPHEN_NOT_LAST_CHARS)
--len;
for (; i < len && tem; tem = tem->next, i++)
if (pos[j] == i) {
tem->hyphen = 1;
j++;
}
}
else {
hbuf[0] = hbuf[len + 1] = '.';
int num[WORD_MAX + 3];
current_language->patterns.hyphenate(hbuf, len + 2, num);
// The position of a hyphenation point gets marked with an odd
// number. Example:
//
// hbuf: . h e l p f u l .
// num: 0 0 0 2 4 3 0 0 0 0
if (!(flags & HYPHEN_FIRST_CHAR))
num[2] = 0;
if (flags & HYPHEN_NOT_FIRST_CHARS)
num[3] = 0;
if (flags & HYPHEN_LAST_CHAR)
++len;
if (flags & HYPHEN_NOT_LAST_CHARS)
--len;
int i;
for (i = 2, tem = h; i < len && tem; tem = tem->next, i++)
if (num[i] & 1)
tem->hyphen = 1;
}
}
}
h = nexth;
}
}
static void do_hyphenation_patterns_file(bool append)
{
symbol name = get_long_name(true /* required */);
if (!name.is_null()) {
if (!current_language)
error("no current hyphenation language");
else
current_language->patterns.read_patterns_file(
name.contents(), append,
¤t_language->exceptions);
}
skip_line();
}
static void hyphenation_patterns_file()
{
do_hyphenation_patterns_file(false /* append */);
}
static void hyphenation_patterns_file_append()
{
do_hyphenation_patterns_file(true /* append */);
}
class hyphenation_language_reg : public reg {
public:
const char *get_string();
};
const char *hyphenation_language_reg::get_string()
{
return current_language ? current_language->name.contents() : "";
}
void init_hyphen_requests()
{
init_request("hw", hyphen_word);
init_request("hla", set_hyphenation_language);
init_request("hpf", hyphenation_patterns_file);
init_request("hpfa", hyphenation_patterns_file_append);
register_dictionary.define(".hla", new hyphenation_language_reg);
}
// Local Variables:
// fill-column: 72
// mode: C++
// End:
// vim: set cindent noexpandtab shiftwidth=2 textwidth=72: