actual/packages/node-libofx/OpenSP-1.5.2/lib/Text.cxx
2022-04-28 22:44:38 -04:00

433 lines
9.5 KiB
C++

// Copyright (c) 1994 James Clark
// See the file COPYING for copying permission.
#ifdef __GNUG__
#pragma implementation
#endif
#include "splib.h"
#include "Text.h"
// for memcmp()
#include <string.h>
#ifdef SP_NAMESPACE
namespace SP_NAMESPACE {
#endif
Text::Text()
{
}
Text::~Text()
{
}
void Text::addChar(Char c, const Location &loc)
{
if (items_.size() == 0
|| items_.back().type != TextItem::data
|| loc.origin().pointer() != items_.back().loc.origin().pointer()
|| loc.index() != (items_.back().loc.index()
+ (chars_.size() - items_.back().index))) {
items_.resize(items_.size() + 1);
items_.back().loc = loc;
items_.back().type = TextItem::data;
items_.back().index = chars_.size();
}
chars_ += c;
}
void Text::addChars(const Char *p, size_t length, const Location &loc)
{
if (items_.size() == 0
|| items_.back().type != TextItem::data
|| loc.origin().pointer() != items_.back().loc.origin().pointer()
|| loc.index() != (items_.back().loc.index()
+ (chars_.size() - items_.back().index))) {
items_.resize(items_.size() + 1);
items_.back().loc = loc;
items_.back().type = TextItem::data;
items_.back().index = chars_.size();
}
chars_.append(p, length);
}
void Text::addCdata(const StringC &str,
const ConstPtr<Origin> &origin)
{
addSimple(TextItem::cdata, Location(origin, 0));
chars_.append(str.data(), str.size());
}
void Text::addSdata(const StringC &str,
const ConstPtr<Origin> &origin)
{
addSimple(TextItem::sdata, Location(origin, 0));
chars_.append(str.data(), str.size());
}
void Text::addNonSgmlChar(Char c, const Location &loc)
{
addSimple(TextItem::nonSgml, loc);
chars_ += c;
}
void Text::addCharsTokenize(const Char *str, size_t n, const Location &loc,
Char space)
{
Location loci(loc);
// FIXME speed this up
for (size_t i = 0; i < n; loci += 1, i++) {
if (str[i] == space && (size() == 0 || lastChar() == space))
ignoreChar(str[i], loci);
else
addChar(str[i], loci);
}
}
void Text::tokenize(Char space, Text &text) const
{
TextIter iter(*this);
TextItem::Type type;
const Char *p;
size_t n;
const Location *loc;
while (iter.next(type, p, n, loc)) {
switch (type) {
case TextItem::data:
text.addCharsTokenize(p, n, *loc, space);
break;
case TextItem::sdata:
case TextItem::cdata:
{
text.addEntityStart(*loc);
text.addCharsTokenize(p, n, *loc, space);
Location tem(*loc);
tem += n;
text.addEntityEnd(tem);
}
break;
case TextItem::ignore:
text.ignoreChar(*p, *loc);
break;
default:
text.addSimple(type, *loc);
break;
}
}
if (text.size() > 0 && text.lastChar() == space)
text.ignoreLastChar();
}
void Text::addSimple(TextItem::Type type, const Location &loc)
{
items_.resize(items_.size() + 1);
items_.back().loc = loc;
items_.back().type = type;
items_.back().index = chars_.size();
}
void Text::ignoreChar(Char c, const Location &loc)
{
items_.resize(items_.size() + 1);
items_.back().loc = loc;
items_.back().type = TextItem::ignore;
items_.back().c = c;
items_.back().index = chars_.size();
}
void Text::ignoreLastChar()
{
size_t lastIndex = chars_.size() - 1;
size_t i;
for (i = items_.size() - 1; items_[i].index > lastIndex; i--)
;
// lastIndex >= items_[i].index
if (items_[i].index != lastIndex) {
items_.resize(items_.size() + 1);
i++;
for (size_t j = items_.size() - 1; j > i; j--)
items_[j] = items_[j - 1];
items_[i].index = lastIndex;
items_[i].loc = items_[i - 1].loc;
items_[i].loc += lastIndex - items_[i - 1].index;
}
items_[i].c = chars_[chars_.size() - 1];
items_[i].type = TextItem::ignore;
for (size_t j = i + 1; j < items_.size(); j++)
items_[j].index = lastIndex;
chars_.resize(chars_.size() - 1);
}
// All characters other than spaces are substed.
void Text::subst(const SubstTable &table, Char space)
{
for (size_t i = 0; i < items_.size(); i++)
if (items_[i].type == TextItem::data) {
size_t lim = (i + 1 < items_.size()
? items_[i + 1].index
: chars_.size());
size_t j;
for (j = items_[i].index; j < lim; j++) {
Char c = chars_[j];
if (c != space && c != table[c])
break;
}
if (j < lim) {
size_t start = items_[i].index;
StringC origChars(chars_.data() + start, lim - start);
for (; j < lim; j++)
if (chars_[j] != space)
table.subst(chars_[j]);
items_[i].loc = Location(new MultiReplacementOrigin(items_[i].loc,
origChars),
0);
}
}
}
void Text::clear()
{
chars_.resize(0);
items_.clear();
}
Boolean Text::startDelimLocation(Location &loc) const
{
if (items_.size() == 0 || items_[0].type != TextItem::startDelim)
return 0;
loc = items_[0].loc;
return 1;
}
Boolean Text::endDelimLocation(Location &loc) const
{
if (items_.size() == 0)
return 0;
switch (items_.back().type) {
case TextItem::endDelim:
case TextItem::endDelimA:
break;
default:
return 0;
}
loc = items_.back().loc;
return 1;
}
Boolean Text::delimType(Boolean &lita) const
{
if (items_.size() == 0)
return 0;
switch (items_.back().type) {
case TextItem::endDelim:
lita = 0;
return 1;
case TextItem::endDelimA:
lita = 1;
return 1;
default:
break;
}
return 0;
}
TextItem::TextItem()
{
}
TextItem::~TextItem()
{
}
TextItem::TextItem(const TextItem& from)
: type(from.type),
c(from.c),
loc(from.loc),
index(from.index)
{
}
TextItem& TextItem::operator=(const TextItem& from)
{
if (this != &from) {
type = from.type;
c = from.c;
loc = from.loc;
index = from.index;
}
return *this;
}
void Text::swap(Text &to)
{
items_.swap(to.items_);
chars_.swap(to.chars_);
}
TextIter::TextIter(const Text &text)
: ptr_(text.items_.begin()), text_(&text)
{
}
const Char *TextIter::chars(size_t &length) const
{
if (ptr_->type == TextItem::ignore) {
length = 1;
return &ptr_->c;
}
else {
const StringC &chars = text_->chars_;
size_t charsIndex = ptr_->index;
if (ptr_ + 1 != text_->items_.begin() + text_->items_.size())
length = ptr_[1].index - charsIndex;
else
length = chars.size() - charsIndex;
return chars.data() + charsIndex;
}
}
Boolean TextIter::next(TextItem::Type &type, const Char *&str, size_t &length,
const Location *&loc)
{
const TextItem *end = text_->items_.begin() + text_->items_.size();
if (ptr_ == end)
return 0;
type = ptr_->type;
loc = &ptr_->loc;
if (type == TextItem::ignore) {
str = &ptr_->c;
length = 1;
}
else {
const StringC &chars = text_->chars_;
size_t charsIndex = ptr_->index;
str = chars.data() + charsIndex;
if (ptr_ + 1 != end)
length = ptr_[1].index - charsIndex;
else
length = chars.size() - charsIndex;
}
ptr_++;
return 1;
}
void Text::insertChars(const StringC &s, const Location &loc)
{
chars_.insert(0, s);
items_.resize(items_.size() + 1);
for (size_t i = items_.size() - 1; i > 0; i--) {
items_[i] = items_[i - 1];
items_[i].index += s.size();
}
items_[0].loc = loc;
items_[0].type = TextItem::data;
items_[0].index = 0;
}
size_t Text::normalizedLength(size_t normsep) const
{
size_t n = size();
n += normsep;
for (size_t i = 0; i < items_.size(); i++)
switch (items_[i].type) {
case TextItem::sdata:
case TextItem::cdata:
n += normsep;
break;
default:
break;
}
return n;
}
// This is used to determine for a FIXED CDATA attribute
// whether a specified value if equal to the default value.
Boolean Text::fixedEqual(const Text &text) const
{
if (string() != text.string())
return 0;
size_t j = 0;
for (size_t i = 0; i < items_.size(); i++)
switch (items_[i].type) {
case TextItem::cdata:
case TextItem::sdata:
for (;;) {
if (j >= text.items_.size())
return 0;
if (text.items_[j].type == TextItem::nonSgml)
return 0;
if (text.items_[j].type == TextItem::cdata
|| text.items_[j].type == TextItem::sdata)
break;
j++;
}
if (text.items_[j].index != items_[i].index
|| (text.items_[j].loc.origin()->entityDecl()
!= items_[i].loc.origin()->entityDecl()))
return 0;
break;
case TextItem::nonSgml:
for (;;) {
if (j >= text.items_.size())
return 0;
if (text.items_[j].type == TextItem::cdata
|| text.items_[j].type == TextItem::sdata)
return 0;
if (text.items_[j].type == TextItem::nonSgml)
break;
j++;
}
if (text.items_[j].index != items_[i].index)
return 0;
break;
default:
break;
}
for (; j < text.items_.size(); j++)
switch (text.items_[j].type) {
case TextItem::cdata:
case TextItem::sdata:
case TextItem::nonSgml:
return 0;
default:
break;
}
return 1;
}
Boolean Text::charLocation(size_t ind, const ConstPtr<Origin> *&origin, Index &index) const
{
// Find the last item whose index <= ind.
// Invariant:
// indexes < i implies index <= ind
// indexes >= lim implies index > ind
// The first item will always have index 0.
size_t i = 1;
size_t lim = items_.size();
while (i < lim) {
size_t mid = i + (lim - i)/2;
if (items_[mid].index > ind)
lim = mid;
else
i = mid + 1;
}
#if 0
for (size_t i = 1; i < items_.size(); i++)
if (items_[i].index > ind)
break;
#endif
i--;
// If items_.size() == 0, then i == lim.
if (i < lim) {
origin = &items_[i].loc.origin();
index = items_[i].loc.index() + (ind - items_[i].index);
}
return 1;
}
#ifdef SP_NAMESPACE
}
#endif