eBookReaderSwitch/mupdf/source/pdf/pdf-unicode.c

132 lines
3.3 KiB
C
Raw Normal View History

#include "mupdf/fitz.h"
#include "mupdf/pdf.h"
#include <string.h>
/* Load or synthesize ToUnicode map for fonts */
static void
pdf_remap_cmap_range(fz_context *ctx, pdf_cmap *ucs_from_gid,
unsigned int cpt, unsigned int gid, unsigned int n, pdf_cmap *ucs_from_cpt)
{
unsigned int k;
int ucsbuf[8];
int ucslen;
for (k = 0; k <= n; ++k)
{
ucslen = pdf_lookup_cmap_full(ucs_from_cpt, cpt + k, ucsbuf);
if (ucslen == 1)
pdf_map_range_to_range(ctx, ucs_from_gid, gid + k, gid + k, ucsbuf[0]);
else if (ucslen > 1)
pdf_map_one_to_many(ctx, ucs_from_gid, gid + k, ucsbuf, ucslen);
}
}
static pdf_cmap *
pdf_remap_cmap(fz_context *ctx, pdf_cmap *gid_from_cpt, pdf_cmap *ucs_from_cpt)
{
pdf_cmap *ucs_from_gid;
unsigned int a, b, x;
int i;
ucs_from_gid = pdf_new_cmap(ctx);
fz_try(ctx)
{
if (gid_from_cpt->usecmap)
ucs_from_gid->usecmap = pdf_remap_cmap(ctx, gid_from_cpt->usecmap, ucs_from_cpt);
for (i = 0; i < gid_from_cpt->codespace_len; i++)
{
pdf_add_codespace(ctx, ucs_from_gid,
gid_from_cpt->codespace[i].low,
gid_from_cpt->codespace[i].high,
gid_from_cpt->codespace[i].n);
}
for (i = 0; i < gid_from_cpt->rlen; ++i)
{
a = gid_from_cpt->ranges[i].low;
b = gid_from_cpt->ranges[i].high;
x = gid_from_cpt->ranges[i].out;
pdf_remap_cmap_range(ctx, ucs_from_gid, a, x, b - a, ucs_from_cpt);
}
for (i = 0; i < gid_from_cpt->xlen; ++i)
{
a = gid_from_cpt->xranges[i].low;
b = gid_from_cpt->xranges[i].high;
x = gid_from_cpt->xranges[i].out;
pdf_remap_cmap_range(ctx, ucs_from_gid, a, x, b - a, ucs_from_cpt);
}
/* Font encoding CMaps don't have one-to-many mappings, so we can ignore the mranges. */
pdf_sort_cmap(ctx, ucs_from_gid);
}
fz_catch(ctx)
{
pdf_drop_cmap(ctx, ucs_from_gid);
fz_rethrow(ctx);
}
return ucs_from_gid;
}
void
pdf_load_to_unicode(fz_context *ctx, pdf_document *doc, pdf_font_desc *font,
const char **strings, char *collection, pdf_obj *cmapstm)
{
unsigned int cpt;
if (pdf_is_stream(ctx, cmapstm))
{
pdf_cmap *ucs_from_cpt = pdf_load_embedded_cmap(ctx, doc, cmapstm);
fz_try(ctx)
font->to_unicode = pdf_remap_cmap(ctx, font->encoding, ucs_from_cpt);
fz_always(ctx)
pdf_drop_cmap(ctx, ucs_from_cpt);
fz_catch(ctx)
fz_rethrow(ctx);
font->size += pdf_cmap_size(ctx, font->to_unicode);
}
else if (collection)
{
if (!strcmp(collection, "Adobe-CNS1"))
font->to_unicode = pdf_load_system_cmap(ctx, "Adobe-CNS1-UCS2");
else if (!strcmp(collection, "Adobe-GB1"))
font->to_unicode = pdf_load_system_cmap(ctx, "Adobe-GB1-UCS2");
else if (!strcmp(collection, "Adobe-Japan1"))
font->to_unicode = pdf_load_system_cmap(ctx, "Adobe-Japan1-UCS2");
else if (!strcmp(collection, "Adobe-Korea1"))
font->to_unicode = pdf_load_system_cmap(ctx, "Adobe-Korea1-UCS2");
return;
}
if (strings)
{
/* TODO one-to-many mappings */
font->cid_to_ucs = fz_malloc_array(ctx, 256, unsigned short);
font->cid_to_ucs_len = 256;
font->size += 256 * sizeof *font->cid_to_ucs;
for (cpt = 0; cpt < 256; cpt++)
{
if (strings[cpt])
font->cid_to_ucs[cpt] = fz_unicode_from_glyph_name(strings[cpt]);
else
font->cid_to_ucs[cpt] = FZ_REPLACEMENT_CHARACTER;
}
}
if (!font->to_unicode && !font->cid_to_ucs)
{
/* TODO: synthesize a ToUnicode if it's a freetype font with
* cmap and/or post tables or if it has glyph names. */
}
}