banan-os/LibFont/PSF.cpp

215 lines
5.5 KiB
C++

#include <BAN/Debug.h>
#include <BAN/Endianness.h>
#include <BAN/UTF8.h>
#include <LibFont/PSF.h>
#define PSF1_MAGIC0 0x36
#define PSF1_MAGIC1 0x04
#define PSF1_MODE512 0x01
#define PSF1_MODEHASTAB 0x02
#define PSF1_MODEHASSEQ 0x04
#define PSF1_STARTSEQ 0xFFFE
#define PSF1_SEPARATOR 0xFFFF
#define PSF2_MAGIC0 0x72
#define PSF2_MAGIC1 0xB5
#define PSF2_MAGIC2 0x4A
#define PSF2_MAGIC3 0x86
#define PSF2_HAS_UNICODE_TABLE 0x01
#define PSF2_STARTSEQ 0xFE
#define PSF2_SEPARATOR 0xFF
namespace LibFont
{
bool is_psf1(BAN::ConstByteSpan font_data)
{
if (font_data.size() < 2)
return false;
return font_data[0] == PSF1_MAGIC0 && font_data[1] == PSF1_MAGIC1;
}
BAN::ErrorOr<Font> parse_psf1(BAN::ConstByteSpan font_data)
{
struct PSF1Header
{
uint8_t magic[2];
uint8_t mode;
uint8_t char_size;
};
if (font_data.size() < sizeof(PSF1Header))
return BAN::Error::from_errno(EINVAL);
const auto& header = font_data.as<const PSF1Header>();
uint32_t glyph_count = header.mode & PSF1_MODE512 ? 512 : 256;
uint32_t glyph_size = header.char_size;
uint32_t glyph_data_size = glyph_size * glyph_count;
if (font_data.size() < sizeof(PSF1Header) + glyph_data_size)
return BAN::Error::from_errno(EINVAL);
BAN::Vector<uint8_t> glyph_data;
TRY(glyph_data.resize(glyph_data_size));
memcpy(glyph_data.data(), font_data.data() + sizeof(PSF1Header), glyph_data_size);
BAN::HashMap<uint32_t, uint32_t> glyph_offsets;
TRY(glyph_offsets.reserve(glyph_count));
bool codepoint_redef = false;
bool codepoint_sequence = false;
if (header.mode & (PSF1_MODEHASTAB | PSF1_MODEHASSEQ))
{
uint32_t current_index = sizeof(PSF1Header) + glyph_data_size;
uint32_t glyph_index = 0;
while (current_index < font_data.size())
{
uint16_t lo = font_data[current_index];
uint16_t hi = font_data[current_index + 1];
uint16_t codepoint = (hi << 8) | lo;
if (codepoint == PSF1_STARTSEQ)
{
codepoint_sequence = true;
break;
}
else if (codepoint == PSF1_SEPARATOR)
{
glyph_index++;
}
else
{
if (glyph_offsets.contains(codepoint))
codepoint_redef = true;
else
TRY(glyph_offsets.insert(codepoint, glyph_index * glyph_size));
}
current_index += 2;
}
}
else
{
for (uint32_t i = 0; i < glyph_count; i++)
TRY(glyph_offsets.insert(i, i * glyph_size));
}
if (codepoint_redef)
dwarnln("Font contains multiple definitions for same codepoint(s)");
if (codepoint_sequence)
dwarnln("Font contains codepoint sequences (not supported)");
return Font(BAN::move(glyph_offsets), BAN::move(glyph_data), 8, header.char_size, 1);
}
bool is_psf2(BAN::ConstByteSpan font_data)
{
if (font_data.size() < 4)
return false;
return font_data[0] == PSF2_MAGIC0 && font_data[1] == PSF2_MAGIC1 && font_data[2] == PSF2_MAGIC2 && font_data[3] == PSF2_MAGIC3;
}
BAN::ErrorOr<Font> parse_psf2(BAN::ConstByteSpan font_data)
{
struct PSF2Header
{
uint8_t magic[4];
BAN::LittleEndian<uint32_t> version;
BAN::LittleEndian<uint32_t> header_size;
BAN::LittleEndian<uint32_t> flags;
BAN::LittleEndian<uint32_t> glyph_count;
BAN::LittleEndian<uint32_t> glyph_size;
BAN::LittleEndian<uint32_t> height;
BAN::LittleEndian<uint32_t> width;
};
if (font_data.size() < sizeof(PSF2Header))
return BAN::Error::from_errno(EINVAL);
const auto& header = font_data.as<const PSF2Header>();
uint32_t glyph_data_size = header.glyph_count * header.glyph_size;
if (font_data.size() < glyph_data_size + header.header_size)
return BAN::Error::from_errno(EINVAL);
BAN::Vector<uint8_t> glyph_data;
TRY(glyph_data.resize(glyph_data_size));
memcpy(glyph_data.data(), font_data.data() + header.header_size, glyph_data_size);
BAN::HashMap<uint32_t, uint32_t> glyph_offsets;
TRY(glyph_offsets.reserve(400));
bool invalid_utf = false;
bool codepoint_redef = false;
bool codepoint_sequence = false;
uint8_t bytes[4] {};
uint32_t byte_index = 0;
if (header.flags & PSF2_HAS_UNICODE_TABLE)
{
uint32_t glyph_index = 0;
for (uint32_t i = glyph_data_size + header.header_size; i < font_data.size(); i++)
{
uint8_t byte = font_data[i];
if (byte == PSF2_STARTSEQ)
{
codepoint_sequence = true;
break;
}
else if (byte == PSF2_SEPARATOR)
{
if (byte_index)
{
invalid_utf = true;
byte_index = 0;
}
glyph_index++;
}
else
{
ASSERT(byte_index < 4);
bytes[byte_index++] = byte;
uint32_t len = BAN::UTF8::byte_length(bytes[0]);
if (len == 0)
{
invalid_utf = true;
byte_index = 0;
}
else if (len == byte_index)
{
uint32_t codepoint = BAN::UTF8::to_codepoint(bytes);
if (codepoint == BAN::UTF8::invalid)
invalid_utf = true;
else if (glyph_offsets.contains(codepoint))
codepoint_redef = true;
else
TRY(glyph_offsets.insert(codepoint, glyph_index * header.glyph_size));
byte_index = 0;
}
}
}
}
else
{
for (uint32_t i = 0; i < header.glyph_count; i++)
TRY(glyph_offsets.insert(i, i * header.glyph_size));
}
if (invalid_utf)
dwarnln("Font contains invalid UTF-8 codepoint(s)");
if (codepoint_redef)
dwarnln("Font contains multiple definitions for same codepoint(s)");
if (codepoint_sequence)
dwarnln("Font contains codepoint sequences (not supported)");
return Font(BAN::move(glyph_offsets), BAN::move(glyph_data), header.width, header.height, header.glyph_size / header.height);
}
}