diff options
Diffstat (limited to 'supervisor/shared')
| -rw-r--r-- | supervisor/shared/translate.c | 26 | ||||
| -rw-r--r-- | supervisor/shared/translate.h | 32 |
2 files changed, 48 insertions, 10 deletions
diff --git a/supervisor/shared/translate.c b/supervisor/shared/translate.c index 187d5ff8a..606f8fa91 100644 --- a/supervisor/shared/translate.c +++ b/supervisor/shared/translate.c @@ -37,7 +37,7 @@ #include "supervisor/serial.h" void serial_write_compressed(const compressed_string_t* compressed) { - char decompressed[compressed->length]; + char decompressed[decompress_length(compressed)]; decompress(compressed, decompressed); serial_write(decompressed); } @@ -58,12 +58,22 @@ STATIC int put_utf8(char *buf, int u) { } } +uint16_t decompress_length(const compressed_string_t* compressed) { + if (compress_max_length_bits <= 8) { + return 1 + (compressed->data >> (8 - compress_max_length_bits)); + } else { + return 1 + ((compressed->data * 256 + compressed->tail[0]) >> (16 - compress_max_length_bits)); + } +} + char* decompress(const compressed_string_t* compressed, char* decompressed) { - uint8_t this_byte = 0; - uint8_t this_bit = 7; - uint8_t b = compressed->data[this_byte]; + uint8_t this_byte = compress_max_length_bits / 8; + uint8_t this_bit = 7 - compress_max_length_bits % 8; + uint8_t b = (&compressed->data)[this_byte]; + uint16_t length = decompress_length(compressed); + // Stop one early because the last byte is always NULL. - for (uint16_t i = 0; i < compressed->length - 1;) { + for (uint16_t i = 0; i < length - 1;) { uint32_t bits = 0; uint8_t bit_length = 0; uint32_t max_code = lengths[0]; @@ -78,7 +88,7 @@ char* decompress(const compressed_string_t* compressed, char* decompressed) { if (this_bit == 0) { this_bit = 7; this_byte += 1; - b = compressed->data[this_byte]; // This may read past the end but its never used. + b = (&compressed->data)[this_byte]; // This may read past the end but its never used. } else { this_bit -= 1; } @@ -91,14 +101,14 @@ char* decompress(const compressed_string_t* compressed, char* decompressed) { i += put_utf8(decompressed + i, values[searched_length + bits - max_code]); } - decompressed[compressed->length-1] = '\0'; + decompressed[length-1] = '\0'; return decompressed; } inline __attribute__((always_inline)) const compressed_string_t* translate(const char* original) { #ifndef NO_QSTR #define QDEF(id, str) - #define TRANSLATION(id, len, compressed...) if (strcmp(original, id) == 0) { static const compressed_string_t v = {.length = len, .data = compressed}; return &v; } else + #define TRANSLATION(id, firstbyte, ...) if (strcmp(original, id) == 0) { static const compressed_string_t v = { .data = firstbyte, .tail = { __VA_ARGS__ } }; return &v; } else #include "genhdr/qstrdefs.generated.h" #undef TRANSLATION #undef QDEF diff --git a/supervisor/shared/translate.h b/supervisor/shared/translate.h index 5e8acbb6a..731b26d12 100644 --- a/supervisor/shared/translate.h +++ b/supervisor/shared/translate.h @@ -29,13 +29,41 @@ #include <stdint.h> +// The format of the compressed data is: +// - the size of the uncompressed string in UTF-8 bytes, encoded as a +// (compress_max_length_bits)-bit number. compress_max_length_bits is +// computed during dictionary generation time, and happens to be 8 +// for all current platforms. However, it'll probably end up being +// 9 in some translations sometime in the future. This length excludes +// the trailing NUL, though notably decompress_length includes it. +// +// - followed by the huffman encoding of the individual UTF-16 code +// points that make up the string. The trailing "\0" is not +// represented by a huffman code, but is implied by the length. +// (building the huffman encoding on UTF-16 code points gave better +// compression than building it on UTF-8 bytes) +// +// The "data" / "tail" construct is so that the struct's last member is a +// "flexible array". However, the _only_ member is not permitted to be +// a flexible member, so we have to declare the first byte as a separte +// member of the structure. +// +// For translations where length needs 8 bits, this saves about 1.5 +// bytes per string on average compared to a structure of {uint16_t, +// flexible array}, but is also future-proofed against strings with +// UTF-8 length above 256, with a savings of about 1.375 bytes per +// string. typedef struct { - uint16_t length; - const uint8_t data[]; + uint8_t data; + const uint8_t tail[]; } compressed_string_t; +// Return the compressed, translated version of a source string +// Usually, due to LTO, this is optimized into a load of a constant +// pointer. const compressed_string_t* translate(const char* c); void serial_write_compressed(const compressed_string_t* compressed); char* decompress(const compressed_string_t* compressed, char* decompressed); +uint16_t decompress_length(const compressed_string_t* compressed); #endif // MICROPY_INCLUDED_SUPERVISOR_TRANSLATE_H |