diff --git a/example/mzcat/mzcat.c b/example/mzcat/mzcat.c index 2952d8cd..3f7a9004 100644 --- a/example/mzcat/mzcat.c +++ b/example/mzcat/mzcat.c @@ -78,18 +78,21 @@ Supported compression formats: #define WUFFS_EXAMPLE_USE_SECCOMP #endif -// LZMA's default dictionary size is 16 MiB. This program's destination-buffer -// size defaults to twice that. #ifndef DST_BUFFER_ARRAY_SIZE -#define DST_BUFFER_ARRAY_SIZE (32 * 1024 * 1024) +#define DST_BUFFER_ARRAY_SIZE (128 * 1024) #endif #ifndef SRC_BUFFER_ARRAY_SIZE #define SRC_BUFFER_ARRAY_SIZE (128 * 1024) #endif +#ifndef WORKBUF_ARRAY_SIZE +#define WORKBUF_ARRAY_SIZE (64 * 1024 * 1024) +#endif + uint8_t g_dst_buffer_array[DST_BUFFER_ARRAY_SIZE]; uint8_t g_src_buffer_array[SRC_BUFFER_ARRAY_SIZE]; +uint8_t g_workbuf_array[WORKBUF_ARRAY_SIZE]; // ---- @@ -269,7 +272,9 @@ main1(int argc, char** argv) { while (true) { wuffs_base__status status = wuffs_base__io_transformer__transform_io( - g_io_transformer, &dst, &src, wuffs_base__empty_slice_u8()); + g_io_transformer, &dst, &src, + wuffs_base__make_slice_u8(&g_workbuf_array[0], + sizeof(g_workbuf_array))); if (dst.meta.ri < dst.meta.wi) { // TODO: handle EINTR and other write errors; see "man 2 write". diff --git a/lang/builtin/builtin.go b/lang/builtin/builtin.go index e5eb44e0..e1ba4963 100644 --- a/lang/builtin/builtin.go +++ b/lang/builtin/builtin.go @@ -245,6 +245,7 @@ var Statuses = [...]string{ `"$mispositioned read"`, `"$mispositioned write"`, `"$short read"`, + `"$short workbuf"`, `"$short write"`, // Errors. @@ -1024,6 +1025,7 @@ var InterfaceFuncs = []string{ // ---- io_transformer + // TODO: "io_transformer.add_history!(hist: slice u8)"? Should it take a workbuf arg? "io_transformer.dst_history_retain_length() optional_u63", "io_transformer.get_quirk(key: u32) u64", "io_transformer.set_quirk!(key: u32, value: u64) status", diff --git a/release/c/wuffs-unsupported-snapshot.c b/release/c/wuffs-unsupported-snapshot.c index d4e4fea0..d051aa2f 100644 --- a/release/c/wuffs-unsupported-snapshot.c +++ b/release/c/wuffs-unsupported-snapshot.c @@ -535,6 +535,7 @@ extern const char wuffs_base__suspension__even_more_information[]; extern const char wuffs_base__suspension__mispositioned_read[]; extern const char wuffs_base__suspension__mispositioned_write[]; extern const char wuffs_base__suspension__short_read[]; +extern const char wuffs_base__suspension__short_workbuf[]; extern const char wuffs_base__suspension__short_write[]; extern const char wuffs_base__error__bad_i_o_position[]; extern const char wuffs_base__error__bad_argument_length_too_short[]; @@ -10345,9 +10346,9 @@ extern const char wuffs_lzma__error__unsupported_properties[]; // ---------------- Public Consts -#define WUFFS_LZMA__DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE 4294967295u +#define WUFFS_LZMA__DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE 0u -#define WUFFS_LZMA__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE 0u +#define WUFFS_LZMA__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE 4294967568u #define WUFFS_LZMA__QUIRK_FORMAT_EXTENSION 1348001792u @@ -10466,6 +10467,8 @@ struct wuffs_lzma__decoder__struct { uint32_t f_pb; uint32_t f_format_extension; uint32_t f_dict_size; + uint32_t f_dict_workbuf_index; + uint32_t f_dict_seen; uint64_t f_decoded_length; uint64_t f_lzma2_prior_decoded_length; uint64_t f_lzma2_encoded_length_have; @@ -10524,6 +10527,7 @@ struct wuffs_lzma__decoder__struct { uint32_t v_index_small_dist_base; uint32_t v_index_small_dist_extra; uint32_t v_index_large_dist; + uint32_t v_dist; uint64_t scratch; } s_decode_bitstream_slow; struct { @@ -13798,9 +13802,9 @@ extern const char wuffs_xz__error__unsupported_filter[]; // ---------------- Public Consts -#define WUFFS_XZ__DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE 4294967295u +#define WUFFS_XZ__DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE 0u -#define WUFFS_XZ__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE 0u +#define WUFFS_XZ__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE 4294967568u // ---------------- Struct Declarations @@ -15627,6 +15631,7 @@ const char wuffs_base__suspension__even_more_information[] = "$base: even more i const char wuffs_base__suspension__mispositioned_read[] = "$base: mispositioned read"; const char wuffs_base__suspension__mispositioned_write[] = "$base: mispositioned write"; const char wuffs_base__suspension__short_read[] = "$base: short read"; +const char wuffs_base__suspension__short_workbuf[] = "$base: short workbuf"; const char wuffs_base__suspension__short_write[] = "$base: short write"; const char wuffs_base__error__bad_i_o_position[] = "#base: bad I/O position"; const char wuffs_base__error__bad_argument_length_too_short[] = "#base: bad argument (length too short)"; @@ -50871,6 +50876,7 @@ const char wuffs_lzma__error__bad_header[] = "#lzma: bad header"; const char wuffs_lzma__error__truncated_input[] = "#lzma: truncated input"; const char wuffs_lzma__error__unsupported_decoded_length[] = "#lzma: unsupported decoded length"; const char wuffs_lzma__error__unsupported_properties[] = "#lzma: unsupported properties"; +const char wuffs_lzma__error__internal_error_inconsistent_dictionary_state[] = "#lzma: internal error: inconsistent dictionary state"; // ---------------- Private Consts @@ -50914,21 +50920,31 @@ static wuffs_base__status wuffs_lzma__decoder__decode_bitstream_slow( wuffs_lzma__decoder* self, wuffs_base__io_buffer* a_dst, - wuffs_base__io_buffer* a_src); + wuffs_base__io_buffer* a_src, + wuffs_base__slice_u8 a_workbuf); + +WUFFS_BASE__GENERATED_C_CODE +static wuffs_base__status +wuffs_lzma__decoder__add_history( + wuffs_lzma__decoder* self, + wuffs_base__slice_u8 a_hist, + wuffs_base__slice_u8 a_workbuf); WUFFS_BASE__GENERATED_C_CODE static wuffs_base__status wuffs_lzma__decoder__do_transform_io( wuffs_lzma__decoder* self, wuffs_base__io_buffer* a_dst, - wuffs_base__io_buffer* a_src); + wuffs_base__io_buffer* a_src, + wuffs_base__slice_u8 a_workbuf); WUFFS_BASE__GENERATED_C_CODE static wuffs_base__status wuffs_lzma__decoder__decode_bitstream( wuffs_lzma__decoder* self, wuffs_base__io_buffer* a_dst, - wuffs_base__io_buffer* a_src); + wuffs_base__io_buffer* a_src, + wuffs_base__slice_u8 a_workbuf); WUFFS_BASE__GENERATED_C_CODE static wuffs_base__empty_struct @@ -51030,7 +51046,8 @@ static wuffs_base__status wuffs_lzma__decoder__decode_bitstream_slow( wuffs_lzma__decoder* self, wuffs_base__io_buffer* a_dst, - wuffs_base__io_buffer* a_src) { + wuffs_base__io_buffer* a_src, + wuffs_base__slice_u8 a_workbuf) { wuffs_base__status status = wuffs_base__make_status(NULL); uint8_t v_c8 = 0; @@ -51069,6 +51086,9 @@ wuffs_lzma__decoder__decode_bitstream_slow( uint32_t v_index_small_dist_extra = 0; uint32_t v_index_small_dist = 0; uint32_t v_index_large_dist = 0; + uint32_t v_dist = 0; + uint32_t v_adj_dist = 0; + uint64_t v_wb_index = 0; uint8_t* iop_a_dst = NULL; uint8_t* io0_a_dst WUFFS_BASE__POTENTIALLY_UNUSED = NULL; @@ -51122,6 +51142,7 @@ wuffs_lzma__decoder__decode_bitstream_slow( v_index_small_dist_base = self->private_data.s_decode_bitstream_slow.v_index_small_dist_base; v_index_small_dist_extra = self->private_data.s_decode_bitstream_slow.v_index_small_dist_extra; v_index_large_dist = self->private_data.s_decode_bitstream_slow.v_index_large_dist; + v_dist = self->private_data.s_decode_bitstream_slow.v_dist; } switch (coro_susp_point) { WUFFS_BASE__COROUTINE_SUSPENSION_POINT_0; @@ -52023,21 +52044,73 @@ wuffs_lzma__decoder__decode_bitstream_slow( v_state = ((uint32_t)(WUFFS_LZMA__STATE_TRANSITION_LONGREP[v_state])); } while (0); } while (0); - if ((((uint64_t)((v_rep0 + 1u))) > ((uint64_t)(v_pos + self->private_impl.f_lzma2_prior_decoded_length))) || (((uint64_t)((v_rep0 + 1u))) > ((uint64_t)(self->private_impl.f_dict_size)))) { + v_dist = (v_rep0 + 1u); + if ((((uint64_t)(v_dist)) > ((uint64_t)(v_pos + self->private_impl.f_lzma2_prior_decoded_length))) || (((uint64_t)(v_dist)) > ((uint64_t)(self->private_impl.f_dict_size)))) { status = wuffs_base__make_status(wuffs_lzma__error__bad_distance); goto exit; } v_pos += ((uint64_t)(v_len)); - while (((uint64_t)(v_len)) > ((uint64_t)(io2_a_dst - iop_a_dst))) { + while (274u > ((uint64_t)(io2_a_dst - iop_a_dst))) { status = wuffs_base__make_status(wuffs_base__suspension__short_write); WUFFS_BASE__COROUTINE_SUSPENSION_POINT_MAYBE_SUSPEND(38); } - if (((uint64_t)((v_rep0 + 1u))) > ((uint64_t)(iop_a_dst - io0_a_dst))) { - status = wuffs_base__make_status(wuffs_base__error__insufficient_history); - goto exit; + if (((uint64_t)(v_dist)) > ((uint64_t)(iop_a_dst - io0_a_dst))) { + v_adj_dist = ((uint32_t)((((uint64_t)(v_dist)) - ((uint64_t)(iop_a_dst - io0_a_dst))))); + if (v_adj_dist > self->private_impl.f_dict_seen) { + status = wuffs_base__make_status(wuffs_lzma__error__bad_distance); + goto exit; + } + v_wb_index = ((uint64_t)(((uint64_t)(self->private_impl.f_dict_workbuf_index)) - ((uint64_t)(v_adj_dist)))); + while (v_wb_index >= 9223372036854775808u) { + v_wb_index += ((uint64_t)(self->private_impl.f_dict_size)); + } + if (v_wb_index > ((uint64_t)(a_workbuf.len))) { + status = wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + goto exit; + } + if (v_len < v_adj_dist) { + wuffs_private_impl__io_writer__limited_copy_u32_from_slice( + &iop_a_dst, io2_a_dst,(v_len + 1u), wuffs_base__slice_u8__subslice_i(a_workbuf, v_wb_index)); + if ( ! (iop_a_dst > io1_a_dst)) { + status = wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + goto exit; + } + v_match_byte = iop_a_dst[-1]; + iop_a_dst--; + if ( ! (iop_a_dst > io1_a_dst)) { + status = wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + goto exit; + } + v_prev_byte = iop_a_dst[-1]; + continue; + } else if (v_len == v_adj_dist) { + wuffs_private_impl__io_writer__limited_copy_u32_from_slice( + &iop_a_dst, io2_a_dst,v_len, wuffs_base__slice_u8__subslice_i(a_workbuf, v_wb_index)); + wuffs_private_impl__io_writer__limited_copy_u32_from_history( + &iop_a_dst, io0_a_dst, io2_a_dst, 1u, v_dist); + if ( ! (iop_a_dst > io1_a_dst)) { + status = wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + goto exit; + } + v_match_byte = iop_a_dst[-1]; + iop_a_dst--; + if ( ! (iop_a_dst > io1_a_dst)) { + status = wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + goto exit; + } + v_prev_byte = iop_a_dst[-1]; + continue; + } + wuffs_private_impl__io_writer__limited_copy_u32_from_slice( + &iop_a_dst, io2_a_dst,v_adj_dist, wuffs_base__slice_u8__subslice_i(a_workbuf, v_wb_index)); + v_len -= v_adj_dist; + if ((((uint64_t)(v_len)) > ((uint64_t)(io2_a_dst - iop_a_dst))) || (((uint64_t)(v_dist)) > ((uint64_t)(iop_a_dst - io0_a_dst)))) { + status = wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + goto exit; + } } v_match_cusp = wuffs_private_impl__io_writer__limited_copy_u32_from_history_fast_return_cusp( - &iop_a_dst, io0_a_dst, io2_a_dst, v_len, (v_rep0 + 1u)); + &iop_a_dst, io0_a_dst, io2_a_dst, v_len, v_dist); v_match_byte = ((uint8_t)((v_match_cusp >> 8u))); v_prev_byte = ((uint8_t)(v_match_cusp)); } @@ -52084,6 +52157,7 @@ wuffs_lzma__decoder__decode_bitstream_slow( self->private_data.s_decode_bitstream_slow.v_index_small_dist_base = v_index_small_dist_base; self->private_data.s_decode_bitstream_slow.v_index_small_dist_extra = v_index_small_dist_extra; self->private_data.s_decode_bitstream_slow.v_index_large_dist = v_index_large_dist; + self->private_data.s_decode_bitstream_slow.v_dist = v_dist; goto exit; exit: @@ -52097,6 +52171,63 @@ wuffs_lzma__decoder__decode_bitstream_slow( return status; } +// -------- func lzma.decoder.add_history + +WUFFS_BASE__GENERATED_C_CODE +static wuffs_base__status +wuffs_lzma__decoder__add_history( + wuffs_lzma__decoder* self, + wuffs_base__slice_u8 a_hist, + wuffs_base__slice_u8 a_workbuf) { + uint64_t v_dict_workbuf_index = 0; + uint64_t v_dict_size = 0; + uint64_t v_hist_length = 0; + wuffs_base__slice_u8 v_s = {0}; + uint64_t v_n_copied = 0; + uint64_t v_n = 0; + + v_dict_workbuf_index = ((uint64_t)(self->private_impl.f_dict_workbuf_index)); + v_dict_size = ((uint64_t)(self->private_impl.f_dict_size)); + if (((uint64_t)(a_hist.len)) == 0u) { + return wuffs_base__make_status(NULL); + } + if (((uint64_t)(a_workbuf.len)) < (v_dict_size + 273u)) { + return wuffs_base__make_status(wuffs_base__error__bad_workbuf_length); + } + v_hist_length = ((uint64_t)(a_hist.len)); + if (v_hist_length > 4294967295u) { + self->private_impl.f_dict_seen = 4294967295u; + } else { + wuffs_private_impl__u32__sat_add_indirect(&self->private_impl.f_dict_seen, ((uint32_t)(v_hist_length))); + } + v_s = a_hist; + if (((uint64_t)(v_s.len)) >= v_dict_size) { + v_s = wuffs_private_impl__slice_u8__suffix(v_s, v_dict_size); + wuffs_private_impl__slice_u8__copy_from_slice(a_workbuf, v_s); + self->private_impl.f_dict_workbuf_index = 0u; + } else if (v_dict_workbuf_index > v_dict_size) { + return wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + } else { + v_n_copied = wuffs_private_impl__slice_u8__copy_from_slice(wuffs_base__slice_u8__subslice_ij(a_workbuf, v_dict_workbuf_index, v_dict_size), v_s); + if (v_n_copied < ((uint64_t)(v_s.len))) { + v_n = wuffs_private_impl__slice_u8__copy_from_slice(a_workbuf, wuffs_base__slice_u8__subslice_i(v_s, v_n_copied)); + self->private_impl.f_dict_workbuf_index = ((uint32_t)(v_n)); + } else { + v_n = ((uint64_t)(v_dict_workbuf_index + v_n_copied)); + if (v_n < v_dict_size) { + self->private_impl.f_dict_workbuf_index = ((uint32_t)(v_n)); + } else { + self->private_impl.f_dict_workbuf_index = 0u; + } + } + } + if ((273u > v_dict_size) || (v_dict_size > ((uint64_t)(a_workbuf.len)))) { + return wuffs_base__make_status(wuffs_lzma__error__internal_error_inconsistent_dictionary_state); + } + wuffs_private_impl__slice_u8__copy_from_slice(wuffs_base__slice_u8__subslice_i(a_workbuf, v_dict_size), wuffs_base__slice_u8__subslice_j(a_workbuf, 273u)); + return wuffs_base__make_status(NULL); +} + // -------- func lzma.decoder.get_quirk WUFFS_BASE__GENERATED_C_CODE @@ -52169,10 +52300,7 @@ wuffs_lzma__decoder__dst_history_retain_length( return wuffs_base__utility__make_optional_u63(false, 0u); } - if (self->private_impl.f_dict_size == 0u) { - return wuffs_base__utility__make_optional_u63(false, 0u); - } - return wuffs_base__utility__make_optional_u63(true, ((uint64_t)(self->private_impl.f_dict_size))); + return wuffs_base__utility__make_optional_u63(true, 0u); } // -------- func lzma.decoder.workbuf_len @@ -52189,7 +52317,13 @@ wuffs_lzma__decoder__workbuf_len( return wuffs_base__utility__empty_range_ii_u64(); } - return wuffs_base__utility__make_range_ii_u64(0u, 0u); + uint64_t v_m = 0; + + if (self->private_impl.f_dict_size == 0u) { + return wuffs_base__utility__make_range_ii_u64(0u, 0u); + } + v_m = (((uint64_t)(self->private_impl.f_dict_size)) + 273u); + return wuffs_base__utility__make_range_ii_u64(v_m, v_m); } // -------- func lzma.decoder.transform_io @@ -52222,22 +52356,59 @@ wuffs_lzma__decoder__transform_io( self->private_impl.active_coroutine = 0; wuffs_base__status status = wuffs_base__make_status(NULL); - wuffs_base__status v_status = wuffs_base__make_status(NULL); + uint64_t v_mark = 0; + wuffs_base__status v_dti_status = wuffs_base__make_status(NULL); + wuffs_base__status v_ah_status = wuffs_base__make_status(NULL); + + uint8_t* iop_a_dst = NULL; + uint8_t* io0_a_dst WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + uint8_t* io1_a_dst WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + uint8_t* io2_a_dst WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + if (a_dst && a_dst->data.ptr) { + io0_a_dst = a_dst->data.ptr; + io1_a_dst = io0_a_dst + a_dst->meta.wi; + iop_a_dst = io1_a_dst; + io2_a_dst = io0_a_dst + a_dst->data.len; + if (a_dst->meta.closed) { + io2_a_dst = iop_a_dst; + } + } uint32_t coro_susp_point = self->private_impl.p_transform_io; switch (coro_susp_point) { WUFFS_BASE__COROUTINE_SUSPENSION_POINT_0; while (true) { + v_mark = ((uint64_t)(iop_a_dst - io0_a_dst)); { - wuffs_base__status t_0 = wuffs_lzma__decoder__do_transform_io(self, a_dst, a_src); - v_status = t_0; + if (a_dst) { + a_dst->meta.wi = ((size_t)(iop_a_dst - a_dst->data.ptr)); + } + wuffs_base__status t_0 = wuffs_lzma__decoder__do_transform_io(self, a_dst, a_src, a_workbuf); + v_dti_status = t_0; + if (a_dst) { + iop_a_dst = a_dst->data.ptr + a_dst->meta.wi; + } } - if ((v_status.repr == wuffs_base__suspension__short_read) && (a_src && a_src->meta.closed)) { + if ( ! wuffs_base__status__is_suspension(&v_dti_status)) { + status = v_dti_status; + if (wuffs_base__status__is_error(&status)) { + goto exit; + } else if (wuffs_base__status__is_suspension(&status)) { + status = wuffs_base__make_status(wuffs_base__error__cannot_return_a_suspension); + goto exit; + } + goto ok; + } else if ((v_dti_status.repr == wuffs_base__suspension__short_read) && (a_src && a_src->meta.closed)) { status = wuffs_base__make_status(wuffs_lzma__error__truncated_input); goto exit; } - status = v_status; + v_ah_status = wuffs_lzma__decoder__add_history(self, wuffs_private_impl__io__since(v_mark, ((uint64_t)(iop_a_dst - io0_a_dst)), io0_a_dst), a_workbuf); + if (wuffs_base__status__is_error(&v_ah_status)) { + status = v_ah_status; + goto exit; + } + status = v_dti_status; WUFFS_BASE__COROUTINE_SUSPENSION_POINT_MAYBE_SUSPEND(1); } @@ -52253,6 +52424,10 @@ wuffs_lzma__decoder__transform_io( goto exit; exit: + if (a_dst && a_dst->data.ptr) { + a_dst->meta.wi = ((size_t)(iop_a_dst - a_dst->data.ptr)); + } + if (wuffs_base__status__is_error(&status)) { self->private_impl.magic = WUFFS_BASE__DISABLED; } @@ -52266,7 +52441,8 @@ static wuffs_base__status wuffs_lzma__decoder__do_transform_io( wuffs_lzma__decoder* self, wuffs_base__io_buffer* a_dst, - wuffs_base__io_buffer* a_src) { + wuffs_base__io_buffer* a_src, + wuffs_base__slice_u8 a_workbuf) { wuffs_base__status status = wuffs_base__make_status(NULL); uint8_t v_header_byte = 0; @@ -52594,6 +52770,10 @@ wuffs_lzma__decoder__do_transform_io( goto exit; } } + while (((uint64_t)(a_workbuf.len)) < (((uint64_t)(self->private_impl.f_dict_size)) + 273u)) { + status = wuffs_base__make_status(wuffs_base__suspension__short_workbuf); + WUFFS_BASE__COROUTINE_SUSPENSION_POINT_MAYBE_SUSPEND(17); + } self->private_impl.f_lzma2_encoded_length_have = 0u; while (true) { v_dmark = ((uint64_t)(iop_a_dst - io0_a_dst)); @@ -52605,7 +52785,7 @@ wuffs_lzma__decoder__do_transform_io( if (a_src) { a_src->meta.ri = ((size_t)(iop_a_src - a_src->data.ptr)); } - wuffs_base__status t_8 = wuffs_lzma__decoder__decode_bitstream(self, a_dst, a_src); + wuffs_base__status t_8 = wuffs_lzma__decoder__decode_bitstream(self, a_dst, a_src, a_workbuf); v_status = t_8; if (a_dst) { iop_a_dst = a_dst->data.ptr + a_dst->meta.wi; @@ -52620,7 +52800,7 @@ wuffs_lzma__decoder__do_transform_io( break; } status = v_status; - WUFFS_BASE__COROUTINE_SUSPENSION_POINT_MAYBE_SUSPEND(17); + WUFFS_BASE__COROUTINE_SUSPENSION_POINT_MAYBE_SUSPEND(18); } if (self->private_impl.f_format_extension == 0u) { break; @@ -52660,7 +52840,8 @@ static wuffs_base__status wuffs_lzma__decoder__decode_bitstream( wuffs_lzma__decoder* self, wuffs_base__io_buffer* a_dst, - wuffs_base__io_buffer* a_src) { + wuffs_base__io_buffer* a_src, + wuffs_base__slice_u8 a_workbuf) { wuffs_base__status status = wuffs_base__make_status(NULL); uint32_t coro_susp_point = self->private_impl.p_decode_bitstream; @@ -52668,7 +52849,7 @@ wuffs_lzma__decoder__decode_bitstream( WUFFS_BASE__COROUTINE_SUSPENSION_POINT_0; WUFFS_BASE__COROUTINE_SUSPENSION_POINT(1); - status = wuffs_lzma__decoder__decode_bitstream_slow(self, a_dst, a_src); + status = wuffs_lzma__decoder__decode_bitstream_slow(self, a_dst, a_src, a_workbuf); if (status.repr) { goto suspend; } @@ -66797,7 +66978,7 @@ wuffs_xz__decoder__workbuf_len( return wuffs_base__utility__empty_range_ii_u64(); } - return wuffs_base__utility__make_range_ii_u64(0u, 0u); + return wuffs_lzma__decoder__workbuf_len(&self->private_data.f_lzma); } // -------- func xz.decoder.transform_io @@ -67097,7 +67278,7 @@ wuffs_xz__decoder__do_transform_io( if (a_src) { a_src->meta.ri = ((size_t)(iop_a_src - a_src->data.ptr)); } - wuffs_base__status t_4 = wuffs_lzma__decoder__transform_io(&self->private_data.f_lzma, a_dst, a_src, wuffs_base__utility__empty_slice_u8()); + wuffs_base__status t_4 = wuffs_lzma__decoder__transform_io(&self->private_data.f_lzma, a_dst, a_src, a_workbuf); v_status = t_4; if (a_dst) { iop_a_dst = a_dst->data.ptr + a_dst->meta.wi; diff --git a/std/lzma/decode_bitstream_slow.wuffs b/std/lzma/decode_bitstream_slow.wuffs index e9e8a6a7..c84f4774 100644 --- a/std/lzma/decode_bitstream_slow.wuffs +++ b/std/lzma/decode_bitstream_slow.wuffs @@ -18,7 +18,7 @@ // its LZMA implementation looks relatively verbose and repetitive (even though // it's equivalent to the C/C++ code, at the machine code level). -pri func decoder.decode_bitstream_slow?(dst: base.io_writer, src: base.io_reader) { +pri func decoder.decode_bitstream_slow?(dst: base.io_writer, src: base.io_reader, workbuf: roslice base.u8) { var c8 : base.u8 var bits : base.u32 @@ -68,6 +68,17 @@ pri func decoder.decode_bitstream_slow?(dst: base.io_writer, src: base.io_reader var index_large_dist : base.u32[..= 15] + // dist is part of the Lempel-Ziv (length, distance) pair. When dist is at + // or below the args.dst history length then the LZ copy can be resolved + // just from args.dst (and adj_dist and wb_index are unused). Otherwise, if + // e.g. dist was 100 and args.dst.history_length() was 40 then adj_dist + // would be 60: the LZ copy starts 60 bytes before the end of the + // args.workbuf ringbuffer. wb_index is that start index, equal to + // ((this.dict_workbuf_index ~mod- adj_dist) modulo this.dict_size). + var dist : base.u32 + var adj_dist : base.u32 + var wb_index : base.u64 + // Initialize the bitstream. c8 = args.src.read_u8?() if c8 <> 0x00 { @@ -804,24 +815,89 @@ pri func decoder.decode_bitstream_slow?(dst: base.io_writer, src: base.io_reader // AlgOve90 labelDoTheLZCopy: // AlgOve94 emitCopy(len, mrud[0]) + assert len >= 1 - if (((rep0 + 1) as base.u64) > (pos ~mod+ this.lzma2_prior_decoded_length)) or - (((rep0 + 1) as base.u64) > (this.dict_size as base.u64)) { + dist = rep0 + 1 + assert dist >= 1 + if ((dist as base.u64) > (pos ~mod+ this.lzma2_prior_decoded_length)) or + ((dist as base.u64) > (this.dict_size as base.u64)) { return "#bad distance" } pos ~mod+= len as base.u64 - while (len as base.u64) > args.dst.length(), + // 274 is at least (len + 1). A limited_copy_u32_from_slice call below + // passes "up_to: len + 1". It's simpler if it's always a full copy. + while 274 > args.dst.length(), inv len >= 1, - post (len as base.u64) <= args.dst.length(), + inv dist >= 1, + post 274 <= args.dst.length(), { yield? base."$short write" } endwhile - if ((rep0 + 1) as base.u64) > args.dst.history_length() { - return base."#insufficient history" + assert (len as base.u64) <= args.dst.length() via "a <= b: a <= c; c <= b"(c: 274) + + // Copy from the args.workbuf history, if necessary. + if (dist as base.u64) > args.dst.history_length() { + adj_dist = (((dist as base.u64) - args.dst.history_length()) & 0xFFFF_FFFF) as base.u32 + if adj_dist > this.dict_seen { + return "#bad distance" + } + wb_index = (this.dict_workbuf_index as base.u64) ~mod- (adj_dist as base.u64) + while wb_index >= 0x8000_0000_0000_0000, + inv dist >= 1, + { + wb_index ~mod+= this.dict_size as base.u64 + } endwhile + if wb_index > args.workbuf.length() { + return "#internal error: inconsistent dictionary state" + } + + if len < adj_dist { + // Copy len bytes of history and one more byte for the later + // half of the two-byte cusp, all from args.workbuf. + args.dst.limited_copy_u32_from_slice!(up_to: len + 1, s: args.workbuf[wb_index ..]) + if not args.dst.can_undo_byte() { + return "#internal error: inconsistent dictionary state" + } + match_byte = args.dst.peek_undo_byte() + args.dst.undo_byte!() + if not args.dst.can_undo_byte() { + return "#internal error: inconsistent dictionary state" + } + prev_byte = args.dst.peek_undo_byte() + continue.outer + } else if len == adj_dist { + // Like above, copy len bytes of history and one more byte for + // the later half of the two-byte cusp, but the copy-source + // crosses over from args.workbuf to args.dst. + args.dst.limited_copy_u32_from_slice!(up_to: len, s: args.workbuf[wb_index ..]) + args.dst.limited_copy_u32_from_history!(up_to: 1, distance: dist) + if not args.dst.can_undo_byte() { + return "#internal error: inconsistent dictionary state" + } + match_byte = args.dst.peek_undo_byte() + args.dst.undo_byte!() + if not args.dst.can_undo_byte() { + return "#internal error: inconsistent dictionary state" + } + prev_byte = args.dst.peek_undo_byte() + continue.outer + } + args.dst.limited_copy_u32_from_slice!(up_to: adj_dist, s: args.workbuf[wb_index ..]) + len -= adj_dist + assert len >= 1 + if ((len as base.u64) > args.dst.length()) or + ((dist as base.u64) > args.dst.history_length()) { + return "#internal error: inconsistent dictionary state" + } } - assert (rep0 + 1) >= 1 + + // Copy from the args.dst history. + assert len >= 1 + assert dist >= 1 + assert (len as base.u64) <= args.dst.length() + assert (dist as base.u64) <= args.dst.history_length() match_cusp = args.dst.limited_copy_u32_from_history_fast_return_cusp!( - up_to: len, distance: (rep0 + 1)) + up_to: len, distance: dist) match_byte = (match_cusp >> 8) as base.u8 prev_byte = (match_cusp & 0xFF) as base.u8 } endwhile.outer diff --git a/std/lzma/decode_lzma.wuffs b/std/lzma/decode_lzma.wuffs index 71878e37..f429199f 100644 --- a/std/lzma/decode_lzma.wuffs +++ b/std/lzma/decode_lzma.wuffs @@ -18,9 +18,10 @@ pub status "#truncated input" pub status "#unsupported decoded length" pub status "#unsupported properties" -pub const DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE : base.u64 = 0xFFFF_FFFF +pri status "#internal error: inconsistent dictionary state" -pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0 +pub const DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE : base.u64 = 0 +pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0xFFFF_FFFF + 273 pub struct decoder? implements base.io_transformer( // In theory, lc ranges in [0 ..= 8]. In practice, to keep the @@ -36,8 +37,19 @@ pub struct decoder? implements base.io_transformer( format_extension : base.u32, + // dict_size is the nominal size of the args.workbuf ringbuffer that + // holds historical decoded bytes. The workbuf_len is a little larger: + // see "ML" below. dict_size : base.u32, + // dict_workbuf_index indexes the args.workbuf ringbuffer, pointing to + // where add_history will next write to. + dict_workbuf_index : base.u32, + + // dict_seen is the cumulative sum (saturating at 0xFFFF_FFFF) of the + // number of bytes passed to add_history. + dict_seen : base.u32, + decoded_length : base.u64, lzma2_prior_decoded_length : base.u64, @@ -166,6 +178,79 @@ pub struct decoder? implements base.io_transformer( // https://github.com/jljusten/LZMA-SDK/blob/f287b63f6bb8b88e281f18a9295340c732245167/Source/LzmaDecode.h#L48 ) +pri func decoder.add_history!(hist: slice base.u8, workbuf: slice base.u8) base.status { + var dict_workbuf_index : base.u64[..= 0xFFFF_FFFF] + var dict_size : base.u64[..= 0xFFFF_FFFF] + var hist_length : base.u64 + var s : slice base.u8 + var n_copied : base.u64 + var n : base.u64 + + dict_workbuf_index = this.dict_workbuf_index as base.u64 + dict_size = this.dict_size as base.u64 + if args.hist.length() == 0 { + return ok + } + + // 273 is ML, where ML is the Maximum Length of an LZMA Lempel-Ziv (length, + // distance) pair. After an add_history call, args.workbuf[dict_size .. + // (dict_size + ML)] duplicates args.workbuf[.. ML]. This simplifies + // copying up to (ML + 1) bytes from the ringbuffer, as there is no need to + // split the copy around the dict_size boundary. + // + // This technique is similar to the one used in the std/deflate package + // (although that package uses a different value of ML, 258). That package + // also extends the nominal dict_size by (ML - 1), not ML as is done here, + // because std/deflate copies up to ML bytes, but std/lzma copies up to (ML + // + 1) bytes. std/lzma also needs to know the later half of the two-byte + // cusp: the first byte after the copied history. + if args.workbuf.length() < (dict_size + 273) { + return base."#bad workbuf length" + } + assert args.workbuf.length() >= dict_size via "a >= b: a >= (b + c); 0 <= c"(c: 273) + assert dict_size <= args.workbuf.length() via "a <= b: b >= a"() + + hist_length = args.hist.length() + if hist_length > 0xFFFF_FFFF { + this.dict_seen = 0xFFFF_FFFF + } else { + this.dict_seen ~sat+= hist_length as base.u32 + } + + s = args.hist + if s.length() >= dict_size { + s = s.suffix(up_to: dict_size) + args.workbuf.copy_from_slice!(s: s) + this.dict_workbuf_index = 0 + + } else if dict_workbuf_index > dict_size { + return "#internal error: inconsistent dictionary state" + + } else { + n_copied = args.workbuf[dict_workbuf_index .. dict_size].copy_from_slice!(s: s) + if n_copied < s.length() { + n = args.workbuf.copy_from_slice!(s: s[n_copied ..]) + this.dict_workbuf_index = (n & 0xFFFF_FFFF) as base.u32 + } else { + n = dict_workbuf_index ~mod+ n_copied + if n < dict_size { + assert n < 0xFFFF_FFFF via "a < b: a < c; c <= b"(c: dict_size) + this.dict_workbuf_index = n as base.u32 + } else { + this.dict_workbuf_index = 0 + } + } + } + + // Have the tail duplicate the head. + if (273 > dict_size) or (dict_size > args.workbuf.length()) { + return "#internal error: inconsistent dictionary state" + } + assert 273 <= args.workbuf.length() via "a <= b: a <= c; c <= b"(c: dict_size) + args.workbuf[dict_size ..].copy_from_slice!(s: args.workbuf[.. 273]) + return ok +} + pub func decoder.get_quirk(key: base.u32) base.u64 { if args.key == QUIRK_FORMAT_EXTENSION { return this.format_extension as base.u64 @@ -196,29 +281,41 @@ pub func decoder.set_quirk!(key: base.u32, value: base.u64) base.status { } pub func decoder.dst_history_retain_length() base.optional_u63 { - if this.dict_size == 0 { - return this.util.make_optional_u63(has_value: false, value: 0) - } - return this.util.make_optional_u63(has_value: true, value: this.dict_size as base.u64) + return this.util.make_optional_u63(has_value: true, value: 0) } pub func decoder.workbuf_len() base.range_ii_u64 { - return this.util.make_range_ii_u64(min_incl: 0, max_incl: 0) + var m : base.u64 + + if this.dict_size == 0 { + return this.util.make_range_ii_u64(min_incl: 0, max_incl: 0) + } + m = (this.dict_size as base.u64) + 273 + return this.util.make_range_ii_u64(min_incl: m, max_incl: m) } pub func decoder.transform_io?(dst: base.io_writer, src: base.io_reader, workbuf: slice base.u8) { - var status : base.status + var mark : base.u64 + var dti_status : base.status + var ah_status : base.status while true { - status =? this.do_transform_io?(dst: args.dst, src: args.src) - if (status == base."$short read") and args.src.is_closed() { + mark = args.dst.mark() + dti_status =? this.do_transform_io?(dst: args.dst, src: args.src, workbuf: args.workbuf) + if not dti_status.is_suspension() { + return dti_status + } else if (dti_status == base."$short read") and args.src.is_closed() { return "#truncated input" } - yield? status + ah_status = this.add_history!(hist: args.dst.since(mark: mark), workbuf: args.workbuf) + if ah_status.is_error() { + return ah_status + } + yield? dti_status } endwhile } -pri func decoder.do_transform_io?(dst: base.io_writer, src: base.io_reader) { +pri func decoder.do_transform_io?(dst: base.io_writer, src: base.io_reader, workbuf: roslice base.u8) { var header_byte : base.u8 var c32 : base.u32 var prop_byte : base.u8 @@ -348,11 +445,15 @@ pri func decoder.do_transform_io?(dst: base.io_writer, src: base.io_reader) { } } + while args.workbuf.length() < ((this.dict_size as base.u64) + 273) { + yield? base."$short workbuf" + } endwhile + this.lzma2_encoded_length_have = 0 while true { dmark = args.dst.mark() smark = args.src.mark() - status =? this.decode_bitstream?(dst: args.dst, src: args.src) + status =? this.decode_bitstream?(dst: args.dst, src: args.src, workbuf: args.workbuf) this.lzma2_encoded_length_have ~sat+= args.src.count_since(mark: smark) this.lzma2_prior_decoded_length ~sat+= args.dst.count_since(mark: dmark) if status.is_ok() { @@ -369,8 +470,8 @@ pri func decoder.do_transform_io?(dst: base.io_writer, src: base.io_reader) { } endwhile } -pri func decoder.decode_bitstream?(dst: base.io_writer, src: base.io_reader) { - this.decode_bitstream_slow?(dst: args.dst, src: args.src) +pri func decoder.decode_bitstream?(dst: base.io_writer, src: base.io_reader, workbuf: roslice base.u8) { + this.decode_bitstream_slow?(dst: args.dst, src: args.src, workbuf: args.workbuf) } // initialize_probs sets all of the probs_foo[i][j] values to 1024. diff --git a/std/xz/decode_xz.wuffs b/std/xz/decode_xz.wuffs index 86bb75c2..e4b643e2 100644 --- a/std/xz/decode_xz.wuffs +++ b/std/xz/decode_xz.wuffs @@ -26,9 +26,8 @@ pub status "#truncated input" pub status "#unsupported checksum algorithm" pub status "#unsupported filter" -pub const DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE : base.u64 = 0xFFFF_FFFF - -pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0 +pub const DECODER_DST_HISTORY_RETAIN_LENGTH_MAX_INCL_WORST_CASE : base.u64 = 0 +pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0xFFFF_FFFF + 273 pub struct decoder? implements base.io_transformer( filters : array[3] base.u32, @@ -79,7 +78,7 @@ pub func decoder.dst_history_retain_length() base.optional_u63 { } pub func decoder.workbuf_len() base.range_ii_u64 { - return this.util.make_range_ii_u64(min_incl: 0, max_incl: 0) + return this.lzma.workbuf_len() } pub func decoder.transform_io?(dst: base.io_writer, src: base.io_reader, workbuf: slice base.u8) { @@ -177,7 +176,7 @@ pri func decoder.do_transform_io?(dst: base.io_writer, src: base.io_reader, work while true { dmark = args.dst.mark() smark = args.src.mark() - status =? this.lzma.transform_io?(dst: args.dst, src: args.src, workbuf: this.util.empty_slice_u8()) + status =? this.lzma.transform_io?(dst: args.dst, src: args.src, workbuf: args.workbuf) compressed_size ~mod+= args.src.count_since(mark: smark) uncompressed_size ~mod+= args.dst.count_since(mark: dmark) diff --git a/test/c/std/xz.c b/test/c/std/xz.c index c0107513..e5d96a19 100644 --- a/test/c/std/xz.c +++ b/test/c/std/xz.c @@ -137,14 +137,12 @@ test_wuffs_xz_decode_enwik5() { UINT64_MAX); } -/* TODO const char* // test_wuffs_xz_decode_one_byte_reads() { CHECK_FOCUS(__func__); return do_test_io_buffers(wuffs_xz_decode, &g_xz_romeo_delta1_gt, UINT64_MAX, 1); } -*/ const char* // test_wuffs_xz_decode_romeo() { @@ -203,7 +201,7 @@ proc g_tests[] = { test_wuffs_xz_decode_enwik5, test_wuffs_xz_decode_interface, - // TODO test_wuffs_xz_decode_one_byte_reads, + test_wuffs_xz_decode_one_byte_reads, test_wuffs_xz_decode_romeo, #ifdef WUFFS_MIMIC