X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fdata%2Fsys-file-writer.c;h=c94d13d85994037042ea236ac85c89600c50f339;hb=refs%2Fheads%2Fctables7;hp=fe96d3cbba86be2212ac81b93788e892a0e3fd4a;hpb=261869b71aa60b8974c4a6b98e35b74af5d11de5;p=pspp diff --git a/src/data/sys-file-writer.c b/src/data/sys-file-writer.c index fe96d3cbba..c94d13d859 100644 --- a/src/data/sys-file-writer.c +++ b/src/data/sys-file-writer.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 1997-9, 2000, 2006, 2007, 2009, 2010, 2011 Free Software Foundation, Inc. + Copyright (C) 1997-2000, 2006-2014 Free Software Foundation, Inc. This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -25,6 +25,7 @@ #include #include #include +#include #include "data/attributes.h" #include "data/case.h" @@ -32,7 +33,6 @@ #include "data/casewriter.h" #include "data/dictionary.h" #include "data/file-handle-def.h" -#include "data/file-name.h" #include "data/format.h" #include "data/make-file.h" #include "data/missing-values.h" @@ -72,28 +72,42 @@ struct sfm_writer FILE *file; /* File stream. */ struct replace_file *rf; /* Ticket for replacing output file. */ - bool compress; /* 1=compressed, 0=not compressed. */ - casenumber case_cnt; /* Number of cases written so far. */ + enum any_compression compression; + casenumber n_cases; /* Number of cases written so far. */ + uint8_t space; /* ' ' in the file's character encoding. */ - /* Compression buffering. + /* Simple compression buffering. - Compressed data is output as groups of 8 1-byte opcodes - followed by up to 8 (depending on the opcodes) 8-byte data - items. Data items and opcodes arrive at the same time but - must be reordered for writing to disk, thus a small amount - of buffering here. */ - uint8_t opcodes[8]; /* Buffered opcodes. */ - int opcode_cnt; /* Number of buffered opcodes. */ - uint8_t data[8][8]; /* Buffered data. */ - int data_cnt; /* Number of buffered data items. */ + Compressed data is output as a series of 8-byte elements, with 1 to 9 + such elements clustered together. The first element in a cluster is 8 + 1-byte opcodes. Some opcodes call for an additional element in the + cluster (hence, if there are eight such opcodes, then the cluster + contains a full 9 elements). + + cbuf[] holds a cluster at a time. */ + uint8_t cbuf[9][8]; + int n_opcodes; /* Number of opcodes in cbuf[0] so far. */ + int n_elements; /* Number of elements in cbuf[] so far. */ + + /* ZLIB compression. */ + z_stream zstream; /* ZLIB deflater. */ + off_t zstart; + struct zblock *blocks; + size_t n_blocks, allocated_blocks; /* Variables. */ struct sfm_var *sfm_vars; /* Variables. */ - size_t sfm_var_cnt; /* Number of variables. */ - size_t segment_cnt; /* Number of variables including extra segments + size_t sfm_n_vars; /* Number of variables. */ + size_t n_segments; /* Number of variables including extra segments for long string variables. */ }; +struct zblock + { + unsigned int uncompressed_size; + unsigned int compressed_size; + }; + static const struct casewriter_class sys_file_casewriter_class; static void write_header (struct sfm_writer *, const struct dictionary *); @@ -115,6 +129,8 @@ static void write_vls_length_table (struct sfm_writer *w, static void write_long_string_value_labels (struct sfm_writer *, const struct dictionary *); +static void write_long_string_missing_values (struct sfm_writer *, + const struct dictionary *); static void write_mrsets (struct sfm_writer *, const struct dictionary *, bool pre_v14); @@ -130,6 +146,7 @@ static void write_variable_attributes (struct sfm_writer *, const struct dictionary *); static void write_int (struct sfm_writer *, int32_t); +static void write_int64 (struct sfm_writer *, int64_t); static inline void convert_double_to_output_format (double, uint8_t[8]); static void write_float (struct sfm_writer *, double); static void write_string (struct sfm_writer *, const char *, size_t); @@ -152,6 +169,10 @@ static void put_cmp_opcode (struct sfm_writer *, uint8_t); static void put_cmp_number (struct sfm_writer *, double); static void put_cmp_string (struct sfm_writer *, const void *, size_t); +static bool start_zstream (struct sfm_writer *); +static void finish_zstream (struct sfm_writer *); +static void write_ztrailer (struct sfm_writer *); + static bool write_error (const struct sfm_writer *); static bool close_writer (struct sfm_writer *); @@ -160,8 +181,10 @@ struct sfm_write_options sfm_writer_default_options (void) { struct sfm_write_options opts; + opts.compression = (settings_get_scompression () + ? ANY_COMP_SIMPLE + : ANY_COMP_NONE); opts.create_writeable = true; - opts.compress = settings_get_scompression (); opts.version = 3; return opts; } @@ -176,7 +199,7 @@ struct casewriter * sfm_open_writer (struct file_handle *fh, struct dictionary *d, struct sfm_write_options opts) { - struct sfm_writer *w; + struct encoding_info encoding_info; mode_t mode; int i; @@ -189,23 +212,30 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, } /* Create and initialize writer. */ - w = xmalloc (sizeof *w); + struct sfm_writer *w = XZALLOC (struct sfm_writer); w->fh = fh_ref (fh); w->lock = NULL; w->file = NULL; w->rf = NULL; - w->compress = opts.compress; - w->case_cnt = 0; + /* Use the requested compression, except that no EBCDIC-based ZLIB compressed + files have been observed, so drop back to simple compression for those + files. */ + w->compression = opts.compression; + if (w->compression == ANY_COMP_ZLIB + && is_encoding_ebcdic_compatible (dict_get_encoding (d))) + w->compression = ANY_COMP_SIMPLE; + + w->n_cases = 0; - w->opcode_cnt = w->data_cnt = 0; + w->n_opcodes = w->n_elements = 0; + memset (w->cbuf[0], 0, 8); /* Figure out how to map in-memory case data to on-disk case data. Also count the number of segments. Very long strings occupy multiple segments, otherwise each variable only takes one segment. */ - w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars, - &w->sfm_var_cnt); + w->n_segments = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars, &w->sfm_n_vars); /* Open file handle as an exclusive writer. */ /* TRANSLATORS: this fragment will be interpolated into @@ -218,8 +248,7 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, mode = 0444; if (opts.create_writeable) mode |= 0222; - w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode, - &w->file, NULL); + w->rf = replace_file_start (fh, "wb", mode, &w->file); if (w->rf == NULL) { msg (ME, _("Error opening `%s' for writing as a system file: %s."), @@ -227,17 +256,20 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, goto error; } + get_encoding_info (&encoding_info, dict_get_encoding (d)); + w->space = encoding_info.space[0]; + /* Write the file header. */ write_header (w, d); /* Write basic variable info. */ short_names_assign (d); - for (i = 0; i < dict_get_var_cnt (d); i++) + for (i = 0; i < dict_get_n_vars (d); i++) write_variable (w, dict_get_var (d, i)); write_value_labels (w, d); - if (dict_get_document_line_cnt (d) > 0) + if (dict_get_document_n_lines (d) > 0) write_documents (w, d); write_integer_info_record (w, d); @@ -253,10 +285,14 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, write_vls_length_table (w, d); write_long_string_value_labels (w, d); + write_long_string_missing_values (w, d); - if (attrset_count (dict_get_attributes (d))) - write_data_file_attributes (w, d); - write_variable_attributes (w, d); + if (opts.version >= 3) + { + if (attrset_count (dict_get_attributes (d))) + write_data_file_attributes (w, d); + write_variable_attributes (w, d); + } write_mrsets (w, d, false); @@ -266,6 +302,20 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, write_int (w, 999); write_int (w, 0); + if (w->compression == ANY_COMP_ZLIB) + { + w->zstream.zalloc = Z_NULL; + w->zstream.zfree = Z_NULL; + w->zstream.opaque = Z_NULL; + w->zstart = ftello (w->file); + + write_int64 (w, w->zstart); + write_int64 (w, 0); + write_int64 (w, 0); + + start_zstream (w); + } + if (write_error (w)) goto error; @@ -297,7 +347,7 @@ calc_oct_idx (const struct dictionary *d, struct variable *target_var) int i; oct_idx = 0; - for (i = 0; i < dict_get_var_cnt (d); i++) + for (i = 0; i < dict_get_n_vars (d); i++) { struct variable *var = dict_get_var (d, i); if (var == target_var) @@ -311,16 +361,20 @@ calc_oct_idx (const struct dictionary *d, struct variable *target_var) static void write_header (struct sfm_writer *w, const struct dictionary *d) { + const char *dict_encoding = dict_get_encoding (d); char prod_name[61]; - char creation_date[10]; - char creation_time[9]; const char *file_label; struct variable *weight; time_t t; /* Record-type code. */ - write_string (w, "$FL2", 4); + if (is_encoding_ebcdic_compatible (dict_encoding)) + write_string (w, EBCDIC_MAGIC, 4); + else if (w->compression == ANY_COMP_ZLIB) + write_string (w, ASCII_ZMAGIC, 4); + else + write_string (w, ASCII_MAGIC, 4); /* Product identification. */ snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s", @@ -334,7 +388,9 @@ write_header (struct sfm_writer *w, const struct dictionary *d) write_int (w, calc_oct_idx (d, NULL)); /* Compressed? */ - write_int (w, w->compress); + write_int (w, (w->compression == ANY_COMP_NONE ? 0 + : w->compression == ANY_COMP_SIMPLE ? 1 + : 2)); /* Weight variable. */ weight = dict_get_weight (d); @@ -349,10 +405,11 @@ write_header (struct sfm_writer *w, const struct dictionary *d) write_float (w, COMPRESSION_BIAS); /* Creation date and time. */ + char *creation_date, *creation_time; if (time (&t) == (time_t) -1) { - strcpy (creation_date, "01 Jan 70"); - strcpy (creation_time, "00:00:00"); + creation_date = xstrdup ("01 Jan 70"); + creation_time = xstrdup ("00:00:00"); } else { @@ -369,13 +426,14 @@ write_header (struct sfm_writer *w, const struct dictionary *d) int min = rerange (tmp->tm_min + 1); int sec = rerange (tmp->tm_sec + 1); - snprintf (creation_date, sizeof creation_date, - "%02d %s %02d", day, month_name[mon - 1], year); - snprintf (creation_time, sizeof creation_time, - "%02d:%02d:%02d", hour - 1, min - 1, sec - 1); + creation_date = xasprintf ("%02d %s %02d", + day, month_name[mon - 1], year); + creation_time = xasprintf ("%02d:%02d:%02d", hour - 1, min - 1, sec - 1); } write_utf8_string (w, dict_encoding, creation_date, 9); write_utf8_string (w, dict_encoding, creation_time, 8); + free (creation_time); + free (creation_date); /* File label. */ file_label = dict_get_label (d); @@ -427,10 +485,9 @@ static void write_variable (struct sfm_writer *w, const struct variable *v) { int width = var_get_width (v); - int segment_cnt = sfm_width_to_segments (width); + int n_segments = sfm_width_to_segments (width); int seg0_width = sfm_segment_alloc_width (width, 0); const char *encoding = var_get_encoding (v); - struct missing_values mv; int i; /* Record type. */ @@ -444,14 +501,21 @@ write_variable (struct sfm_writer *w, const struct variable *v) /* Number of missing values. If there is a range, then the range counts as 2 missing values and causes the number to be - negated. */ - mv_copy (&mv, var_get_missing_values (v)); - if (mv_get_width (&mv) > 8) - mv_resize (&mv, 8); - if (mv_has_range (&mv)) - write_int (w, -2 - mv_n_values (&mv)); + negated. + + Missing values for long string variables are written in a separate + record. */ + enum { MAX_SHORT_STRING = 8 }; + if (width <= MAX_SHORT_STRING) + { + const struct missing_values *mv = var_get_missing_values (v); + if (mv_has_range (mv)) + write_int (w, -2 - mv_n_values (mv)); + else + write_int (w, mv_n_values (mv)); + } else - write_int (w, mv_n_values (&mv)); + write_int (w, 0); /* Print and write formats. */ write_format (w, *var_get_print_format (v), seg0_width); @@ -474,20 +538,24 @@ write_variable (struct sfm_writer *w, const struct variable *v) } /* Write the missing values, if any, range first. */ - if (mv_has_range (&mv)) + if (width <= MAX_SHORT_STRING) { - double x, y; - mv_get_range (&mv, &x, &y); - write_float (w, x); - write_float (w, y); + const struct missing_values *mv = var_get_missing_values (v); + if (mv_has_range (mv)) + { + double x, y; + mv_get_range (mv, &x, &y); + write_float (w, x); + write_float (w, y); + } + for (i = 0; i < mv_n_values (mv); i++) + write_value (w, mv_get_value (mv, i), width); } - for (i = 0; i < mv_n_values (&mv); i++) - write_value (w, mv_get_value (&mv, i), mv_get_width (&mv)); write_variable_continuation_records (w, seg0_width); /* Write additional segments for very long string variables. */ - for (i = 1; i < segment_cnt; i++) + for (i = 1; i < n_segments; i++) { int seg_width = sfm_segment_alloc_width (width, i); struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0); @@ -502,8 +570,6 @@ write_variable (struct sfm_writer *w, const struct variable *v) write_variable_continuation_records (w, seg_width); } - - mv_destroy (&mv); } /* Writes the value labels to system file W. @@ -532,7 +598,7 @@ write_value_labels (struct sfm_writer *w, const struct dictionary *d) hmap_init (&same_sets); idx = 0; - for (i = 0; i < dict_get_var_cnt (d); i++) + for (i = 0; i < dict_get_n_vars (d); i++) { struct variable *v = dict_get_var (d, i); @@ -648,14 +714,14 @@ put_attrset (struct string *string, const struct attrset *attrs) struct attrset_iterator i; for (attr = attrset_first (attrs, &i); attr != NULL; - attr = attrset_next (attrs, &i)) + attr = attrset_next (attrs, &i)) { size_t n_values = attribute_get_n_values (attr); size_t j; ds_put_cstr (string, attribute_get_name (attr)); ds_put_byte (string, '('); - for (j = 0; j < n_values; j++) + for (j = 0; j < n_values; j++) ds_put_format (string, "'%s'\n", attribute_get_value (attr, j)); ds_put_byte (string, ')'); } @@ -671,25 +737,67 @@ write_data_file_attributes (struct sfm_writer *w, ds_destroy (&s); } +static void +add_role_attribute (enum var_role role, struct attrset *attrs) +{ + struct attribute *attr; + const char *s; + + switch (role) + { + case ROLE_INPUT: + default: + s = "0"; + break; + + case ROLE_TARGET: + s = "1"; + break; + + case ROLE_BOTH: + s = "2"; + break; + + case ROLE_NONE: + s = "3"; + break; + + case ROLE_PARTITION: + s = "4"; + break; + + case ROLE_SPLIT: + s = "5"; + break; + } + attrset_delete (attrs, "$@Role"); + + attr = attribute_create ("$@Role"); + attribute_add_value (attr, s); + attrset_add (attrs, attr); +} + static void write_variable_attributes (struct sfm_writer *w, const struct dictionary *d) { struct string s = DS_EMPTY_INITIALIZER; - size_t n_vars = dict_get_var_cnt (d); + size_t n_vars = dict_get_n_vars (d); size_t n_attrsets = 0; size_t i; for (i = 0; i < n_vars; i++) - { + { struct variable *v = dict_get_var (d, i); - struct attrset *attrs = var_get_attributes (v); - if (attrset_count (attrs)) - { - if (n_attrsets++) - ds_put_byte (&s, '/'); - ds_put_format (&s, "%s:", var_get_name (v)); - put_attrset (&s, attrs); - } + struct attrset attrs; + + attrset_clone (&attrs, var_get_attributes (v)); + + add_role_attribute (var_get_role (v), &attrs); + if (n_attrsets++) + ds_put_byte (&s, '/'); + ds_put_format (&s, "%s:", var_get_name (v)); + put_attrset (&s, &attrs); + attrset_destroy (&attrs); } if (n_attrsets) write_utf8_record (w, dict_get_encoding (d), &s, 18); @@ -708,6 +816,12 @@ write_mrsets (struct sfm_writer *w, const struct dictionary *dict, size_t n_mrsets; size_t i; + if (is_encoding_ebcdic_compatible (encoding)) + { + /* FIXME. */ + return; + } + n_mrsets = dict_get_n_mrsets (dict); if (n_mrsets == 0) return; @@ -738,8 +852,7 @@ write_mrsets (struct sfm_writer *w, const struct dictionary *dict, if (mrset->width == 0) counted = xasprintf ("%.0f", mrset->counted.f); else - counted = xmemdup0 (value_str (&mrset->counted, mrset->width), - mrset->width); + counted = xmemdup0 (mrset->counted.s, mrset->width); ds_put_format (&s, "%zu %s", strlen (counted), counted); free (counted); } @@ -759,11 +872,12 @@ write_mrsets (struct sfm_writer *w, const struct dictionary *dict, for (j = 0; j < mrset->n_vars; j++) { const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0); + char *lower_name_utf8 = utf8_to_lower (short_name_utf8); char *short_name = recode_string (encoding, "UTF-8", - short_name_utf8, -1); - str_lowercase (short_name); + lower_name_utf8, -1); ds_put_format (&s, " %s", short_name); free (short_name); + free (lower_name_utf8); } ds_put_byte (&s, '\n'); } @@ -783,13 +897,13 @@ write_variable_display_parameters (struct sfm_writer *w, write_int (w, 7); /* Record type. */ write_int (w, 11); /* Record subtype. */ write_int (w, 4); /* Data item (int32) size. */ - write_int (w, w->segment_cnt * 3); /* Number of data items. */ + write_int (w, w->n_segments * 3); /* Number of data items. */ - for (i = 0; i < dict_get_var_cnt (dict); ++i) + for (i = 0; i < dict_get_n_vars (dict); ++i) { struct variable *v = dict_get_var (dict, i); int width = var_get_width (v); - int segment_cnt = sfm_width_to_segments (width); + int n_segments = sfm_width_to_segments (width); int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1 : var_get_measure (v) == MEASURE_ORDINAL ? 2 : 3); @@ -798,7 +912,7 @@ write_variable_display_parameters (struct sfm_writer *w, : 2); int i; - for (i = 0; i < segment_cnt; i++) + for (i = 0; i < n_segments; i++) { int width_left = width - sfm_segment_effective_offset (width, i); write_int (w, measure); @@ -818,7 +932,7 @@ write_vls_length_table (struct sfm_writer *w, int i; ds_init_empty (&map); - for (i = 0; i < dict_get_var_cnt (dict); ++i) + for (i = 0; i < dict_get_n_vars (dict); ++i) { const struct variable *v = dict_get_var (dict, i); if (sfm_width_to_segments (var_get_width (v)) > 1) @@ -830,14 +944,13 @@ write_vls_length_table (struct sfm_writer *w, ds_destroy (&map); } - static void write_long_string_value_labels (struct sfm_writer *w, const struct dictionary *dict) { - size_t n_vars = dict_get_var_cnt (dict); + const char *encoding = dict_get_encoding (dict); + size_t n_vars = dict_get_n_vars (dict); size_t size, i; - off_t start UNUSED; /* Figure out the size in advance. */ size = 0; @@ -845,7 +958,6 @@ write_long_string_value_labels (struct sfm_writer *w, { struct variable *var = dict_get_var (dict, i); const struct val_labs *val_labs = var_get_value_labels (var); - const char *encoding = var_get_encoding (var); int width = var_get_width (var); const struct val_lab *val_lab; @@ -870,12 +982,10 @@ write_long_string_value_labels (struct sfm_writer *w, write_int (w, 1); /* Data item (byte) size. */ write_int (w, size); /* Number of data items. */ - start = ftello (w->file); for (i = 0; i < n_vars; i++) { struct variable *var = dict_get_var (dict, i); const struct val_labs *val_labs = var_get_value_labels (var); - const char *encoding = var_get_encoding (var); int width = var_get_width (var); const struct val_lab *val_lab; char *var_name; @@ -897,8 +1007,7 @@ write_long_string_value_labels (struct sfm_writer *w, size_t len; write_int (w, width); - write_bytes (w, value_str (val_lab_get_value (val_lab), width), - width); + write_bytes (w, val_lab_get_value (val_lab)->s, width); label = recode_string (var_get_encoding (var), "UTF-8", val_lab_get_escaped_label (val_lab), -1); @@ -908,7 +1017,68 @@ write_long_string_value_labels (struct sfm_writer *w, free (label); } } - assert (ftello (w->file) == start + size); +} + +static void +write_long_string_missing_values (struct sfm_writer *w, + const struct dictionary *dict) +{ + const char *encoding = dict_get_encoding (dict); + size_t n_vars = dict_get_n_vars (dict); + size_t size, i; + + /* Figure out the size in advance. */ + size = 0; + for (i = 0; i < n_vars; i++) + { + struct variable *var = dict_get_var (dict, i); + const struct missing_values *mv = var_get_missing_values (var); + int width = var_get_width (var); + + if (mv_is_empty (mv) || width < 9) + continue; + + size += 4; + size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1); + size += 1; + size += mv_n_values (mv) * (4 + 8); + } + if (size == 0) + return; + + write_int (w, 7); /* Record type. */ + write_int (w, 22); /* Record subtype */ + write_int (w, 1); /* Data item (byte) size. */ + write_int (w, size); /* Number of data items. */ + + for (i = 0; i < n_vars; i++) + { + struct variable *var = dict_get_var (dict, i); + const struct missing_values *mv = var_get_missing_values (var); + int width = var_get_width (var); + uint8_t n_missing_values; + char *var_name; + int j; + + if (mv_is_empty (mv) || width < 9) + continue; + + var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1); + write_int (w, strlen (var_name)); + write_bytes (w, var_name, strlen (var_name)); + free (var_name); + + n_missing_values = mv_n_values (mv); + write_bytes (w, &n_missing_values, 1); + + for (j = 0; j < n_missing_values; j++) + { + const union value *value = mv_get_value (mv, j); + + write_int (w, 8); + write_bytes (w, value->s, 8); + } + } } static void @@ -935,7 +1105,7 @@ write_longvar_table (struct sfm_writer *w, const struct dictionary *dict) size_t i; ds_init_empty (&map); - for (i = 0; i < dict_get_var_cnt (dict); i++) + for (i = 0; i < dict_get_n_vars (dict); i++) { struct variable *v = dict_get_var (dict, i); if (i) @@ -952,6 +1122,7 @@ static void write_integer_info_record (struct sfm_writer *w, const struct dictionary *d) { + const char *dict_encoding = dict_get_encoding (d); int version_component[3]; int float_format; int codepage; @@ -973,13 +1144,21 @@ write_integer_info_record (struct sfm_writer *w, abort (); /* Choose codepage. */ - codepage = sys_get_codepage_from_encoding (dict_get_encoding (d)); + codepage = sys_get_codepage_from_encoding (dict_encoding); if (codepage == 0) { - /* Default to "7-bit ASCII" if the codepage number is unknown, because + /* The codepage is unknown. Choose a default. + + For an EBCDIC-compatible encoding, use the value for EBCDIC. + + For an ASCII-compatible encoding, default to "7-bit ASCII", because many files use this codepage number regardless of their actual - encoding. */ - codepage = 2; + encoding. + */ + if (is_encoding_ascii_compatible (dict_encoding)) + codepage = 2; + else if (is_encoding_ebcdic_compatible (dict_encoding)) + codepage = 1; } /* Write record. */ @@ -1024,9 +1203,9 @@ sys_file_casewriter_write (struct casewriter *writer, void *w_, return; } - w->case_cnt++; + w->n_cases++; - if (!w->compress) + if (w->compression == ANY_COMP_NONE) write_case_uncompressed (w, c); else write_case_compressed (w, c); @@ -1064,8 +1243,12 @@ close_writer (struct sfm_writer *w) if (w->file != NULL) { /* Flush buffer. */ - if (w->opcode_cnt > 0) - flush_compressed (w); + flush_compressed (w); + if (w->compression == ANY_COMP_ZLIB) + { + finish_zstream (w); + write_ztrailer (w); + } fflush (w->file); ok = !write_error (w); @@ -1073,9 +1256,9 @@ close_writer (struct sfm_writer *w) /* Seek back to the beginning and update the number of cases. This is just a courtesy to later readers, so there's no need to check return values or report errors. */ - if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET)) + if (ok && w->n_cases <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET)) { - write_int (w, w->case_cnt); + write_int (w, w->n_cases); clearerr (w->file); } @@ -1090,6 +1273,8 @@ close_writer (struct sfm_writer *w) ok = false; } + free (w->blocks); + fh_unlock (w->lock); fh_unref (w->fh); @@ -1113,7 +1298,7 @@ write_case_uncompressed (struct sfm_writer *w, const struct ccase *c) { size_t i; - for (i = 0; i < w->sfm_var_cnt; i++) + for (i = 0; i < w->sfm_n_vars; i++) { struct sfm_var *v = &w->sfm_vars[i]; @@ -1134,7 +1319,7 @@ write_case_compressed (struct sfm_writer *w, const struct ccase *c) { size_t i; - for (i = 0; i < w->sfm_var_cnt; i++) + for (i = 0; i < w->sfm_n_vars; i++) { struct sfm_var *v = &w->sfm_vars[i]; @@ -1148,10 +1333,7 @@ write_case_compressed (struct sfm_writer *w, const struct ccase *c) && d == (int) d) put_cmp_opcode (w, (int) d + COMPRESSION_BIAS); else - { - put_cmp_opcode (w, 253); - put_cmp_number (w, d); - } + put_cmp_number (w, d); } else { @@ -1169,10 +1351,7 @@ write_case_compressed (struct sfm_writer *w, const struct ccase *c) if (!memcmp (data, " ", chunk_size)) put_cmp_opcode (w, 254); else - { - put_cmp_opcode (w, 253); - put_cmp_string (w, data, chunk_size); - } + put_cmp_string (w, data, chunk_size); } /* This code deals properly with padding that is not a @@ -1186,19 +1365,146 @@ write_case_compressed (struct sfm_writer *w, const struct ccase *c) } } -/* Flushes buffered compressed opcodes and data to W. - The compression buffer must not be empty. */ +static bool +start_zstream (struct sfm_writer *w) +{ + int error; + + error = deflateInit (&w->zstream, 1); + if (error != Z_OK) + { + msg (ME, _("Failed to initialize ZLIB for compression (%s)."), + w->zstream.msg); + return false; + } + return true; +} + static void -flush_compressed (struct sfm_writer *w) +finish_zstream (struct sfm_writer *w) { - assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8); + struct zblock *block; + int error; - write_bytes (w, w->opcodes, w->opcode_cnt); - write_zeros (w, 8 - w->opcode_cnt); + assert (w->zstream.total_in <= ZBLOCK_SIZE); - write_bytes (w, w->data, w->data_cnt * sizeof *w->data); + w->zstream.next_in = NULL; + w->zstream.avail_in = 0; + do + { + uint8_t buf[4096]; - w->opcode_cnt = w->data_cnt = 0; + w->zstream.next_out = buf; + w->zstream.avail_out = sizeof buf; + error = deflate (&w->zstream, Z_FINISH); + write_bytes (w, buf, w->zstream.next_out - buf); + } + while (error == Z_OK); + + if (error != Z_STREAM_END) + msg (ME, _("Failed to complete ZLIB stream compression (%s)."), + w->zstream.msg); + + if (w->n_blocks >= w->allocated_blocks) + w->blocks = x2nrealloc (w->blocks, &w->allocated_blocks, + sizeof *w->blocks); + block = &w->blocks[w->n_blocks++]; + block->uncompressed_size = w->zstream.total_in; + block->compressed_size = w->zstream.total_out; + deflateEnd (&w->zstream); +} + +static void +write_zlib (struct sfm_writer *w, const void *data_, unsigned int n) +{ + const uint8_t *data = data_; + + while (n > 0) + { + unsigned int chunk; + + if (w->zstream.total_in >= ZBLOCK_SIZE) + { + finish_zstream (w); + start_zstream (w); + } + + chunk = MIN (n, ZBLOCK_SIZE - w->zstream.total_in); + + w->zstream.next_in = CONST_CAST (uint8_t *, data); + w->zstream.avail_in = chunk; + do + { + uint8_t buf[4096]; + int error; + + w->zstream.next_out = buf; + w->zstream.avail_out = sizeof buf; + error = deflate (&w->zstream, Z_NO_FLUSH); + write_bytes (w, buf, w->zstream.next_out - buf); + if (error != Z_OK) + { + msg (ME, _("ZLIB stream compression failed (%s)."), + w->zstream.msg); + return; + } + } + while (w->zstream.avail_in > 0 || w->zstream.avail_out == 0); + data += chunk; + n -= chunk; + } +} + +static void +write_ztrailer (struct sfm_writer *w) +{ + long long int uncompressed_ofs; + long long int compressed_ofs; + const struct zblock *block; + + write_int64 (w, -COMPRESSION_BIAS); + write_int64 (w, 0); + write_int (w, ZBLOCK_SIZE); + write_int (w, w->n_blocks); + + uncompressed_ofs = w->zstart; + compressed_ofs = w->zstart + 24; + for (block = w->blocks; block < &w->blocks[w->n_blocks]; block++) + { + write_int64 (w, uncompressed_ofs); + write_int64 (w, compressed_ofs); + write_int (w, block->uncompressed_size); + write_int (w, block->compressed_size); + + uncompressed_ofs += block->uncompressed_size; + compressed_ofs += block->compressed_size; + } + + if (!fseeko (w->file, w->zstart + 8, SEEK_SET)) + { + write_int64 (w, compressed_ofs); + write_int64 (w, 24 + (w->n_blocks * 24)); + } + else + msg (ME, _("%s: Seek failed (%s)."), + fh_get_file_name (w->fh), strerror (errno)); +} + +/* Flushes buffered compressed opcodes and data to W. */ +static void +flush_compressed (struct sfm_writer *w) +{ + if (w->n_opcodes) + { + unsigned int n = 8 * (1 + w->n_elements); + if (w->compression == ANY_COMP_SIMPLE) + write_bytes (w, w->cbuf, n); + else + write_zlib (w, w->cbuf, n); + + w->n_opcodes = w->n_elements = 0; + memset (w->cbuf[0], 0, 8); + } } /* Appends OPCODE to the buffered set of compression opcodes in @@ -1206,41 +1512,32 @@ flush_compressed (struct sfm_writer *w) static void put_cmp_opcode (struct sfm_writer *w, uint8_t opcode) { - if (w->opcode_cnt >= 8) + if (w->n_opcodes >= 8) flush_compressed (w); - w->opcodes[w->opcode_cnt++] = opcode; + w->cbuf[0][w->n_opcodes++] = opcode; } -/* Appends NUMBER to the buffered compression data in W. The - buffer must not be full; the way to assure that is to call - this function only just after a call to put_cmp_opcode, which - will flush the buffer as necessary. */ +/* Appends NUMBER to the buffered compression data in W. */ static void put_cmp_number (struct sfm_writer *w, double number) { - assert (w->opcode_cnt > 0); - assert (w->data_cnt < 8); - - convert_double_to_output_format (number, w->data[w->data_cnt++]); + put_cmp_opcode (w, 253); + convert_double_to_output_format (number, w->cbuf[++w->n_elements]); } /* Appends SIZE bytes of DATA to the buffered compression data in W, followed by enough spaces to pad the output data to exactly - 8 bytes (thus, SIZE must be no greater than 8). The buffer - must not be full; the way to assure that is to call this - function only just after a call to put_cmp_opcode, which will - flush the buffer as necessary. */ + 8 bytes (thus, SIZE must be no greater than 8). */ static void put_cmp_string (struct sfm_writer *w, const void *data, size_t size) { - assert (w->opcode_cnt > 0); - assert (w->data_cnt < 8); assert (size <= 8); - memset (w->data[w->data_cnt], ' ', 8); - memcpy (w->data[w->data_cnt], data, size); - w->data_cnt++; + put_cmp_opcode (w, 253); + w->n_elements++; + memset (w->cbuf[w->n_elements], w->space, 8); + memcpy (w->cbuf[w->n_elements], data, size); } /* Writes 32-bit integer X to the output file for writer W. */ @@ -1250,6 +1547,13 @@ write_int (struct sfm_writer *w, int32_t x) write_bytes (w, &x, sizeof x); } +/* Writes 64-bit integer X to the output file for writer W. */ +static void +write_int64 (struct sfm_writer *w, int64_t x) +{ + write_bytes (w, &x, sizeof x); +} + /* Converts NATIVE to the 64-bit format used in output files in OUTPUT. */ static inline void @@ -1285,7 +1589,7 @@ write_value (struct sfm_writer *w, const union value *value, int width) write_float (w, value->f); else { - write_bytes (w, value_str (value, width), width); + write_bytes (w, value->s, width); write_zeros (w, 8 - width); } } @@ -1300,7 +1604,7 @@ write_string (struct sfm_writer *w, const char *string, size_t width) size_t pad_bytes = width - data_bytes; write_bytes (w, string, data_bytes); while (pad_bytes-- > 0) - putc (' ', w->file); + putc (w->space, w->file); } /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the @@ -1361,5 +1665,5 @@ static void write_spaces (struct sfm_writer *w, size_t n) { while (n-- > 0) - putc (' ', w->file); + putc (w->space, w->file); }