X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fdata%2Fsys-file-writer.c;h=05d42b080ea9152ae280f6b2ccc327f87a551608;hb=5882d195ad6891b87dba6ded8a5caa1a949f165f;hp=d2522c23f82d1093be8edcea269c5841fc3fde69;hpb=cf5e75010ed9fa9b8eb75a71d1419d40e1a8ddd8;p=pspp diff --git a/src/data/sys-file-writer.c b/src/data/sys-file-writer.c index d2522c23f8..05d42b080e 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" @@ -72,20 +72,28 @@ struct sfm_writer FILE *file; /* File stream. */ struct replace_file *rf; /* Ticket for replacing output file. */ - bool compress; /* 1=compressed, 0=not compressed. */ + enum any_compression compression; casenumber case_cnt; /* 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. */ @@ -94,13 +102,20 @@ struct sfm_writer 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 *); static void write_variable (struct sfm_writer *, const struct variable *); -static void write_value_labels (struct sfm_writer *, struct variable *, - int idx); -static void write_integer_info_record (struct sfm_writer *); +static void write_value_labels (struct sfm_writer *, + const struct dictionary *); +static void write_integer_info_record (struct sfm_writer *, + const struct dictionary *); static void write_float_info_record (struct sfm_writer *); static void write_longvar_table (struct sfm_writer *w, @@ -114,6 +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); @@ -129,9 +146,16 @@ 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); +static void write_utf8_string (struct sfm_writer *, const char *encoding, + const char *string, size_t width); +static void write_utf8_record (struct sfm_writer *, const char *encoding, + const struct string *content, int subtype); +static void write_string_record (struct sfm_writer *, + const struct substring content, int subtype); static void write_bytes (struct sfm_writer *, const void *, size_t); static void write_zeros (struct sfm_writer *, size_t); static void write_spaces (struct sfm_writer *, size_t); @@ -145,6 +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 *); @@ -153,8 +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; } @@ -169,9 +199,9 @@ struct casewriter * sfm_open_writer (struct file_handle *fh, struct dictionary *d, struct sfm_write_options opts) { + struct encoding_info encoding_info; struct sfm_writer *w; mode_t mode; - int idx; int i; /* Check version. */ @@ -183,16 +213,24 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, } /* Create and initialize writer. */ - w = xmalloc (sizeof *w); + w = xzalloc (sizeof *w); w->fh = fh_ref (fh); w->lock = NULL; w->file = NULL; w->rf = NULL; - w->compress = opts.compress; + /* 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->case_cnt = 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 @@ -212,8 +250,7 @@ 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."), @@ -221,6 +258,9 @@ 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); @@ -229,20 +269,12 @@ sfm_open_writer (struct file_handle *fh, struct dictionary *d, for (i = 0; i < dict_get_var_cnt (d); i++) write_variable (w, dict_get_var (d, i)); - /* Write out value labels. */ - idx = 0; - for (i = 0; i < dict_get_var_cnt (d); i++) - { - struct variable *v = dict_get_var (d, i); - - write_value_labels (w, v, idx); - idx += sfm_width_to_octs (var_get_width (v)); - } + write_value_labels (w, d); if (dict_get_document_line_cnt (d) > 0) write_documents (w, d); - write_integer_info_record (w); + write_integer_info_record (w, d); write_float_info_record (w); write_mrsets (w, d, true); @@ -255,10 +287,14 @@ 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); @@ -268,6 +304,20 @@ 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; @@ -313,6 +363,7 @@ 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]; @@ -322,12 +373,17 @@ write_header (struct sfm_writer *w, const struct dictionary *d) 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", version, host_system); - write_string (w, prod_name, 60); + write_utf8_string (w, dict_encoding, prod_name, 60); /* Layout code. */ write_int (w, 2); @@ -336,7 +392,9 @@ 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); @@ -376,14 +434,14 @@ write_header (struct sfm_writer *w, const struct dictionary *d) snprintf (creation_time, sizeof creation_time, "%02d:%02d:%02d", hour - 1, min - 1, sec - 1); } - write_string (w, creation_date, 9); - write_string (w, creation_time, 8); + write_utf8_string (w, dict_encoding, creation_date, 9); + write_utf8_string (w, dict_encoding, creation_time, 8); /* File label. */ file_label = dict_get_label (d); if (file_label == NULL) file_label = ""; - write_string (w, file_label, 64); + write_utf8_string (w, dict_encoding, file_label, 64); /* Padding. */ write_zeros (w, 3); @@ -431,7 +489,7 @@ 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 seg0_width = sfm_segment_alloc_width (width, 0); - struct missing_values mv; + const char *encoding = var_get_encoding (v); int i; /* Record type. */ @@ -445,14 +503,20 @@ 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. */ + 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); @@ -461,12 +525,12 @@ write_variable (struct sfm_writer *w, const struct variable *v) /* Short name. The full name is in a translation table written separately. */ - write_string (w, var_get_short_name (v, 0), 8); + write_utf8_string (w, encoding, var_get_short_name (v, 0), 8); /* Value label. */ if (var_has_label (v)) { - char *label = recode_string (var_get_encoding (v), UTF8, var_get_label (v), -1); + char *label = recode_string (encoding, UTF8, var_get_label (v), -1); size_t label_len = MIN (strlen (label), 255); size_t padded_len = ROUND_UP (label_len, 4); write_int (w, label_len); @@ -475,15 +539,19 @@ 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); @@ -499,55 +567,123 @@ write_variable (struct sfm_writer *w, const struct variable *v) write_int (w, 0); /* No missing values. */ write_format (w, fmt, seg_width); /* Print format. */ write_format (w, fmt, seg_width); /* Write format. */ - write_string (w, var_get_short_name (v, i), 8); + write_utf8_string (w, encoding, var_get_short_name (v, i), 8); write_variable_continuation_records (w, seg_width); } - - mv_destroy (&mv); } -/* Writes the value labels for variable V having system file - variable index IDX to system file W. +/* Writes the value labels to system file W. Value labels for long string variables are written separately, by write_long_string_value_labels. */ static void -write_value_labels (struct sfm_writer *w, struct variable *v, int idx) +write_value_labels (struct sfm_writer *w, const struct dictionary *d) { - const struct val_labs *val_labs; - const struct val_lab **labels; - size_t n_labels; + struct label_set + { + struct hmap_node hmap_node; + const struct val_labs *val_labs; + int *indexes; + size_t n_indexes, allocated_indexes; + }; + + size_t n_sets, allocated_sets; + struct label_set **sets; + struct hmap same_sets; size_t i; + int idx; - val_labs = var_get_value_labels (v); - n_labels = val_labs_count (val_labs); - if (n_labels == 0 || var_get_width (v) > 8) - return; + n_sets = allocated_sets = 0; + sets = NULL; + hmap_init (&same_sets); - /* Value label record. */ - write_int (w, 3); /* Record type. */ - write_int (w, val_labs_count (val_labs)); - labels = val_labs_sorted (val_labs); - for (i = 0; i < n_labels; i++) + idx = 0; + for (i = 0; i < dict_get_var_cnt (d); i++) { - const struct val_lab *vl = labels[i]; - char *label = recode_string (var_get_encoding (v), UTF8, - val_lab_get_escaped_label (vl), -1); - uint8_t len = MIN (strlen (label), 255); - - write_value (w, val_lab_get_value (vl), var_get_width (v)); - write_bytes (w, &len, 1); - write_bytes (w, label, len); - write_zeros (w, REM_RND_UP (len + 1, 8)); - free (label); + struct variable *v = dict_get_var (d, i); + + if (var_has_value_labels (v) && var_get_width (v) <= 8) + { + const struct val_labs *val_labs = var_get_value_labels (v); + unsigned int hash = val_labs_hash (val_labs, 0); + struct label_set *set; + + HMAP_FOR_EACH_WITH_HASH (set, struct label_set, hmap_node, + hash, &same_sets) + { + if (val_labs_equal (set->val_labs, val_labs)) + { + if (set->n_indexes >= set->allocated_indexes) + set->indexes = x2nrealloc (set->indexes, + &set->allocated_indexes, + sizeof *set->indexes); + set->indexes[set->n_indexes++] = idx; + goto next_var; + } + } + + set = xmalloc (sizeof *set); + set->val_labs = val_labs; + set->indexes = xmalloc (sizeof *set->indexes); + set->indexes[0] = idx; + set->n_indexes = 1; + set->allocated_indexes = 1; + hmap_insert (&same_sets, &set->hmap_node, hash); + + if (n_sets >= allocated_sets) + sets = x2nrealloc (sets, &allocated_sets, sizeof *sets); + sets[n_sets++] = set; + } + + next_var: + idx += sfm_width_to_octs (var_get_width (v)); + } + + for (i = 0; i < n_sets; i++) + { + const struct label_set *set = sets[i]; + const struct val_labs *val_labs = set->val_labs; + size_t n_labels = val_labs_count (val_labs); + int width = val_labs_get_width (val_labs); + const struct val_lab **labels; + size_t j; + + /* Value label record. */ + write_int (w, 3); /* Record type. */ + write_int (w, n_labels); + labels = val_labs_sorted (val_labs); + for (j = 0; j < n_labels; j++) + { + const struct val_lab *vl = labels[j]; + char *label = recode_string (dict_get_encoding (d), UTF8, + val_lab_get_escaped_label (vl), -1); + uint8_t len = MIN (strlen (label), 255); + + write_value (w, val_lab_get_value (vl), width); + write_bytes (w, &len, 1); + write_bytes (w, label, len); + write_zeros (w, REM_RND_UP (len + 1, 8)); + free (label); + } + free (labels); + + /* Value label variable record. */ + write_int (w, 4); /* Record type. */ + write_int (w, set->n_indexes); + for (j = 0; j < set->n_indexes; j++) + write_int (w, set->indexes[j] + 1); } - free (labels); - /* Value label variable record. */ - write_int (w, 4); /* Record type. */ - write_int (w, 1); /* Number of variables. */ - write_int (w, idx + 1); /* Variable's dictionary index. */ + for (i = 0; i < n_sets; i++) + { + struct label_set *set = sets[i]; + + free (set->indexes); + free (set); + } + free (sets); + hmap_destroy (&same_sets); } /* Writes record type 6, document record. */ @@ -592,27 +728,56 @@ put_attrset (struct string *string, const struct attrset *attrs) } } -static void -write_attribute_record (struct sfm_writer *w, const struct string *content, - int subtype) -{ - write_int (w, 7); - write_int (w, subtype); - write_int (w, 1); - write_int (w, ds_length (content)); - write_bytes (w, ds_data (content), ds_length (content)); -} - static void write_data_file_attributes (struct sfm_writer *w, const struct dictionary *d) { struct string s = DS_EMPTY_INITIALIZER; put_attrset (&s, dict_get_attributes (d)); - write_attribute_record (w, &s, 17); + write_utf8_record (w, dict_get_encoding (d), &s, 17); 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) { @@ -624,17 +789,19 @@ write_variable_attributes (struct sfm_writer *w, const struct dictionary *d) 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_short_name (v, 0)); - 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_attribute_record (w, &s, 18); + if (n_attrsets) + write_utf8_record (w, dict_get_encoding (d), &s, 18); ds_destroy (&s); } @@ -645,10 +812,17 @@ static void write_mrsets (struct sfm_writer *w, const struct dictionary *dict, bool pre_v14) { + const char *encoding = dict_get_encoding (dict); struct string s = DS_EMPTY_INITIALIZER; 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; @@ -656,14 +830,17 @@ write_mrsets (struct sfm_writer *w, const struct dictionary *dict, for (i = 0; i < n_mrsets; i++) { const struct mrset *mrset = dict_get_mrset (dict, i); - const char *label; + char *name; size_t j; if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES) != pre_v14) continue; - ds_put_format (&s, "%s=", mrset->name); + name = recode_string (encoding, "UTF-8", mrset->name, -1); + ds_put_format (&s, "%s=", name); + free (name); + if (mrset->type == MRSET_MD) { char *counted; @@ -685,14 +862,30 @@ write_mrsets (struct sfm_writer *w, const struct dictionary *dict, ds_put_byte (&s, 'C'); ds_put_byte (&s, ' '); - label = mrset->label && !mrset->label_from_var_label ? mrset->label : ""; - ds_put_format (&s, "%zu %s", strlen (label), label); + if (mrset->label && !mrset->label_from_var_label) + { + char *label = recode_string (encoding, "UTF-8", mrset->label, -1); + ds_put_format (&s, "%zu %s", strlen (label), label); + free (label); + } + else + ds_put_cstr (&s, "0 "); for (j = 0; j < mrset->n_vars; j++) - ds_put_format (&s, " %s", var_get_short_name (mrset->vars[j], 0)); + { + 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", + lower_name_utf8, -1); + ds_put_format (&s, " %s", short_name); + free (short_name); + free (lower_name_utf8); + } ds_put_byte (&s, '\n'); } - write_attribute_record (w, &s, pre_v14 ? 7 : 19); + + if (!ds_is_empty (&s)) + write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19); ds_destroy (&s); } @@ -749,21 +942,15 @@ write_vls_length_table (struct sfm_writer *w, var_get_short_name (v, 0), var_get_width (v), 0); } if (!ds_is_empty (&map)) - { - write_int (w, 7); /* Record type. */ - write_int (w, 14); /* Record subtype. */ - write_int (w, 1); /* Data item (char) size. */ - write_int (w, ds_length (&map)); /* Number of data items. */ - write_bytes (w, ds_data (&map), ds_length (&map)); - } + write_utf8_record (w, dict_get_encoding (dict), &map, 14); ds_destroy (&map); } - static void write_long_string_value_labels (struct sfm_writer *w, const struct dictionary *dict) { + const char *encoding = dict_get_encoding (dict); size_t n_vars = dict_get_var_cnt (dict); size_t size, i; off_t start UNUSED; @@ -780,10 +967,15 @@ write_long_string_value_labels (struct sfm_writer *w, if (val_labs_count (val_labs) == 0 || width < 9) continue; - size += 12 + strlen (var_get_name (var)); + size += 12; + size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1); for (val_lab = val_labs_first (val_labs); val_lab != NULL; val_lab = val_labs_next (val_labs, val_lab)) - size += 8 + width + strlen (val_lab_get_escaped_label (val_lab)); + { + size += 8 + width; + size += recode_string_len (encoding, "UTF-8", + val_lab_get_escaped_label (val_lab), -1); + } } if (size == 0) return; @@ -798,49 +990,121 @@ 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 *var_name = var_get_name (var); int width = var_get_width (var); const struct val_lab *val_lab; + char *var_name; if (val_labs_count (val_labs) == 0 || 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); + write_int (w, width); write_int (w, val_labs_count (val_labs)); for (val_lab = val_labs_first (val_labs); val_lab != NULL; val_lab = val_labs_next (val_labs, val_lab)) { - const char *label = val_lab_get_escaped_label (val_lab); - size_t label_length = strlen (label); + char *label; + size_t len; write_int (w, width); write_bytes (w, value_str (val_lab_get_value (val_lab), width), width); - write_int (w, label_length); - write_bytes (w, label, label_length); + + label = recode_string (var_get_encoding (var), "UTF-8", + val_lab_get_escaped_label (val_lab), -1); + len = strlen (label); + write_int (w, len); + write_bytes (w, label, len); + free (label); } } assert (ftello (w->file) == start + size); } static void -write_encoding_record (struct sfm_writer *w, - const struct dictionary *d) +write_long_string_missing_values (struct sfm_writer *w, + const struct dictionary *dict) { - const char *enc = dict_get_encoding (d); + const char *encoding = dict_get_encoding (dict); + size_t n_vars = dict_get_var_cnt (dict); + size_t size, i; + off_t start UNUSED; + + /* 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 ( NULL == enc) + 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, 20); /* Record subtype. */ - write_int (w, 1); /* Data item (char) size. */ - write_int (w, strlen (enc)); /* Number of data items. */ - write_string (w, enc, strlen (enc)); + write_int (w, 22); /* Record subtype */ + 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 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_str (value, width), 8); + } + } + assert (ftello (w->file) == start + size); } +static void +write_encoding_record (struct sfm_writer *w, + const struct dictionary *d) +{ + /* IANA says "...character set names may be up to 40 characters taken + from the printable characters of US-ASCII," so character set names + don't need to be recoded to be in UTF-8. + + We convert encoding names to uppercase because SPSS writes encoding + names in uppercase. */ + char *encoding = xstrdup (dict_get_encoding (d)); + str_uppercase (encoding); + write_string_record (w, ss_cstr (encoding), 20); + free (encoding); +} /* Writes the long variable name table. */ static void @@ -853,30 +1117,24 @@ write_longvar_table (struct sfm_writer *w, const struct dictionary *dict) for (i = 0; i < dict_get_var_cnt (dict); i++) { struct variable *v = dict_get_var (dict, i); - char *longname = recode_string (dict_get_encoding (dict), UTF8, var_get_name (v), -1); - if (i) ds_put_byte (&map, '\t'); ds_put_format (&map, "%s=%s", - var_get_short_name (v, 0), longname); - free (longname); + var_get_short_name (v, 0), var_get_name (v)); } - - write_int (w, 7); /* Record type. */ - write_int (w, 13); /* Record subtype. */ - write_int (w, 1); /* Data item (char) size. */ - write_int (w, ds_length (&map)); /* Number of data items. */ - write_bytes (w, ds_data (&map), ds_length (&map)); - + write_utf8_record (w, dict_get_encoding (dict), &map, 13); ds_destroy (&map); } /* Write integer information record. */ static void -write_integer_info_record (struct sfm_writer *w) +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; /* Parse the version string. */ memset (version_component, 0, sizeof version_component); @@ -894,6 +1152,24 @@ write_integer_info_record (struct sfm_writer *w) else abort (); + /* Choose codepage. */ + codepage = sys_get_codepage_from_encoding (dict_encoding); + if (codepage == 0) + { + /* 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. + */ + if (is_encoding_ascii_compatible (dict_encoding)) + codepage = 2; + else if (is_encoding_ebcdic_compatible (dict_encoding)) + codepage = 1; + } + /* Write record. */ write_int (w, 7); /* Record type. */ write_int (w, 3); /* Record subtype. */ @@ -906,7 +1182,7 @@ write_integer_info_record (struct sfm_writer *w) write_int (w, float_format); write_int (w, 1); /* Compression code. */ write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2); - write_int (w, 2); /* 7-bit ASCII. */ + write_int (w, codepage); } /* Write floating-point information record. */ @@ -938,7 +1214,7 @@ sys_file_casewriter_write (struct casewriter *writer, void *w_, w->case_cnt++; - if (!w->compress) + if (w->compression == ANY_COMP_NONE) write_case_uncompressed (w, c); else write_case_compressed (w, c); @@ -976,8 +1252,12 @@ 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); @@ -1002,6 +1282,8 @@ close_writer (struct sfm_writer *w) ok = false; } + free (w->blocks); + fh_unlock (w->lock); fh_unref (w->fh); @@ -1060,10 +1342,7 @@ 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 { @@ -1081,10 +1360,7 @@ 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 @@ -1098,19 +1374,145 @@ 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; +} + +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 @@ -1118,41 +1520,32 @@ 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. */ @@ -1162,6 +1555,13 @@ 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 @@ -1202,9 +1602,9 @@ write_value (struct sfm_writer *w, const union value *value, int width) } } -/* Writes null-terminated STRING in a field of the given WIDTH to - W. If STRING is longer than WIDTH, it is truncated; if WIDTH - is narrowed, it is padded on the right with spaces. */ +/* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING + is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it + is padded on the right with spaces. */ static void write_string (struct sfm_writer *w, const char *string, size_t width) { @@ -1212,7 +1612,45 @@ 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 + recoded version in a field of the given WIDTH to W. The string is truncated + or padded on the right with spaces to exactly WIDTH bytes. */ +static void +write_utf8_string (struct sfm_writer *w, const char *encoding, + const char *string, size_t width) +{ + char *s = recode_string (encoding, "UTF-8", string, -1); + write_string (w, s, width); + free (s); +} + +/* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded + from UTF-8 encoded into ENCODING. */ +static void +write_utf8_record (struct sfm_writer *w, const char *encoding, + const struct string *content, int subtype) +{ + struct substring s; + + s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL); + write_string_record (w, s, subtype); + ss_dealloc (&s); +} + +/* Writes a record with type 7, subtype SUBTYPE that contains the string + CONTENT. */ +static void +write_string_record (struct sfm_writer *w, + const struct substring content, int subtype) +{ + write_int (w, 7); + write_int (w, subtype); + write_int (w, 1); + write_int (w, ss_length (content)); + write_bytes (w, ss_data (content), ss_length (content)); } /* Writes SIZE bytes of DATA to W's output file. */ @@ -1235,5 +1673,5 @@ static void write_spaces (struct sfm_writer *w, size_t n) { while (n-- > 0) - putc (' ', w->file); + putc (w->space, w->file); }