X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fdata%2Fsys-file-writer.c;h=7a0fb5a691280ca3e4c542c1dbc3a9ebc09e92c3;hb=1a26ffe4489b411bda630d9e5b87e2d2e4538fdb;hp=c78e04d55f24e41ef5ecce805664c896fc15c44e;hpb=128f9567c68a176ab88fd0f54b2398de1f23ba9d;p=pspp diff --git a/src/data/sys-file-writer.c b/src/data/sys-file-writer.c index c78e04d55f..7a0fb5a691 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) 1997-2000, 2006-2013 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" @@ -72,11 +72,11 @@ 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 a series of 8-byte elements, with 1 to 9 such elements clustered together. The first element in a cluster is 8 @@ -89,6 +89,12 @@ struct sfm_writer 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. */ @@ -96,6 +102,12 @@ 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 *); @@ -134,6 +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); @@ -156,6 +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 *); @@ -164,8 +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; } @@ -194,13 +213,20 @@ 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->n_opcodes = w->n_elements = 0; @@ -224,8 +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."), @@ -279,6 +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; @@ -326,8 +365,6 @@ 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; @@ -336,6 +373,8 @@ write_header (struct sfm_writer *w, const struct dictionary *d) /* Record-type code. */ 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); @@ -351,7 +390,9 @@ 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); @@ -366,10 +407,11 @@ 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 { @@ -386,13 +428,14 @@ 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); @@ -464,6 +507,7 @@ write_variable (struct sfm_writer *w, const struct variable *v) 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); @@ -672,14 +716,14 @@ 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, ')'); } @@ -744,7 +788,7 @@ write_variable_attributes (struct sfm_writer *w, const struct dictionary *d) size_t i; for (i = 0; i < n_vars; i++) - { + { struct variable *v = dict_get_var (d, i); struct attrset attrs; @@ -810,8 +854,7 @@ 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); } @@ -910,7 +953,6 @@ write_long_string_value_labels (struct sfm_writer *w, 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; @@ -942,7 +984,6 @@ 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); @@ -968,8 +1009,7 @@ 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); @@ -979,7 +1019,6 @@ write_long_string_value_labels (struct sfm_writer *w, free (label); } } - assert (ftello (w->file) == start + size); } static void @@ -989,7 +1028,6 @@ write_long_string_missing_values (struct sfm_writer *w, 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; @@ -1015,7 +1053,6 @@ write_long_string_missing_values (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); @@ -1041,10 +1078,9 @@ write_long_string_missing_values (struct sfm_writer *w, const union value *value = mv_get_value (mv, j); write_int (w, 8); - write_bytes (w, value_str (value, width), 8); + write_bytes (w, value->s, 8); } } - assert (ftello (w->file) == start + size); } static void @@ -1171,7 +1207,7 @@ 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); @@ -1210,6 +1246,11 @@ close_writer (struct sfm_writer *w) { /* Flush buffer. */ flush_compressed (w); + if (w->compression == ANY_COMP_ZLIB) + { + finish_zstream (w); + write_ztrailer (w); + } fflush (w->file); ok = !write_error (w); @@ -1234,6 +1275,8 @@ close_writer (struct sfm_writer *w) ok = false; } + free (w->blocks); + fh_unlock (w->lock); fh_unref (w->fh); @@ -1324,13 +1367,143 @@ write_case_compressed (struct sfm_writer *w, const struct ccase *c) } } +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 +finish_zstream (struct sfm_writer *w) +{ + struct zblock *block; + int error; + + assert (w->zstream.total_in <= ZBLOCK_SIZE); + + w->zstream.next_in = NULL; + w->zstream.avail_in = 0; + do + { + uint8_t buf[4096]; + + 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) { - write_bytes (w, w->cbuf, 8 * (1 + w->n_elements)); + 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); } @@ -1376,6 +1549,13 @@ 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 @@ -1411,7 +1591,7 @@ 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); } }