#include <stdlib.h>
#include <sys/stat.h>
#include <time.h>
+#include <zlib.h>
#include "data/attributes.h"
#include "data/case.h"
FILE *file; /* File stream. */
struct replace_file *rf; /* Ticket for replacing output file. */
- bool compress; /* 1=compressed, 0=not compressed. */
+ enum sfm_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. */
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 *);
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 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 *);
sfm_writer_default_options (void)
{
struct sfm_write_options opts;
+ opts.compression = (settings_get_scompression ()
+ ? SFM_COMP_SIMPLE
+ : SFM_COMP_NONE);
opts.create_writeable = true;
- opts.compress = settings_get_scompression ();
opts.version = 3;
return opts;
}
}
/* 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 == SFM_COMP_ZLIB
+ && is_encoding_ebcdic_compatible (dict_get_encoding (d)))
+ w->compression = SFM_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
write_int (w, 999);
write_int (w, 0);
+ if (w->compression == SFM_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;
/* Record-type code. */
if (is_encoding_ebcdic_compatible (dict_encoding))
write_string (w, EBCDIC_MAGIC, 4);
+ else if (w->compression == SFM_COMP_ZLIB)
+ write_string (w, ASCII_ZMAGIC, 4);
else
write_string (w, ASCII_MAGIC, 4);
write_int (w, calc_oct_idx (d, NULL));
/* Compressed? */
- write_int (w, w->compress);
+ write_int (w, (w->compression == SFM_COMP_NONE ? 0
+ : w->compression == SFM_COMP_SIMPLE ? 1
+ : 2));
/* Weight variable. */
weight = dict_get_weight (d);
s = "0";
break;
- case ROLE_OUTPUT:
+ case ROLE_TARGET:
s = "1";
break;
w->case_cnt++;
- if (!w->compress)
+ if (w->compression == SFM_COMP_NONE)
write_case_uncompressed (w, c);
else
write_case_compressed (w, c);
if (w->file != NULL)
{
/* Flush buffer. */
- if (w->opcode_cnt > 0)
- flush_compressed (w);
+ flush_compressed (w);
+ if (w->compression == SFM_COMP_ZLIB)
+ {
+ finish_zstream (w);
+ write_ztrailer (w);
+ }
fflush (w->file);
ok = !write_error (w);
ok = false;
}
+ free (w->blocks);
+
fh_unlock (w->lock);
fh_unref (w->fh);
&& 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
{
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
}
}
-/* 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 == SFM_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
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], w->space, 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);
}
\f
/* Writes 32-bit integer X to the output file for writer W. */
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