/* PSPP - a program for statistical analysis.
- Copyright (C) 2006, 2009, 2010, 2011, 2012 Free Software Foundation, Inc.
+ Copyright (C) 2006, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 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
#include <errno.h>
#include <iconv.h>
#include <langinfo.h>
-#include <libintl.h>
#include <locale.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <unicase.h>
#include <unigbrk.h>
#include "libpspp/assertion.h"
#include "libpspp/str.h"
#include "libpspp/version.h"
+#include "gl/c-ctype.h"
#include "gl/c-strcase.h"
#include "gl/localcharset.h"
+#include "gl/minmax.h"
#include "gl/xalloc.h"
#include "gl/relocatable.h"
#include "gl/xstrndup.h"
#define _(msgid) gettext (msgid)
struct converter
- {
- char *tocode;
- char *fromcode;
- iconv_t conv;
- int error;
- };
+{
+ char *tocode;
+ char *fromcode;
+ iconv_t conv;
+ int null_char_width;
+};
static char *default_encoding;
static struct hmapx map;
/* A wrapper around iconv_open */
static struct converter *
-create_iconv__ (const char* tocode, const char* fromcode)
+create_iconv (const char* tocode, const char* fromcode)
{
size_t hash;
struct hmapx_node *node;
hash = hash_string (tocode, hash_string (fromcode, 0));
HMAPX_FOR_EACH_WITH_HASH (converter, node, hash, &map)
- if (!strcmp (tocode, converter->tocode)
- && !strcmp (fromcode, converter->fromcode))
- return converter;
+ {
+ if (!converter)
+ return NULL;
+
+ if (!strcmp (tocode, converter->tocode)
+ && !strcmp (fromcode, converter->fromcode))
+ return converter;
+ }
converter = xmalloc (sizeof *converter);
converter->tocode = xstrdup (tocode);
converter->fromcode = xstrdup (fromcode);
converter->conv = iconv_open (tocode, fromcode);
- converter->error = converter->conv == (iconv_t) -1 ? errno : 0;
- hmapx_insert (&map, converter, hash);
-
- return converter;
-}
-
-static iconv_t
-create_iconv (const char* tocode, const char* fromcode)
-{
- struct converter *converter;
-
- converter = create_iconv__ (tocode, fromcode);
-
+ int error = converter->conv == (iconv_t) ~0 ? errno : 0;
/* I don't think it's safe to translate this string or to use messaging
as the converters have not yet been set up */
- if (converter->error && strcmp (tocode, fromcode))
+ if (error && strcmp (tocode, fromcode))
{
fprintf (stderr,
"Warning: "
"cannot create a converter for `%s' to `%s': %s\n",
- fromcode, tocode, strerror (converter->error));
- converter->error = 0;
+ fromcode, tocode, strerror (error));
+
+ free (converter->tocode);
+ free (converter->fromcode);
+ free (converter);
+
+ hmapx_insert (&map, NULL, hash);
+ return NULL;
}
- return converter->conv;
+ /* Find out how many bytes there are in a null char in the target
+ encoding */
+ iconv_t bconv = iconv_open (tocode, "ASCII");
+ if (bconv != (iconv_t) -1)
+ {
+ ICONV_CONST char *nullstr = strdup ("");
+ ICONV_CONST char *outbuf = strdup ("XXXXXXXX");
+ ICONV_CONST char *snullstr = nullstr;
+ ICONV_CONST char *soutbuf = outbuf;
+
+ size_t inbytes = 1;
+ const size_t bytes = 8;
+ size_t outbytes = bytes;
+ if (-1 != iconv (bconv, &nullstr, &inbytes, &outbuf, &outbytes))
+ converter->null_char_width = bytes - outbytes;
+ free (snullstr);
+ free (soutbuf);
+ iconv_close (bconv);
+ }
+
+ hmapx_insert (&map, converter, hash);
+
+ return converter;
}
+
/* Converts the single byte C from encoding FROM to TO, returning the first
byte of the result.
Returns the output length if successful, -1 if the output buffer is too
small. */
static ssize_t
-try_recode (iconv_t conv,
- const char *ip, size_t inbytes,
- char *op_, size_t outbytes)
+try_recode (struct converter *cvtr, char fallbackchar,
+ const char *in, size_t inbytes,
+ char *out_, size_t outbytes)
{
- /* FIXME: Need to ensure that this char is valid in the target encoding */
- const char fallbackchar = '?';
- char *op = op_;
+ char *out = out_;
+ int i, j;
+
+ int null_bytes = cvtr->null_char_width;
/* Put the converter into the initial shift state, in case there was any
state information left over from its last usage. */
- iconv (conv, NULL, 0, NULL, 0);
+ iconv (cvtr->conv, NULL, 0, NULL, 0);
- while (iconv (conv, (ICONV_CONST char **) &ip, &inbytes,
- &op, &outbytes) == -1)
- switch (errno)
- {
- case EINVAL:
- if (outbytes < 2)
- return -1;
- *op++ = fallbackchar;
- *op = '\0';
- return op - op_;
-
- case EILSEQ:
- if (outbytes == 0)
- return -1;
- *op++ = fallbackchar;
- outbytes--;
- ip++;
- inbytes--;
- break;
-
- case E2BIG:
- return -1;
-
- default:
- /* should never happen */
- fprintf (stderr, "Character conversion error: %s\n", strerror (errno));
- NOT_REACHED ();
- break;
- }
+ /* Do two rounds of iconv() calls:
+
+ - The first round does the bulk of the conversion using the
+ caller-supplied input data..
+
+ - The second round flushes any leftover output. This has a real effect
+ with input encodings that use combining diacritics, e.g. without the
+ second round the last character tends to gets dropped when converting
+ from windows-1258 to other encodings.
+ */
+ for (i = 0; i < 2; i++)
+ {
+ ICONV_CONST char **inp = i ? NULL : (ICONV_CONST char **) ∈
+ size_t *inbytesp = i ? NULL : &inbytes;
+
+ while (iconv (cvtr->conv, inp, inbytesp, &out, &outbytes) == -1)
+ switch (errno)
+ {
+ case EINVAL:
+ if (outbytes < null_bytes + 1)
+ return -E2BIG;
+ if (!fallbackchar)
+ return -EINVAL;
+ *out++ = fallbackchar;
+ for (j = 0 ; j < null_bytes ; ++j)
+ *out++ = '\0';
+ return out - 1 - out_;
+
+ case EILSEQ:
+ if (outbytes == 0)
+ return -E2BIG;
+ if (!fallbackchar)
+ return -EILSEQ;
+ *out++ = fallbackchar;
+ outbytes--;
+ if (inp)
+ {
+ in++;
+ inbytes--;
+ }
+ break;
+
+ case E2BIG:
+ return -E2BIG;
+
+ default:
+ /* should never happen */
+ fprintf (stderr, "Character conversion error: %s\n",
+ strerror (errno));
+ NOT_REACHED ();
+ break;
+ }
+ }
+
+ if (outbytes <= null_bytes - 1)
+ return -E2BIG;
- if (outbytes == 0)
- return -1;
+ for (i = 0 ; i < null_bytes ; ++i)
+ *out++ = '\0';
- *op = '\0';
- return op - op_;
+ return out - 1 - out_;
}
/* Converts the string TEXT, which should be encoded in FROM-encoding, to a
return NULL;
if ( length == -1 )
- length = strlen (text);
+ length = strlen (text);
out = recode_substring_pool (to, from, ss_buffer (text, length), pool);
return out.string;
fits or until HEAD_LEN reaches 0.
[*] Actually this function drops grapheme clusters instead of characters, so
- that, e.g. a Unicode character followed by a combining accent character
- is either completely included or completely excluded from HEAD_LEN. See
- UAX #29 at http://unicode.org/reports/tr29/ for more information on
- grapheme clusters.
+ that, e.g. a Unicode character followed by a combining accent character
+ is either completely included or completely excluded from HEAD_LEN. See
+ UAX #29 at http://unicode.org/reports/tr29/ for more information on
+ grapheme clusters.
A null ENCODING is treated as UTF-8.
Simple examples for encoding="UTF-8", max_len=6:
- head="abc", tail="xyz" => 3
- head="abcd", tail="xyz" => 3 ("d" dropped).
- head="abc", tail="uvwxyz" => 0 ("abc" dropped).
- head="abc", tail="tuvwxyz" => 0 ("abc" dropped).
+ head="abc", tail="xyz" => 3
+ head="abcd", tail="xyz" => 3 ("d" dropped).
+ head="abc", tail="uvwxyz" => 0 ("abc" dropped).
+ head="abc", tail="tuvwxyz" => 0 ("abc" dropped).
Examples for encoding="ISO-8859-1", max_len=6:
- head="éèä", tail="xyz" => 6
- (each letter in head is only 1 byte in ISO-8859-1 even though they
- each take 2 bytes in UTF-8 encoding)
+ head="éèä", tail="xyz" => 6
+ (each letter in head is only 1 byte in ISO-8859-1 even though they
+ each take 2 bytes in UTF-8 encoding)
*/
static size_t
utf8_encoding_concat__ (const char *head, size_t head_len,
included, even if TAIL by itself is longer than MAX_LEN in ENCODING.
[*] Actually this function drops grapheme clusters instead of characters, so
- that, e.g. a Unicode character followed by a combining accent character
- is either completely included or completely excluded from the returned
- string. See UAX #29 at http://unicode.org/reports/tr29/ for more
- information on grapheme clusters.
+ that, e.g. a Unicode character followed by a combining accent character
+ is either completely included or completely excluded from the returned
+ string. See UAX #29 at http://unicode.org/reports/tr29/ for more
+ information on grapheme clusters.
A null ENCODING is treated as UTF-8.
Simple examples for encoding="UTF-8", max_len=6:
- head="abc", tail="xyz" => "abcxyz"
- head="abcd", tail="xyz" => "abcxyz"
- head="abc", tail="uvwxyz" => "uvwxyz"
- head="abc", tail="tuvwxyz" => "tuvwxyz"
+ head="abc", tail="xyz" => "abcxyz"
+ head="abcd", tail="xyz" => "abcxyz"
+ head="abc", tail="uvwxyz" => "uvwxyz"
+ head="abc", tail="tuvwxyz" => "tuvwxyz"
Examples for encoding="ISO-8859-1", max_len=6:
- head="éèä", tail="xyz" => "éèäxyz"
- (each letter in HEAD is only 1 byte in ISO-8859-1 even though they
- each take 2 bytes in UTF-8 encoding)
+ head="éèä", tail="xyz" => "éèäxyz"
+ (each letter in HEAD is only 1 byte in ISO-8859-1 even though they
+ each take 2 bytes in UTF-8 encoding)
*/
char *
utf8_encoding_concat (const char *head, const char *tail,
ENCODING. Both S and the returned string are encoded in UTF-8.
[*] Actually this function drops grapheme clusters instead of characters, so
- that, e.g. a Unicode character followed by a combining accent character
- is either completely included or completely excluded from the returned
- string. See UAX #29 at http://unicode.org/reports/tr29/ for more
- information on grapheme clusters.
+ that, e.g. a Unicode character followed by a combining accent character
+ is either completely included or completely excluded from the returned
+ string. See UAX #29 at http://unicode.org/reports/tr29/ for more
+ information on grapheme clusters.
A null ENCODING is treated as UTF-8.
*/
return recode_string ("UTF-8", filename_encoding (), filename, -1);
}
-/* Converts the string TEXT, which should be encoded in FROM-encoding, to a
- dynamically allocated string in TO-encoding. Any characters which cannot be
- converted will be represented by '?'.
-
- The returned string will be null-terminated and allocated on POOL.
-
- This function's behaviour differs from that of g_convert_with_fallback
- provided by GLib. The GLib function will fail (returns NULL) if any part of
- the input string is not valid in the declared input encoding. This function
- however perseveres even in the presence of badly encoded input. */
-struct substring
-recode_substring_pool (const char *to, const char *from,
- struct substring text, struct pool *pool)
+static int
+recode_substring_pool__ (const char *to, const char *from,
+ struct substring text, char fallbackchar,
+ struct pool *pool, struct substring *out)
{
- size_t outbufferlength;
- iconv_t conv ;
+ size_t bufsize;
+ struct converter *conv;
if (to == NULL)
to = default_encoding;
conv = create_iconv (to, from);
- if ( (iconv_t) -1 == conv )
+ if ( NULL == conv )
{
- struct substring out;
- ss_alloc_substring_pool (&out, text, pool);
- return out;
+ if (fallbackchar)
+ {
+ out->string = pool_malloc (pool, text.length + 1);
+ out->length = text.length;
+ memcpy (out->string, text.string, text.length);
+ out->string[out->length] = '\0';
+ return 0;
+ }
+ else
+ return EPROTO;
}
- for ( outbufferlength = 1 ; outbufferlength != 0; outbufferlength <<= 1 )
- if ( outbufferlength > text.length)
- {
- char *output = pool_malloc (pool, outbufferlength);
- ssize_t output_len = try_recode (conv, text.string, text.length,
- output, outbufferlength);
- if (output_len >= 0)
- return ss_buffer (output, output_len);
- pool_free (pool, output);
- }
+ for (bufsize = text.length + 1; bufsize > text.length; bufsize *= 2)
+ {
+ char *output = pool_malloc (pool, bufsize);
+ ssize_t retval;
+
+ retval = try_recode (conv, fallbackchar, text.string, text.length,
+ output, bufsize);
+ if (retval >= 0)
+ {
+ *out = ss_buffer (output, retval);
+ return 0;
+ }
+ pool_free (pool, output);
+
+ if (retval != -E2BIG)
+ return -retval;
+ }
NOT_REACHED ();
}
+/* Converts the string TEXT, which should be encoded in FROM-encoding, to a
+ dynamically allocated string in TO-encoding. Any characters which cannot be
+ converted will be represented by '?'.
+
+ The returned string will be null-terminated and allocated on POOL with
+ pool_malloc().
+
+ This function's behaviour differs from that of g_convert_with_fallback
+ provided by GLib. The GLib function will fail (returns NULL) if any part of
+ the input string is not valid in the declared input encoding. This function
+ however perseveres even in the presence of badly encoded input. */
+struct substring
+recode_substring_pool (const char *to, const char *from,
+ struct substring text, struct pool *pool)
+{
+ struct substring out;
+
+ recode_substring_pool__ (to, from, text, '?', pool, &out);
+ return out;
+}
+
+/* Converts the string TEXT, which should be encoded in FROM-encoding, to a
+ dynamically allocated string in TO-encoding. On success, returns 0, and the
+ converted null-terminated string, allocated from POOL with pool_malloc(), is
+ stored in *OUT. On failure, returns a positive errno value.
+
+ The function fails with an error if any part of the input string is not
+ valid in the declared input encoding. */
+int
+recode_pedantically (const char *to, const char *from,
+ struct substring text, struct pool *pool,
+ struct substring *out)
+{
+ int error;
+
+ error = recode_substring_pool__ (to, from, text, 0, pool, out);
+ if (error)
+ *out = ss_empty ();
+ return error;
+}
+\f
void
i18n_init (void)
{
- setlocale (LC_CTYPE, "");
- setlocale (LC_COLLATE, "");
- setlocale (LC_MESSAGES, "");
-#if HAVE_LC_PAPER
- setlocale (LC_PAPER, "");
-#endif
+ setlocale (LC_ALL, "");
bindtextdomain (PACKAGE, relocate(locale_dir));
textdomain (PACKAGE);
/* Attempts to set the encoding from a locale name
- returns true if successfull.
+ returns true if successful.
This function does not (should not!) alter the current locale.
*/
bool
ok = false;
}
-
setlocale (LC_CTYPE, tmp);
free (tmp);
HMAPX_FOR_EACH (cvtr, node, &map)
{
+ if (cvtr == NULL)
+ continue;
free (cvtr->tocode);
free (cvtr->fromcode);
if (cvtr->conv != (iconv_t) -1)
/* Return the system local's idea of the
- decimal seperator character */
+ decimal separator character */
char
get_system_decimal (void)
{
char radix_char;
- char *ol = xstrdup (setlocale (LC_NUMERIC, NULL));
- setlocale (LC_NUMERIC, "");
-
#if HAVE_NL_LANGINFO
radix_char = nl_langinfo (RADIXCHAR)[0];
#else
}
#endif
- /* We MUST leave LC_NUMERIC untouched, since it would
- otherwise interfere with data_{in,out} */
- setlocale (LC_NUMERIC, ol);
- free (ol);
return radix_char;
}
return buffer;
}
\f
+/* UTF-8 functions that deal with uppercase/lowercase distinctions. */
+
+/* Returns a hash value for the N bytes of UTF-8 encoded data starting at S,
+ with lowercase and uppercase letters treated as equal, starting from
+ BASIS. */
+unsigned int
+utf8_hash_case_bytes (const char *s, size_t n, unsigned int basis)
+{
+ uint8_t folded_buf[2048];
+ size_t folded_len = sizeof folded_buf;
+ uint8_t *folded_s;
+ unsigned int hash;
+
+ folded_s = u8_casefold (CHAR_CAST (const uint8_t *, s), n,
+ NULL, UNINORM_NFKD, folded_buf, &folded_len);
+ if (folded_s != NULL)
+ {
+ hash = hash_bytes (folded_s, folded_len, basis);
+ if (folded_s != folded_buf)
+ free (folded_s);
+ }
+ else
+ {
+ if (errno == ENOMEM)
+ xalloc_die ();
+ hash = hash_bytes (s, n, basis);
+ }
+
+ return hash;
+}
+
+/* Returns a hash value for null-terminated UTF-8 string S, with lowercase and
+ uppercase letters treated as equal, starting from BASIS. */
+unsigned int
+utf8_hash_case_string (const char *s, unsigned int basis)
+{
+ return utf8_hash_case_bytes (s, strlen (s), basis);
+}
+
+/* Compares UTF-8 strings A and B case-insensitively.
+ Returns a negative value if A < B, zero if A == B, positive if A > B. */
+int
+utf8_strcasecmp (const char *a, const char *b)
+{
+ return utf8_strncasecmp (a, strlen (a), b, strlen (b));
+}
+
+/* Compares UTF-8 strings A (with length AN) and B (with length BN)
+ case-insensitively.
+ Returns a negative value if A < B, zero if A == B, positive if A > B. */
+int
+utf8_strncasecmp (const char *a, size_t an, const char *b, size_t bn)
+{
+ int result;
+
+ if (u8_casecmp (CHAR_CAST (const uint8_t *, a), an,
+ CHAR_CAST (const uint8_t *, b), bn,
+ NULL, UNINORM_NFKD, &result))
+ {
+ if (errno == ENOMEM)
+ xalloc_die ();
+
+ result = memcmp (a, b, MIN (an, bn));
+ if (result == 0)
+ result = an < bn ? -1 : an > bn;
+ }
+
+ return result;
+}
+
+static bool
+is_all_digits (const uint8_t *s, size_t len)
+{
+ for (size_t i = 0; i < len; i++)
+ if (!c_isdigit (s[i]))
+ return false;
+ return true;
+}
+
+/* Compares UTF-8 strings A and B case-insensitively. If the strings end in a
+ number, then they are compared numerically. Returns a negative value if A <
+ B, zero if A == B, positive if A > B. */
+int
+utf8_strverscasecmp (const char *a, const char *b)
+{
+ /* Normalize A. */
+ uint8_t a_stub[64];
+ size_t a_len = sizeof a_stub;
+ uint8_t *a_norm = u8_casefold (CHAR_CAST (uint8_t *, a), strlen (a), NULL,
+ UNINORM_NFKD, a_stub, &a_len);
+
+ /* Normalize B. */
+ uint8_t b_stub[64];
+ size_t b_len = sizeof b_stub;
+ uint8_t *b_norm = u8_casefold (CHAR_CAST (uint8_t *, b), strlen (b), NULL,
+ UNINORM_NFKD, b_stub, &b_len);
+
+ int result;
+ if (!a_norm || !b_norm)
+ {
+ result = strcmp (a, b);
+ goto exit;
+ }
+
+ size_t len = MIN (a_len, b_len);
+ for (size_t i = 0; i < len; i++)
+ if (a_norm[i] != b_norm[i])
+ {
+ /* If both strings end in digits, compare them numerically. */
+ if (is_all_digits (&a_norm[i], a_len - i)
+ && is_all_digits (&b_norm[i], b_len - i))
+ {
+ /* Start by stripping leading zeros, since those don't matter for
+ numerical comparison. */
+ size_t ap, bp;
+ for (ap = i; ap < a_len; ap++)
+ if (a_norm[ap] != '0')
+ break;
+ for (bp = i; bp < b_len; bp++)
+ if (b_norm[bp] != '0')
+ break;
+
+ /* The number with more digits, if there is one, is larger. */
+ size_t a_digits = a_len - ap;
+ size_t b_digits = b_len - bp;
+ if (a_digits != b_digits)
+ result = a_digits > b_digits ? 1 : -1;
+ else
+ result = memcmp (&a_norm[ap], &b_norm[bp], a_digits);
+ }
+ else
+ result = a_norm[i] > b_norm[i] ? 1 : -1;
+ goto exit;
+ }
+ result = a_len < b_len ? -1 : a_len > b_len;
+
+exit:
+ if (a_norm != a_stub)
+ free (a_norm);
+ if (b_norm != b_stub)
+ free (b_norm);
+ return result;
+}
+
+static char *
+utf8_casemap (const char *s,
+ uint8_t *(*f) (const uint8_t *, size_t, const char *, uninorm_t,
+ uint8_t *, size_t *))
+{
+ char *result;
+ size_t size;
+
+ result = CHAR_CAST (char *,
+ f (CHAR_CAST (const uint8_t *, s), strlen (s) + 1,
+ NULL, NULL, NULL, &size));
+ if (result == NULL)
+ {
+ if (errno == ENOMEM)
+ xalloc_die ();
+
+ result = xstrdup (s);
+ }
+ return result;
+}
+
+char *
+utf8_to_upper (const char *s)
+{
+ return utf8_casemap (s, u8_toupper);
+}
+
+char *
+utf8_to_lower (const char *s)
+{
+ return utf8_casemap (s, u8_tolower);
+}
+\f
bool
get_encoding_info (struct encoding_info *e, const char *name)
{
const struct substring in = SS_LITERAL_INITIALIZER (
- "\t\n\v\f\r "
- "!\"#$%&'()*+,-./0123456789:;<=>?@"
- "ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`"
- "abcdefghijklmnopqrstuvwxyz{|}~");
+ "\t\n\v\f\r "
+ "!\"#$%&'()*+,-./0123456789:;<=>?@"
+ "ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`"
+ "abcdefghijklmnopqrstuvwxyz{|}~");
struct substring out, cr, lf, space;
bool ok;
bool
is_encoding_supported (const char *encoding)
{
- return (create_iconv__ ("UTF-8", encoding)->conv != (iconv_t) -1
- && create_iconv__ (encoding, "UTF-8")->conv != (iconv_t) -1);
+ return (create_iconv ("UTF-8", encoding)
+ && create_iconv (encoding, "UTF-8"));
}
/* Returns true if E is the name of a UTF-8 encoding.
static int n_categories;
static void SENTINEL (0)
-add_category (size_t *allocated_categories, const char *category, ...)
+ add_category (size_t *allocated_categories, const char *category, ...)
{
struct encoding_category *c;
const char *encodings[16];
projects / pspp / blobdiff