paraslash 0.7.3

[paraslash.git] / string.c

/* Copyright (C) 2004 Andre Noll <maan@tuebingen.mpg.de>, see file COPYING. */

/** \file string.c Memory allocation and string handling functions. */

#include "para.h"

#include <pwd.h>
#include <sys/utsname.h> /* uname() */
#include <regex.h>
#include <langinfo.h>
#include <wchar.h>
#include <wctype.h>

#include "string.h"
#include "error.h"

/**
 * Reallocate an array, abort on failure or bugs.
 *
 * \param ptr Pointer to the memory block, may be NULL.
 * \param nmemb Number of elements.
 * \param size The size of one element in bytes.
 *
 * A wrapper for realloc(3) which aborts on invalid arguments or integer
 * overflow. The wrapper also terminates the current process on allocation
 * errors, so the caller does not need to check for failure.
 *
 * \return A pointer to newly allocated memory which is suitably aligned for
 * any kind of variable and may be different from ptr.
 *
 * \sa realloc(3).
 */
__must_check void *arr_realloc(void *ptr, size_t nmemb, size_t size)
{
        size_t pr;

        assert(size > 0);
        assert(nmemb > 0);
        assert(!__builtin_mul_overflow(nmemb, size, &pr));
        assert(pr != 0);
        ptr = realloc(ptr, pr);
        assert(ptr);
        return ptr;
}

/**
 * Allocate an array, abort on failure or bugs.
 *
 * \param nmemb See \ref arr_realloc().
 * \param size See \ref arr_realloc().
 *
 * Like \ref arr_realloc(), this aborts on invalid arguments, integer overflow
 * and allocation errors.
 *
 * \return A pointer to newly allocated memory which is suitably aligned for
 * any kind of variable.
 *
 * \sa See \ref arr_realloc().
 */
__must_check __malloc void *arr_alloc(size_t nmemb, size_t size)
{
        return arr_realloc(NULL, nmemb, size);
}

/**
 * Allocate and initialize an array, abort on failure or bugs.
 *
 * \param nmemb See \ref arr_realloc().
 * \param size See \ref arr_realloc().
 *
 * This calls \ref arr_alloc() and zeroes-out the array.
 *
 * \return See \ref arr_alloc().
 */
__must_check __malloc void *arr_zalloc(size_t nmemb, size_t size)
{
        void *ptr = arr_alloc(nmemb, size);

        /*
         * This multiplication can not overflow because the above call to \ref
         * arr_alloc() aborts on overflow.
         */
        memset(ptr, 0, nmemb * size);
        return ptr;
}

/**
 * Allocate and initialize memory.
 *
 * \param size The desired new size.
 *
 * \return A pointer to the allocated and zeroed-out memory, which is suitably
 * aligned for any kind of variable.
 *
 * \sa \ref alloc(), calloc(3).
 */
__must_check void *zalloc(size_t size)
{
        return arr_zalloc(1, size);
}

/**
 * Paraslash's version of realloc().
 *
 * \param p Pointer to the memory block, may be \p NULL.
 * \param size The desired new size.
 *
 * A wrapper for realloc(3). It calls \p exit(\p EXIT_FAILURE) on errors,
 * i.e. there is no need to check the return value in the caller.
 *
 * \return A pointer to newly allocated memory which is suitably aligned for
 * any kind of variable and may be different from \a p.
 *
 * \sa realloc(3).
 */
__must_check void *para_realloc(void *p, size_t size)
{
        return arr_realloc(p, 1, size);
}

/**
 * Paraslash's version of malloc().
 *
 * \param size The desired new size.
 *
 * A wrapper for malloc(3) which exits on errors.
 *
 * \return A pointer to the allocated memory, which is suitably aligned for any
 * kind of variable.
 *
 * \sa malloc(3).
 */
__must_check __malloc void *alloc(size_t size)
{
        return arr_alloc(1, size);
}

/**
 * Paraslash's version of strdup().
 *
 * \param s The string to be duplicated.
 *
 * A strdup(3)-like function which aborts if insufficient memory was available
 * to allocate the duplicated string, absolving the caller from the
 * responsibility to check for failure.
 *
 * \return A pointer to the duplicated string. Unlike strdup(3), the caller may
 * pass NULL, in which case the function returns a pointer to an empty string.
 * Regardless of whether or not NULL was passed, the returned string is
 * allocated on the heap and has to be freed by the caller.
 *
 * \sa strdup(3).
 */
__must_check __malloc char *para_strdup(const char *s)
{
        char *dupped_string = strdup(s? s: "");

        assert(dupped_string);
        return dupped_string;
}

/**
 * Print a formatted message to a dynamically allocated string.
 *
 * \param result The formatted string is returned here.
 * \param fmt The format string.
 * \param ap Initialized list of arguments.
 *
 * This function is similar to vasprintf(), a GNU extension which is not in C
 * or POSIX. It allocates a string large enough to hold the output including
 * the terminating null byte. The allocated string is returned via the first
 * argument and must be freed by the caller. However, unlike vasprintf(), this
 * function calls exit() if insufficient memory is available, while vasprintf()
 * returns -1 in this case.
 *
 * \return Number of bytes written, not including the terminating \p NULL
 * character.
 *
 * \sa printf(3), vsnprintf(3), va_start(3), vasprintf(3), \ref xasprintf().
 */
__printf_2_0 unsigned xvasprintf(char **result, const char *fmt, va_list ap)
{
        int ret;
        size_t size = 150;
        va_list aq;

        *result = alloc(size + 1);
        va_copy(aq, ap);
        ret = vsnprintf(*result, size, fmt, aq);
        va_end(aq);
        assert(ret >= 0);
        if (ret < size) /* OK */
                return ret;
        size = ret + 1;
        *result = para_realloc(*result, size);
        va_copy(aq, ap);
        ret = vsnprintf(*result, size, fmt, aq);
        va_end(aq);
        assert(ret >= 0 && ret < size);
        return ret;
}

/**
 * Print to a dynamically allocated string, variable number of arguments.
 *
 * \param result See \ref xvasprintf().
 * \param fmt Usual format string.
 *
 * \return The return value of the underlying call to \ref xvasprintf().
 *
 * \sa \ref xvasprintf() and the references mentioned there.
 */
__printf_2_3 unsigned xasprintf(char **result, const char *fmt, ...)
{
        va_list ap;
        unsigned ret;

        va_start(ap, fmt);
        ret = xvasprintf(result, fmt, ap);
        va_end(ap);
        return ret;
}

/**
 * Allocate a sufficiently large string and print into it.
 *
 * \param fmt A usual format string.
 *
 * Produce output according to \p fmt. No artificial bound on the length of the
 * resulting string is imposed.
 *
 * \return This function either returns a pointer to a string that must be
 * freed by the caller or aborts without returning.
 *
 * \sa printf(3), \ref xasprintf().
 */
__must_check __printf_1_2 __malloc char *make_message(const char *fmt, ...)
{
        char *msg;
        va_list ap;

        va_start(ap, fmt);
        xvasprintf(&msg, fmt, ap);
        va_end(ap);
        return msg;
}

/**
 * Free the content of a pointer and set it to NULL.
 *
 * \param arg A pointer to the pointer whose content should be freed.
 *
 * If arg is NULL, the function returns immediately. Otherwise it frees the
 * memory pointed to by *arg and sets *arg to NULL. Hence callers have to pass
 * the *address* of the pointer variable that points to the memory which should
 * be freed.
 */
void freep(void *arg)
{
        if (arg) {
                void **ptr = arg;
                free(*ptr);
                *ptr = NULL;
        }
}

/**
 * Paraslash's version of strcat().
 *
 * \param a String to be appended to.
 * \param b String to append.
 *
 * Append \p b to \p a.
 *
 * \return If \a a is \p NULL, return a pointer to a copy of \a b, i.e.
 * para_strcat(NULL, b) is equivalent to para_strdup(b). If \a b is \p NULL,
 * return \a a without making a copy of \a a.  Otherwise, construct the
 * concatenation \a c, free \a a (but not \a b) and return \a c.
 *
 * \sa strcat(3).
 */
__must_check __malloc char *para_strcat(char *a, const char *b)
{
        char *tmp;

        if (!a)
                return para_strdup(b);
        if (!b)
                return a;
        tmp = make_message("%s%s", a, b);
        free(a);
        return tmp;
}

/**
 * Get the logname of the current user.
 *
 * \return A dynamically allocated string that must be freed by the caller. On
 * errors, the string "unknown_user" is returned, i.e. this function never
 * returns \p NULL.
 *
 * \sa getpwuid(3).
 */
__must_check __malloc char *para_logname(void)
{
        struct passwd *pw = getpwuid(getuid());
        return para_strdup(pw? pw->pw_name : "unknown_user");
}

/**
 * Get the home directory of the current user.
 *
 * \return A dynamically allocated string that must be freed by the caller. If
 * the home directory could not be found, this function returns "/tmp".
 */
__must_check __malloc char *para_homedir(void)
{
        struct passwd *pw = getpwuid(getuid());
        return para_strdup(pw? pw->pw_dir : "/tmp");
}

/**
 * Get the own hostname.
 *
 * \return A dynamically allocated string containing the hostname.
 *
 * \sa uname(2).
 */
__malloc char *para_hostname(void)
{
        struct utsname u;

        uname(&u);
        return para_strdup(u.nodename);
}

/**
 * Call a custom function for each complete line.
 *
 * \param flags Any combination of flags defined in \ref for_each_line_flags.
 * \param buf The buffer containing data separated by newlines.
 * \param size The number of bytes in \a buf.
 * \param line_handler The custom function.
 * \param private_data Pointer passed to \a line_handler.
 *
 * For each complete line in \p buf, \p line_handler is called. The first
 * argument to \p line_handler is (a copy of) the current line, and \p
 * private_data is passed as the second argument.  If the \p FELF_READ_ONLY
 * flag is unset, a pointer into \a buf is passed to the line handler,
 * otherwise a pointer to a copy of the current line is passed instead. This
 * copy is freed immediately after the line handler returns.
 *
 * The function returns if \p line_handler returns a negative value or no more
 * lines are in the buffer.  The rest of the buffer (last chunk containing an
 * incomplete line) is moved to the beginning of the buffer if FELF_READ_ONLY is
 * unset.
 *
 * \return On success this function returns the number of bytes not handled to
 * \p line_handler. The only possible error is a negative return value from the
 * line handler. In this case processing stops and the return value of the line
 * handler is returned to indicate failure.
 *
 * \sa \ref for_each_line_flags.
 */
int for_each_line(unsigned flags, char *buf, size_t size,
                line_handler_t *line_handler, void *private_data)
{
        char *start = buf, *end;
        int ret, i, num_lines = 0;

//      PARA_NOTICE_LOG("buf: %s\n", buf);
        while (start < buf + size) {
                char *next_null;
                char *next_cr;

                next_cr = memchr(start, '\n', buf + size - start);
                next_null = memchr(start, '\0', next_cr?
                        next_cr - start : buf + size - start);
                if (!next_cr && !next_null)
                        break;
                if (next_null)
                        end = next_null;
                else
                        end = next_cr;
                num_lines++;
                if (!(flags & FELF_DISCARD_FIRST) || start != buf) {
                        if (flags & FELF_READ_ONLY) {
                                size_t s = end - start;
                                char *b = alloc(s + 1);
                                memcpy(b, start, s);
                                b[s] = '\0';
                                ret = line_handler(b, private_data);
                                free(b);
                        } else {
                                *end = '\0';
                                ret = line_handler(start, private_data);
                        }
                        if (ret < 0)
                                return ret;
                }
                start = ++end;
        }
        i = buf + size - start;
        if (i && i != size && !(flags & FELF_READ_ONLY))
                memmove(buf, start, i);
        return i;
}

/** Return the hex characters of the lower 4 bits. */
#define hex(a) (hexchar[(a) & 15])

static void write_size_header(char *buf, int n)
{
        static char hexchar[] = "0123456789abcdef";

        buf[0] = hex(n >> 12);
        buf[1] = hex(n >> 8);
        buf[2] = hex(n >> 4);
        buf[3] = hex(n);
        buf[4] = ' ';
}

/**
 * Read a four-byte hex-number and return its value.
 *
 * Each status item sent by para_server is prefixed with such a hex number in
 * ASCII which describes the size of the status item.
 *
 * \param buf The buffer which must be at least four bytes long.
 *
 * \return The value of the hex number on success, \p -E_SIZE_PREFIX if the
 * buffer did not contain only hex digits.
 */
int read_size_header(const char *buf)
{
        int i, len = 0;

        for (i = 0; i < 4; i++) {
                unsigned char c = buf[i];
                len <<= 4;
                if (c >= '0' && c <= '9') {
                        len += c - '0';
                        continue;
                }
                if (c >= 'a' && c <= 'f') {
                        len += c - 'a' + 10;
                        continue;
                }
                return -E_SIZE_PREFIX;
        }
        if (buf[4] != ' ')
                return -E_SIZE_PREFIX;
        return len;
}

/**
 * Safely print into a buffer at a given offset.
 *
 * \param b Determines the buffer, its size, and the offset.
 * \param fmt The format string.
 *
 * This function prints into the buffer given by \a b at the offset which is
 * also given by \a b. If there is not enough space to hold the result, the
 * buffer size is doubled until the underlying call to vsnprintf() succeeds
 * or the size of the buffer exceeds the maximal size specified in \a b.
 *
 * In the latter case the unmodified \a buf and \a offset values as well as the
 * private_data pointer of \a b are passed to the \a max_size_handler of \a b.
 * If this function succeeds, i.e. returns a non-negative value, the offset of
 * \a b is reset to zero and the given data is written to the beginning of the
 * buffer. If \a max_size_handler() returns a negative value, this value is
 * returned by \a para_printf().
 *
 * Upon return, the offset of \a b is adjusted accordingly so that subsequent
 * calls to this function append data to what is already contained in the
 * buffer.
 *
 * It's OK to call this function with \p b->buf being \p NULL. In this case, an
 * initial buffer is allocated.
 *
 * \return The number of bytes printed into the buffer (not including the
 * terminating \p NULL byte) on success, negative on errors. If there is no
 * size-bound on \a b, i.e. if \p b->max_size is zero, this function never
 * fails.
 *
 * \sa make_message(), vsnprintf(3).
 */
__printf_2_3 int para_printf(struct para_buffer *b, const char *fmt, ...)
{
        int ret, sz_off = (b->flags & PBF_SIZE_PREFIX)? 5 : 0;

        if (!b->buf) {
                b->buf = alloc(128);
                b->size = 128;
                b->offset = 0;
        }
        while (1) {
                char *p = b->buf + b->offset;
                size_t size = b->size - b->offset;
                va_list ap;

                if (size > sz_off) {
                        va_start(ap, fmt);
                        ret = vsnprintf(p + sz_off, size - sz_off, fmt, ap);
                        va_end(ap);
                        if (ret > -1 && ret < size - sz_off) { /* success */
                                b->offset += ret + sz_off;
                                if (sz_off)
                                        write_size_header(p, ret);
                                return ret + sz_off;
                        }
                }
                /* check if we may grow the buffer */
                if (!b->max_size || 2 * b->size < b->max_size) { /* yes */
                        /* try again with more space */
                        b->size *= 2;
                        b->buf = para_realloc(b->buf, b->size);
                        continue;
                }
                /* can't grow buffer */
                if (!b->offset || !b->max_size_handler) /* message too large */
                        return -ERRNO_TO_PARA_ERROR(ENOSPC);
                ret = b->max_size_handler(b->buf, b->offset, b->private_data);
                if (ret < 0)
                        return ret;
                b->offset = 0;
        }
}

/** \cond llong_minmax */
/* LLONG_MAX and LLONG_MIN might not be defined. */
#ifndef LLONG_MAX
#define LLONG_MAX 9223372036854775807LL
#endif
#ifndef LLONG_MIN
#define LLONG_MIN (-LLONG_MAX - 1LL)
#endif
/** \endcond llong_minmax */

/**
 * Convert a string to a 64-bit signed integer value.
 *
 * \param str The string to be converted.
 * \param value Result pointer.
 *
 * \return Standard.
 *
 * \sa \ref para_atoi32(), strtol(3), atoi(3).
 */
int para_atoi64(const char *str, int64_t *value)
{
        char *endptr;
        long long tmp;

        errno = 0; /* To distinguish success/failure after call */
        tmp = strtoll(str, &endptr, 10);
        if (errno == ERANGE && (tmp == LLONG_MAX || tmp == LLONG_MIN))
                return -E_ATOI_OVERFLOW;
        /*
         * If there were no digits at all, strtoll() stores the original value
         * of str in *endptr.
         */
        if (endptr == str)
                return -E_ATOI_NO_DIGITS;
        /*
         * The implementation may also set errno and return 0 in case no
         * conversion was performed.
         */
        if (errno != 0 && tmp == 0)
                return -E_ATOI_NO_DIGITS;
        if (*endptr != '\0') /* Further characters after number */
                return -E_ATOI_JUNK_AT_END;
        *value = tmp;
        return 1;
}

/**
 * Convert a string to a 32-bit signed integer value.
 *
 * \param str The string to be converted.
 * \param value Result pointer.
 *
 * \return Standard.
 *
 * \sa \ref para_atoi64().
*/
int para_atoi32(const char *str, int32_t *value)
{
        int64_t tmp;
        int ret;
        const int32_t max = 2147483647;

        ret = para_atoi64(str, &tmp);
        if (ret < 0)
                return ret;
        if (tmp > max || tmp < -max - 1)
                return -E_ATOI_OVERFLOW;
        *value = tmp;
        return 1;
}

static int get_next_word(const char *buf, const char *delim, char **word)
{
        enum line_state_flags {LSF_HAVE_WORD = 1, LSF_BACKSLASH = 2,
                LSF_SINGLE_QUOTE = 4, LSF_DOUBLE_QUOTE = 8};
        const char *in;
        char *out;
        int ret, state = 0;

        out = alloc(strlen(buf) + 1);
        *out = '\0';
        *word = out;
        for (in = buf; *in; in++) {
                const char *p;

                switch (*in) {
                case '\\':
                        if (state & LSF_BACKSLASH) /* \\ */
                                goto copy_char;
                        state |= LSF_BACKSLASH;
                        state |= LSF_HAVE_WORD;
                        continue;
                case 'n':
                case 't':
                        if (state & LSF_BACKSLASH) { /* \n or \t */
                                *out++ = (*in == 'n')? '\n' : '\t';
                                state &= ~LSF_BACKSLASH;
                                continue;
                        }
                        goto copy_char;
                case '"':
                        if (state & LSF_BACKSLASH) /* \" */
                                goto copy_char;
                        if (state & LSF_SINGLE_QUOTE) /* '" */
                                goto copy_char;
                        if (state & LSF_DOUBLE_QUOTE) {
                                state &= ~LSF_DOUBLE_QUOTE;
                                continue;
                        }
                        state |= LSF_HAVE_WORD;
                        state |= LSF_DOUBLE_QUOTE;
                        continue;
                case '\'':
                        if (state & LSF_BACKSLASH) /* \' */
                                goto copy_char;
                        if (state & LSF_DOUBLE_QUOTE) /* "' */
                                goto copy_char;
                        if (state & LSF_SINGLE_QUOTE) {
                                state &= ~LSF_SINGLE_QUOTE;
                                continue;
                        }
                        state |= LSF_HAVE_WORD;
                        state |= LSF_SINGLE_QUOTE;
                        continue;
                }
                for (p = delim; *p; p++) {
                        if (*in != *p)
                                continue;
                        if (state & LSF_BACKSLASH)
                                goto copy_char;
                        if (state & LSF_SINGLE_QUOTE)
                                goto copy_char;
                        if (state & LSF_DOUBLE_QUOTE)
                                goto copy_char;
                        if (state & LSF_HAVE_WORD)
                                goto success;
                        break;
                }
                if (*p) /* ignore delimiter at the beginning */
                        continue;
copy_char:
                state |= LSF_HAVE_WORD;
                *out++ = *in;
                state &= ~LSF_BACKSLASH;
        }
        ret = 0;
        if (!(state & LSF_HAVE_WORD))
                goto out;
        ret = -ERRNO_TO_PARA_ERROR(EINVAL);
        if (state & LSF_BACKSLASH) {
                PARA_ERROR_LOG("trailing backslash\n");
                goto out;
        }
        if ((state & LSF_SINGLE_QUOTE) || (state & LSF_DOUBLE_QUOTE)) {
                PARA_ERROR_LOG("unmatched quote character\n");
                goto out;
        }
success:
        *out = '\0';
        return in - buf;
out:
        free(*word);
        *word = NULL;
        return ret;
}

/**
 * Get the number of the word the cursor is on.
 *
 * \param buf The zero-terminated line buffer.
 * \param delim Characters that separate words.
 * \param point The cursor position.
 *
 * \return Zero-based word number.
 */
int compute_word_num(const char *buf, const char *delim, int point)
{
        int ret, num_words;
        const char *p;
        char *word;

        for (p = buf, num_words = 0; ; p += ret, num_words++) {
                ret = get_next_word(p, delim, &word);
                if (ret <= 0)
                        break;
                free(word);
                if (p + ret >= buf + point)
                        break;
        }
        return num_words;
}

/**
 * Free an array of words created by create_argv() or create_shifted_argv().
 *
 * \param argv A pointer previously obtained by \ref create_argv().
 */
void free_argv(char **argv)
{
        int i;

        if (!argv)
                return;
        for (i = 0; argv[i]; i++)
                free(argv[i]);
        free(argv);
}

static int create_argv_offset(int offset, const char *buf, const char *delim,
                char ***result)
{
        char *word, **argv = arr_alloc(offset + 1, sizeof(char *));
        const char *p;
        int i, ret;

        for (i = 0; i < offset; i++)
                argv[i] = NULL;
        for (p = buf; p && *p; p += ret, i++) {
                ret = get_next_word(p, delim, &word);
                if (ret < 0)
                        goto err;
                if (!ret)
                        break;
                argv = arr_realloc(argv, i + 2, sizeof(char*));
                argv[i] = word;
        }
        argv[i] = NULL;
        *result = argv;
        return i;
err:
        while (i > 0)
                free(argv[--i]);
        free(argv);
        *result = NULL;
        return ret;
}

/**
 * Split a buffer into words.
 *
 * This parser honors single and double quotes, backslash-escaped characters
 * and special characters like \\n. The result contains pointers to copies of
 * the words contained in buf and has to be freed by using \ref free_argv().
 *
 * \param buf The buffer to be split.
 * \param delim Each character in this string is treated as a separator.
 * \param result The array of words is returned here.
 *
 * It's OK to pass NULL as the buffer argument. This is equivalent to passing
 * the empty string.
 *
 * \return Number of words in buf, negative on errors. The array returned
 * through the result pointer is NULL terminated.
 */
int create_argv(const char *buf, const char *delim, char ***result)
{
        return create_argv_offset(0, buf, delim, result);
}

/**
 * Split a buffer into words, offset one.
 *
 * This is similar to \ref create_argv() but the returned array is one element
 * larger, words start at index one and element zero is initialized to \p NULL.
 * Callers must set element zero to a non-NULL value before calling free_argv()
 * on the returned array to avoid a memory leak.
 *
 * \param buf See \ref create_argv().
 * \param delim See \ref create_argv().
 * \param result See \ref create_argv().
 *
 * \return Number of words plus one on success, negative on errors.
 */
int create_shifted_argv(const char *buf, const char *delim, char ***result)
{
        return create_argv_offset(1, buf, delim, result);
}

/**
 * Compile a regular expression.
 *
 * This simple wrapper calls regcomp() and logs a message on errors.
 *
 * \param preg See regcomp(3).
 * \param regex See regcomp(3).
 * \param cflags See regcomp(3).
 *
 * \return Standard.
 */
int para_regcomp(regex_t *preg, const char *regex, int cflags)
{
        char *buf;
        size_t size;
        int ret = regcomp(preg, regex, cflags);

        if (ret == 0)
                return 1;
        size = regerror(ret, preg, NULL, 0);
        buf = alloc(size);
        regerror(ret, preg, buf, size);
        PARA_ERROR_LOG("%s\n", buf);
        free(buf);
        return -E_REGEX;
}

/**
 * strdup() for not necessarily zero-terminated strings.
 *
 * \param src The source buffer.
 * \param len The number of bytes to be copied.
 *
 * \return A 0-terminated buffer of length \a len + 1.
 *
 * This is similar to strndup(), which is a GNU extension. However, one
 * difference is that strndup() returns \p NULL if insufficient memory was
 * available while this function aborts in this case.
 *
 * \sa strdup(), \ref para_strdup().
 */
char *safe_strdup(const char *src, size_t len)
{
        char *p;

        assert(len < (size_t)-1);
        p = alloc(len + 1);
        if (len > 0)
                memcpy(p, src, len);
        p[len] = '\0';
        return p;
}

/**
 * Copy the value of a key=value pair.
 *
 * This checks whether the given buffer starts with "key=", ignoring case. If
 * yes, a copy of the value is returned. The source buffer may not be
 * zero-terminated.
 *
 * \param src The source buffer.
 * \param len The number of bytes of the tag.
 * \param key Only copy if it is the value of this key.
 *
 * \return A zero-terminated buffer, or \p NULL if the key was
 * not of the given type.
 */
char *key_value_copy(const char *src, size_t len, const char *key)
{
        int keylen = strlen(key);

        if (len <= keylen)
                return NULL;
        if (strncasecmp(src, key, keylen))
                return NULL;
        if (src[keylen] != '=')
                return NULL;
        return safe_strdup(src + keylen + 1, len - keylen - 1);
}

static bool utf8_mode(void)
{
        static bool initialized, have_utf8;

        if (!initialized) {
                char *info = nl_langinfo(CODESET);
                have_utf8 = (info && strcmp(info, "UTF-8") == 0);
                initialized = true;
                PARA_INFO_LOG("%susing UTF-8 character encoding\n",
                        have_utf8? "" : "not ");
        }
        return have_utf8;
}

static int xwcwidth(wchar_t wc, size_t pos)
{
        int n;

        /* special-case for tab */
        if (wc == 0x09) /* tab */
                return (pos | 7) + 1 - pos;
        n = wcwidth(wc);
        /* wcswidth() returns -1 for non-printable characters */
        return n >= 0? n : 1;
}

static size_t xwcswidth(const wchar_t *s, size_t n)
{
        size_t w = 0;

        while (n--)
                w += xwcwidth(*s++, w);
        return w;
}

/**
 * Skip a given number of cells at the beginning of a string.
 *
 * \param s The input string.
 * \param cells_to_skip Desired number of cells that should be skipped.
 * \param bytes_to_skip Result.
 *
 * This function computes how many input bytes must be skipped to advance a
 * string by the given width. If the current character encoding is not UTF-8,
 * this is simply the given number of cells, i.e. \a cells_to_skip. Otherwise,
 * \a s is treated as a multibyte string and on successful return, \a s +
 * bytes_to_skip points to the start of a multibyte string such that the total
 * width of the multibyte characters that are skipped by advancing \a s that
 * many bytes equals at least \a cells_to_skip.
 *
 * \return Standard.
 */
int skip_cells(const char *s, size_t cells_to_skip, size_t *bytes_to_skip)
{
        wchar_t wc;
        mbstate_t ps;
        size_t n, bytes_parsed, cells_skipped;

        *bytes_to_skip = 0;
        if (cells_to_skip == 0)
                return 0;
        if (!utf8_mode()) {
                *bytes_to_skip = cells_to_skip;
                return 0;
        }
        bytes_parsed = cells_skipped = 0;
        memset(&ps, 0, sizeof(ps));
        n = strlen(s);
        while (cells_to_skip > cells_skipped) {
                size_t mbret;

                mbret = mbrtowc(&wc, s + bytes_parsed, n - bytes_parsed, &ps);
                assert(mbret != 0);
                if (mbret == (size_t)-1 || mbret == (size_t)-2)
                        return -ERRNO_TO_PARA_ERROR(EILSEQ);
                bytes_parsed += mbret;
                cells_skipped += xwcwidth(wc, cells_skipped);
        }
        *bytes_to_skip = bytes_parsed;
        return 1;
}

/**
 * Compute the width of an UTF-8 string.
 *
 * \param s The string.
 * \param result The width of \a s is returned here.
 *
 * If not in UTF8-mode. this function is just a wrapper for strlen(3).
 * Otherwise \a s is treated as an UTF-8 string and its display width is
 * computed. Note that this function may fail if the underlying call to
 * mbsrtowcs(3) fails, so the caller must check the return value.
 *
 * \sa nl_langinfo(3), wcswidth(3).
 *
 * \return Standard.
 */
__must_check int strwidth(const char *s, size_t *result)
{
        const char *src = s;
        mbstate_t state;
        static wchar_t *dest;
        size_t num_wchars;

        /*
         * Never call any log function here. This may result in an endless loop
         * as para_gui's para_log() calls this function.
         */

        if (!utf8_mode()) {
                *result = strlen(s);
                return 0;
        }
        memset(&state, 0, sizeof(state));
        *result = 0;
        num_wchars = mbsrtowcs(NULL, &src, 0, &state);
        if (num_wchars == (size_t)-1)
                return -ERRNO_TO_PARA_ERROR(errno);
        if (num_wchars == 0)
                return 0;
        dest = arr_alloc(num_wchars + 1, sizeof(*dest));
        src = s;
        memset(&state, 0, sizeof(state));
        num_wchars = mbsrtowcs(dest, &src, num_wchars, &state);
        assert(num_wchars > 0 && num_wchars != (size_t)-1);
        *result = xwcswidth(dest, num_wchars);
        free(dest);
        return 1;
}

/**
 * Truncate and sanitize a (wide character) string.
 *
 * This replaces all non-printable characters by spaces and makes sure that the
 * modified string does not exceed the given maximal width.
 *
 * \param src The source string in multi-byte form.
 * \param max_width The maximal number of cells the result may occupy.
 * \param result Sanitized multi-byte string, must be freed by caller.
 * \param width The width of the sanitized string, always <= max_width.
 *
 * The function is wide-character aware but falls back to C strings for
 * non-UTF-8 locales.
 *
 * \return Standard. On success, *result points to a sanitized copy of the
 * given string. This copy was allocated with malloc() and should hence be
 * freed when the caller is no longer interested in the result.
 *
 * The function fails if the given string contains an invalid multibyte
 * sequence. In this case, *result is set to NULL, and *width to zero.
 */
__must_check int sanitize_str(const char *src, size_t max_width,
                char **result, size_t *width)
{
        mbstate_t state;
        static wchar_t *wcs;
        size_t num_wchars, n;

        if (!utf8_mode()) {
                *result = para_strdup(src);
                /* replace non-printable characters by spaces */
                for (n = 0; n < max_width && src[n]; n++) {
                        if (!isprint((unsigned char)src[n]))
                                (*result)[n] = ' ';
                }
                (*result)[n] = '\0';
                *width = n;
                return 0;
        }
        *result = NULL;
        *width = 0;
        memset(&state, 0, sizeof(state));
        num_wchars = mbsrtowcs(NULL, &src, 0, &state);
        if (num_wchars == (size_t)-1)
                return -ERRNO_TO_PARA_ERROR(errno);
        wcs = arr_alloc(num_wchars + 1, sizeof(*wcs));
        memset(&state, 0, sizeof(state));
        num_wchars = mbsrtowcs(wcs, &src, num_wchars + 1, &state);
        assert(num_wchars != (size_t)-1);
        for (n = 0; n < num_wchars && *width < max_width; n++) {
                if (!iswprint(wcs[n]))
                        wcs[n] = L' ';
                *width += xwcwidth(wcs[n], *width);
        }
        wcs[n] = L'\0';
        n = wcstombs(NULL, wcs, 0) + 1;
        *result = alloc(n);
        num_wchars = wcstombs(*result, wcs, n);
        assert(num_wchars != (size_t)-1);
        free(wcs);
        return 1;
}