sbuff.c

Go to the documentation of this file.

/*
 *   This library is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU Lesser General Public
 *   License as published by the Free Software Foundation; either
 *   version 2.1 of the License, or (at your option) any later version.
 *
 *   This library is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
 *   Lesser General Public License for more details.
 *
 *   You should have received a copy of the GNU Lesser General Public
 *   License along with this library; if not, write to the Free Software
 *   Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
 */
 
/** A generic string buffer structure for string printing and parsing
 *
 * @file src/lib/util/sbuff.c
 *
 * @copyright 2020 Arran Cudbard-Bell <a.cudbardb@freeradius.org>
 */
RCSID("$Id: 3b9ede50b7229da25ea38aa61238fc75956e57cc $")
 
#include <freeradius-devel/util/misc.h>
#include <freeradius-devel/util/syserror.h>
#include <freeradius-devel/util/atexit.h>
 
 
static _Thread_local char *sbuff_scratch;
 
/** When true, prevent use of the scratch space
 *
 * This prevents us from initialising a pool after the thread local destructors have run.
 *
 * The destructors may be called manually before thread exit, and we don't want to re-initialise the pool
 */
static _Thread_local bool sbuff_scratch_freed;
 
static_assert(sizeof(long long) >= sizeof(int64_t), "long long must be as wide or wider than an int64_t");
static_assert(sizeof(unsigned long long) >= sizeof(uint64_t), "long long must be as wide or wider than an uint64_t");
 
fr_table_num_ordered_t const sbuff_parse_error_table[] = {
        { L("ok"),                      FR_SBUFF_PARSE_OK                               },
        { L("token not found"),         FR_SBUFF_PARSE_ERROR_NOT_FOUND                  },
        { L("trailing data"),           FR_SBUFF_PARSE_ERROR_TRAILING                   },
        { L("token format invalid"),    FR_SBUFF_PARSE_ERROR_FORMAT                     },
        { L("out of space"),            FR_SBUFF_PARSE_ERROR_OUT_OF_SPACE               },
        { L("integer overflow"),        FR_SBUFF_PARSE_ERROR_NUM_OVERFLOW               },
        { L("integer underflow"),       FR_SBUFF_PARSE_ERROR_NUM_UNDERFLOW              },
        { L("empty input is invalid"),  FR_SBUFF_PARSE_ERROR_INPUT_EMPTY                },
};
size_t sbuff_parse_error_table_len = NUM_ELEMENTS(sbuff_parse_error_table);
 
#if defined(STATIC_ANALYZER) || !defined(NDEBUG)
#  define CHECK_SBUFF_INIT(_sbuff)      do { if (!(_sbuff)->extend && (unlikely(!(_sbuff)->buff) || unlikely(!(_sbuff)->start) || unlikely(!(_sbuff)->end) || unlikely(!(_sbuff)->p))) return 0; } while (0)
#  define CHECK_SBUFF_WRITEABLE(_sbuff) do { CHECK_SBUFF_INIT(_sbuff); if (unlikely((_sbuff)->is_const)) return 0; } while (0)
 
#else
#  define CHECK_SBUFF_INIT(_sbuff)
#  define CHECK_SBUFF_WRITEABLE(_sbuff)
#endif
 
bool const sbuff_char_class_uint[SBUFF_CHAR_CLASS] = {
        SBUFF_CHAR_CLASS_NUM,
        ['+'] = true
};
 
bool const sbuff_char_class_int[SBUFF_CHAR_CLASS] = {
        SBUFF_CHAR_CLASS_NUM,
        ['+'] = true, ['-'] = true
};
 
bool const sbuff_char_class_float[SBUFF_CHAR_CLASS] = {
        SBUFF_CHAR_CLASS_NUM,
        ['-'] = true, ['+'] = true, ['e'] = true, ['E'] = true, ['.'] = true,
};
 
bool const sbuff_char_class_zero[SBUFF_CHAR_CLASS] = {
        ['0'] = true
};
 
/*
 *      Anything which vaguely resembles an IP address, prefix, or host name.
 */
bool const sbuff_char_class_hostname[SBUFF_CHAR_CLASS] = {
        SBUFF_CHAR_CLASS_ALPHA_NUM,
        ['.'] = true,           /* only for IPv4 and host names */
        [':'] = true,           /* only for IPv6 numerical addresses */
        ['-'] = true,           /* only for host names */
        ['/'] = true,           /* only for prefixes */
        ['['] = true,           /* only for IPv6 numerical addresses */
        [']'] = true,           /* only for IPv6 numerical addresses */
        ['_'] = true,           /* only for certain host name labels */
        ['*'] = true,           /* really only for ipv4 addresses */
};
 
bool const sbuff_char_class_hex[SBUFF_CHAR_CLASS] = { SBUFF_CHAR_CLASS_HEX };
bool const sbuff_char_alpha_num[SBUFF_CHAR_CLASS] = { SBUFF_CHAR_CLASS_ALPHA_NUM };
bool const sbuff_char_word[SBUFF_CHAR_CLASS] = {
        SBUFF_CHAR_CLASS_ALPHA_NUM,
        ['-'] = true, ['_'] = true,
};
bool const sbuff_char_whitespace[SBUFF_CHAR_CLASS] = {
        ['\t'] = true, ['\n'] = true, ['\r'] = true, ['\f'] = true, ['\v'] = true, [' '] = true,
};
 
bool const sbuff_char_line_endings[SBUFF_CHAR_CLASS] = {
        ['\n'] = true, ['\r'] = true
};
 
bool const sbuff_char_blank[SBUFF_CHAR_CLASS] = {
        ['\t'] = true, [' '] = true,
};
 
/** Copy function that allows overlapping memory ranges to be copied
 *
 * @param[out] o_start          start of output buffer.
 * @param[in] o_end             end of the output buffer.
 * @param[in] i_start           start of the input buffer.
 * @param[in] i_end             end of data to copy.
 * @return
 *      - >0 the number of bytes copied.
 *      - 0 invalid args.
 *      - -1 the copy would not fit in the output buffer.
 */
static inline CC_HINT(always_inline) ssize_t safecpy(char *o_start, char *o_end,
                                                     char const *i_start, char const *i_end)
{
        ssize_t diff;
        size_t  i_len = i_end - i_start;
 
        if (unlikely((o_end < o_start) || (i_end < i_start))) return 0; /* sanity check */
 
        diff = (o_end - o_start) - (i_len);
        if (diff < 0) return -1;
 
        if ((i_start > o_end) || (i_end < o_start)) {                   /* no-overlap */
                memcpy(o_start,  i_start, i_len);
        } else {                                                        /* overlap */
                memmove(o_start, i_start, i_len);
        }
 
        return (i_len);
}
 
static inline CC_HINT(always_inline) size_t min(size_t x, size_t y)
{
        return x < y ? x : y;
}
 
/** Update all markers and pointers in the set of sbuffs to point to new_buff
 *
 * This function should be used if the underlying buffer is realloced.
 *
 * @param[in] sbuff     to update.
 * @param[in] new_buff  to assign to to sbuff.
 * @param[in] new_len   Length of the new buffer.
 */
void fr_sbuff_update(fr_sbuff_t *sbuff, char *new_buff, size_t new_len)
{
        fr_sbuff_t              *sbuff_i;
        char                    *old_buff;      /* Current buff */
 
        old_buff = sbuff->buff;
 
        /*
         *      Update pointers to point to positions
         *      in new buffer based on their relative
         *      offsets in the old buffer... but not
         *      past the end of the new buffer.
         */
        for (sbuff_i = sbuff; sbuff_i; sbuff_i = sbuff_i->parent) {
                fr_sbuff_marker_t       *m_i;
 
                sbuff_i->buff = new_buff;
                sbuff_i->start = new_buff + min(new_len, sbuff_i->start - old_buff);
                sbuff_i->end = sbuff_i->buff + new_len;
                *(sbuff_i->end) = '\0'; /* Re-terminate */
 
                sbuff_i->p = new_buff + min(new_len, sbuff_i->p - old_buff);
 
                for (m_i = sbuff_i->m; m_i; m_i = m_i->next) m_i->p = new_buff + min(new_len, m_i->p - old_buff);
        }
}
 
/** Shift the contents of the sbuff, returning the number of bytes we managed to shift
 *
 * @param[in] sbuff     to shift.
 * @param[in] shift     the contents of the buffer this many bytes
 *                      towards the start of the buffer.
 * @param[in] move_end  If the buffer is used for reading, then this should be true
 *                      so we cannot read passed the end of valid data.
 * @return
 *      - 0 the shift failed due to constraining pointers.
 *      - >0 the number of bytes we managed to shift pointers
 *        in the sbuff.  memmove should be used to move the
 *        existing contents of the buffer, and fill the free
 *        space at the end of the buffer with additional data.
 */
size_t fr_sbuff_shift(fr_sbuff_t *sbuff, size_t shift, bool move_end)
{
        fr_sbuff_t              *sbuff_i;
        char                    *buff, *end;            /* Current start */
        size_t                  max_shift = shift;
        bool                    reterminate = false;
 
        CHECK_SBUFF_INIT(sbuff);
 
        buff = sbuff->buff;
        end = sbuff->end;
 
        /*
         *      If the sbuff is already \0 terminated
         *      and we're not working on a const buffer
         *      then assume we need to re-terminate
         *      later.
         */
        reterminate = (sbuff->p < sbuff->end) && (*sbuff->p == '\0') && !sbuff->is_const;
 
        /*
         *      First pass: find the maximum shift, which is the minimum
         *      of the distances from buff to any of the current pointers
         *      or current pointers of markers of dbuff and its ancestors.
         *      (We're also constrained by the requested shift count.)
         */
        for (sbuff_i = sbuff; sbuff_i; sbuff_i = sbuff_i->parent) {
                fr_sbuff_marker_t *m_i;
 
                max_shift = min(max_shift, sbuff_i->p - buff);
                if (!max_shift) return 0;
 
                for (m_i = sbuff_i->m; m_i; m_i = m_i->next) {
                        max_shift = min(max_shift, m_i->p - buff);
                        if (!max_shift) return 0;
                }
        }
 
        /*
         *      Second pass: adjust pointers.
         *      The first pass means we need only subtract shift from
         *      current pointers.  Start pointers can't constrain shift,
         *      or we'd never free any space, so they require the added
         *      check.
         */
        for (sbuff_i = sbuff; sbuff_i; sbuff_i = sbuff_i->parent) {
                fr_sbuff_marker_t       *m_i;
                char                    *start = sbuff_i->start;
 
                sbuff_i->start -= min(max_shift, sbuff_i->start - buff);
                sbuff_i->p -= max_shift;
                if (move_end) sbuff_i->end -= max_shift;
                sbuff_i->shifted += (max_shift - (start - sbuff_i->start));
                for (m_i = sbuff_i->m; m_i; m_i = m_i->next) m_i->p -= max_shift;
        }
 
        /*
         *      Only memmove if the shift wasn't the
         *      entire contents of the buffer.
         */
        if ((buff + max_shift) < end) memmove(buff, buff + max_shift, end - (buff + max_shift));
 
        if (reterminate) *sbuff->p = '\0';
 
        return max_shift;
}
 
/** Refresh the buffer with more data from the file
 *
 */
size_t fr_sbuff_extend_file(fr_sbuff_extend_status_t *status, fr_sbuff_t *sbuff, size_t extension)
{
        fr_sbuff_t              *sbuff_i;
        size_t                  read, available, total_read, shift;
        fr_sbuff_uctx_file_t    *fctx;
 
        CHECK_SBUFF_INIT(sbuff);
 
        fctx = sbuff->uctx;
        if (fctx->eof) return 0;
 
        if (extension == SIZE_MAX) extension = 0;
 
        total_read = fctx->shifted + (sbuff->end - sbuff->buff);
        if (total_read >= fctx->max) {
                fr_strerror_const("Can't satisfy extension request, max bytes read");
                return 0;       /* There's no way we could satisfy the extension request */
        }
 
        /*
         *      Shift out the maximum number of bytes we can
         *      irrespective of the amount that was requested
         *      as the extension.  It's more efficient to do
         *      this than lots of small shifts, and just
         *      looking and the number of bytes used in the
         *      deepest sbuff, and using that as the shift
         *      amount, might mean we don't shift anything at
         *      all!
         *
         *      fr_sbuff_shift will cap the max shift amount,
         *      so markers and positions will remain valid for
         *      all sbuffs in the chain.
         */
        shift = fr_sbuff_current(sbuff) - fr_sbuff_buff(sbuff);
        if (shift) {
                /*
                 *      Try and shift as much as we can out
                 *      of the buffer to make space.
                 *
                 *      Note: p and markers are constraints here.
                 */
                fctx->shifted += fr_sbuff_shift(sbuff, shift, true);
        }
 
        available = fctx->buff_end - sbuff->end;
        if (available > (fctx->max - total_read)) available = fctx->max - total_read;
        if (available < extension) {
                fr_strerror_printf("Can't satisfy extension request for %zu bytes", extension);
                return 0;       /* There's no way we could satisfy the extension request */
        }
 
        read = fread(sbuff->end, 1, available, fctx->file);
        for (sbuff_i = sbuff; sbuff_i; sbuff_i = sbuff_i->parent) {
                sbuff_i->end += read;   /* Advance end, which increases fr_sbuff_remaining() */
        }
 
        /** Check for errors
         */
        if (read < available) {
                if (!feof(fctx->file)) {
                        /*
                         *      It's an error, but ferror() returns a ??? error number,
                         *      and not errno.
                         *
                         *      Posix says "The ferror() function shall not change the setting of errno if
                         *      stream is valid".  And the return value is defined to be non-zero, but with no
                         *      meaning associated with any non-zero values.
                         */
                        fr_strerror_printf("Error extending buffer: %d", ferror(fctx->file));
                        *status |= FR_SBUFF_FLAG_EXTEND_ERROR;
                        return 0;
                }
 
                fctx->eof = true;
        }
 
        return read;
}
 
/** Accessor function for the EOF state of the file extendor
 *
 */
bool fr_sbuff_eof_file(fr_sbuff_t *sbuff)
{
        fr_sbuff_uctx_file_t    *fctx = sbuff->uctx;
        return fctx->eof;
}
 
/** Reallocate the current buffer
 *
 * @param[in] status            Extend status.
 * @param[in] sbuff             to be extended.
 * @param[in] extension         How many additional bytes should be allocated
 *                              in the buffer.
 * @return
 *      - 0 the extension operation failed.
 *      - >0 the number of bytes the buffer was extended by.
 */
size_t fr_sbuff_extend_talloc(fr_sbuff_extend_status_t *status, fr_sbuff_t *sbuff, size_t extension)
{
        fr_sbuff_uctx_talloc_t  *tctx = sbuff->uctx;
        size_t                  clen, nlen, elen = extension;
        char                    *new_buff;
 
        CHECK_SBUFF_INIT(sbuff);
 
        clen = sbuff->buff ? talloc_array_length(sbuff->buff) : 0;
        /*
         *      If the current buffer size + the extension
         *      is less than init, extend the buffer to init.
         *
         *      This can happen if the buffer has been
         *      trimmed, and then additional data is added.
         */
        if ((clen + elen) < tctx->init) {
                elen = (tctx->init - clen) + 1; /* add \0 */
        /*
         *      Double the buffer size if it's more than the
         *      requested amount.
         */
        } else if (elen < clen) {
                elen = clen - 1;                /* Don't double alloc \0 */
        }
 
        /*
         *      Check we don't exceed the maximum buffer
         *      length, including the NUL byte.
         */
        if (tctx->max && ((clen + elen + 1) > tctx->max)) {
                elen = tctx->max - clen;
                if (elen == 0) {
                        fr_strerror_printf("Failed extending buffer by %zu bytes to "
                                           "%zu bytes, max is %zu bytes",
                                           extension, clen + extension, tctx->max);
                        return 0;
                }
                elen += 1;                      /* add \0 */
        }
        nlen = clen + elen;
 
        new_buff = talloc_realloc(tctx->ctx, sbuff->buff, char, nlen);
        if (unlikely(!new_buff)) {
                fr_strerror_printf("Failed extending buffer by %zu bytes to %zu bytes", elen, nlen);
                *status |= FR_SBUFF_FLAG_EXTEND_ERROR;
                return 0;
        }
 
        (void)fr_sbuff_update(sbuff, new_buff, nlen - 1);       /* Shouldn't fail as we're extending */
 
        return elen;
}
 
/** Trim a talloced sbuff to the minimum length required to represent the contained string
 *
 * @param[in] sbuff     to trim.
 * @param[in] len       Length to trim to.  Passing SIZE_MAX will
 *                      result in the buffer being trimmed to the
 *                      length of the content.
 * @return
 *      - 0 on success.
 *      - -1 on failure - markers present pointing past the end of string data.
 */
int fr_sbuff_trim_talloc(fr_sbuff_t *sbuff, size_t len)
{
        size_t                  clen = 0, nlen = 1;
        char                    *new_buff;
        fr_sbuff_uctx_talloc_t  *tctx = sbuff->uctx;
 
        CHECK_SBUFF_INIT(sbuff);
 
        if (sbuff->buff) clen = talloc_array_length(sbuff->buff);
 
        if (len != SIZE_MAX) {
                nlen += len;
        } else if (sbuff->buff){
                nlen += (sbuff->p - sbuff->start);
        }
 
        if (nlen != clen) {
                new_buff = talloc_realloc(tctx->ctx, sbuff->buff, char, nlen);
                if (unlikely(!new_buff)) {
                        fr_strerror_printf("Failed trimming buffer from %zu to %zu", clen, nlen);
                        return -1;
                }
                fr_sbuff_update(sbuff, new_buff, nlen - 1);
        }
 
        return 0;
}
 
/** Reset a talloced buffer to its initial length, clearing any data stored
 *
 * @param[in] sbuff to reset.
 * @return
 *      - 0 on success.
 *      - -1 on failure - markers present pointing past the end of string data.
 */
int fr_sbuff_reset_talloc(fr_sbuff_t *sbuff)
{
        fr_sbuff_uctx_talloc_t  *tctx = sbuff->uctx;
 
        CHECK_SBUFF_INIT(sbuff);
 
        fr_sbuff_set_to_start(sbuff);   /* Clear data */
        sbuff->m = NULL;                /* Remove any maker references */
 
        if (fr_sbuff_used(sbuff) != tctx->init) {
                char *new_buff;
 
                new_buff = talloc_realloc(tctx->ctx, sbuff->buff, char, tctx->init);
                if (!new_buff) {
                        fr_strerror_printf("Failed reallocing from %zu to %zu",
                                           talloc_array_length(sbuff->buff), tctx->init);
                        return -1;
                }
                sbuff->buff = new_buff;
                fr_sbuff_update(sbuff, new_buff, tctx->init - 1);
        }
 
        return 0;
}
 
/** Fill as much of the output buffer we can and break on partial copy
 *
 * @param[in] _out      sbuff to write to.
 * @param[in] _in       sbuff to copy from.
 * @param[in] _len      maximum amount to copy.
 */
#define FILL_OR_GOTO_DONE(_out, _in, _len) if (fr_sbuff_move(_out, _in, _len) < (size_t)(_len)) goto done
 
/** Constrain end pointer to prevent advancing more than the amount the caller specified
 *
 * @param[in] _sbuff    to constrain.
 * @param[in] _max      maximum amount to advance.
 * @param[in] _used     how much we've advanced so far.
 * @return a temporary end pointer.
 */
#define CONSTRAINED_END(_sbuff, _max, _used) \
        (((_max) - (_used)) > fr_sbuff_remaining(_sbuff) ? (_sbuff)->end : (_sbuff)->p + ((_max) - (_used)))
 
 
/** Populate a terminal index
 *
 * @param[out] needle_len       the longest needle.  Will not be set
 *                              if the terminal array is empty.
 * @param[out] idx              to populate.
 * @param[in] term              Terminals to populate the index with.
 */
static inline CC_HINT(always_inline) void fr_sbuff_terminal_idx_init(size_t *needle_len,
                                                                     uint8_t idx[static SBUFF_CHAR_CLASS],
                                                                     fr_sbuff_term_t const *term)
{
        size_t i, len, max = 0;
 
        if (!term) return;
 
        memset(idx, 0, SBUFF_CHAR_CLASS);
 
        for (i = 0; i < term->len; i++) {
                len = term->elem[i].len;
                if (len > max) max = len;
 
                idx[(uint8_t)term->elem[i].str[0]] = i + 1;
        }
 
        if (i > 0) *needle_len = max;
}
 
/** Efficient terminal string search
 *
 * Caller should ensure that a buffer extension of needle_len bytes has been requested
 * before calling this function.
 *
 * @param[in] in                Sbuff to search in.
 * @param[in] p                 Current position (may be ahead of in->p).
 * @param[in] idx               Fastpath index, populated by
 *                              fr_sbuff_terminal_idx_init.
 * @param[in] term              terminals to search in.
 * @param[in] needle_len        Length of the longest needle.
 * @return
 *      - true if found.
 *      - false if not.
 */
static inline bool fr_sbuff_terminal_search(fr_sbuff_t *in, char const *p,
                                            uint8_t idx[static SBUFF_CHAR_CLASS],
                                            fr_sbuff_term_t const *term, UNUSED size_t needle_len)
{
        uint8_t         term_idx;
 
        ssize_t         start = 0;
        ssize_t         end;
        ssize_t         mid;
 
        size_t          remaining;
 
        if (!term) return false;                        /* If there's no terminals, we don't need to search */
 
        if (p > in->end) return false; /* paranoia */
 
        /*
         *      "p" may be ahead of "in->p", as the caller can scan forward without advancing "in->p`".  So we
         *      need to measure bytes available from "p".  Othwrwise using fr_sbuff_remaining(in) would
         *      over-state the available bytes by (p - in->p) and read past in->end.
         */
        remaining = (size_t)(in->end - p);
 
        /*
         *      Special case for EOFlike states.
         *
         *      This MUST be checked before dereferencing "*p" below.  When the buffer is fully consumed, we
         *      have "p == in->end".  A dereference of "*p" is one byte past the end of the buffer, and would result in an overflow.
         */
        if (remaining == 0) {
                if (!fr_sbuff_is_extendable(in) && (idx['\0'] != 0)) return true;
                return false;
        }
 
        term_idx = idx[(uint8_t)*p];                    /* Fast path */
        if (!term_idx) return false;
 
        end = term->len - 1;
        mid = term_idx - 1;                             /* Inform the mid point from the index */
 
        while (start <= end) {
                fr_sbuff_term_elem_t const      *elem;
                size_t                          tlen;
                int                             ret;
 
                elem = &term->elem[mid];
                tlen = elem->len;
 
                ret = memcmp(p, elem->str, tlen < (size_t)remaining ? tlen : (size_t)remaining);
                if (ret == 0) {
                        /*
                         *      If we have more text than the table element, that's fine
                         */
                        if (remaining >= tlen) return true;
 
                        /*
                         *      If input was shorter than the table element we need to
                         *      keep searching.
                         */
                        ret = -1;
                }
 
                if (ret < 0) {
                        end = mid - 1;
                } else {
                        start = mid + 1;
                }
 
                mid = start + ((end - start) / 2);      /* Avoid overflow */
        }
 
        return false;
}
 
/** Compare two terminal elements for ordering purposes
 *
 * @param[in] a         first terminal to compare.
 * @param[in] b         second terminal to compare.
 * @return CMP(a,b)
 */
static inline int8_t terminal_cmp(fr_sbuff_term_elem_t const *a, fr_sbuff_term_elem_t const *b)
{
        return MEMCMP_FIELDS(a, b, str, len);
}
 
#if 0
static void fr_sbuff_terminal_debug_tmp(fr_sbuff_term_elem_t const *elem[], size_t len)
{
        size_t i;
 
        FR_FAULT_LOG("Terminal count %zu", len);
 
        for (i = 0; i < len; i++) FR_FAULT_LOG("\t\"%s\" (%zu)", elem[i] ? elem[i]->str : "NULL", elem[i] ? elem[i]->len : 0);
}
#endif
 
/** Merge two sets of terminal strings
 *
 * @param[in] ctx       to allocate the new terminal array in.
 * @param[in] a         first set of terminals to merge.
 * @param[in] b         second set of terminals to merge.
 * @return A new set of de-duplicated and sorted terminals.
 */
fr_sbuff_term_t *fr_sbuff_terminals_amerge(TALLOC_CTX *ctx, fr_sbuff_term_t const *a, fr_sbuff_term_t const *b)
{
        size_t                          i, j, num;
        fr_sbuff_term_t                 *out;
        fr_sbuff_term_elem_t const      *tmp[SBUFF_CHAR_CLASS];
 
        /*
         *      Check all inputs are pre-sorted.  It doesn't break this
         *      function, but it's useful in case the terminal arrays
         *      are defined elsewhere without merging.
         */
#if !defined(NDEBUG) && defined(WITH_VERIFY_PTR)
        if (a->len) for (i = 0; i < a->len - 1; i++) fr_assert(terminal_cmp(&a->elem[i], &a->elem[i + 1]) < 0);
        if (b->len) for (i = 0; i < b->len - 1; i++) fr_assert(terminal_cmp(&b->elem[i], &b->elem[i + 1]) < 0);
#endif
 
        /*
         *      Since the inputs are sorted, we can just do an O(n+m)
         *      walk through the arrays, comparing entries across the
         *      two arrays.
         *
         *      If there are duplicates, we prefer "a", for no particular reason.
         */
        num = i = j = 0;
        while ((i < a->len) && (j < b->len)) {
                int8_t cmp;
 
                cmp = terminal_cmp(&a->elem[i], &b->elem[j]);
                if (cmp == 0) {
                        j++;
                        tmp[num++] = &a->elem[i++];
 
                } else if (cmp < 0) {
                        tmp[num++] = &a->elem[i++];
 
                } else if (cmp > 0) {
                        tmp[num++] = &b->elem[j++];
                }
 
                fr_assert(num < SBUFF_CHAR_CLASS);
        }
 
        /*
         *      Only one of these will be hit, and it's simpler than nested "if" statements.
         */
        while (i < a->len) tmp[num++] = &a->elem[i++];
        while (j < b->len) tmp[num++] = &b->elem[j++];
 
        out = talloc_pooled_object(ctx, fr_sbuff_term_t, num, num * sizeof(fr_sbuff_term_elem_t));
        if (unlikely(!out)) return NULL;
 
        out->elem = talloc_array(out, fr_sbuff_term_elem_t, num);
        if (unlikely(!out->elem)) {
                talloc_free(out);
                return NULL;
        }
        out->len = num;
 
        for (i = 0; i < num; i++) out->elem[i] = *tmp[i]; /* copy merged results back */
 
#if !defined(NDEBUG) && defined(WITH_VERIFY_PTR)
        for (i = 0; i < num - 1; i++) fr_assert(terminal_cmp(&out->elem[i], &out->elem[i + 1]) < 0);
#endif
 
        return out;
}
 
/** Copy as many bytes as possible from a sbuff to a sbuff
 *
 * Copy size is limited by available data in sbuff and space in output sbuff.
 *
 * @param[out] out      Where to copy to.
 * @param[in] in        Where to copy from.  Will copy len bytes from current position in buffer.
 * @param[in] len       How many bytes to copy.  If SIZE_MAX the entire buffer will be copied.
 * @return
 *      - 0 no bytes copied.
 *      - >0 the number of bytes copied.
 */
size_t fr_sbuff_out_bstrncpy(fr_sbuff_t *out, fr_sbuff_t *in, size_t len)
{
        fr_sbuff_t      our_in = FR_SBUFF_BIND_CURRENT(in);
        size_t          remaining;
 
        CHECK_SBUFF_INIT(in);
 
        while (fr_sbuff_used_total(&our_in) < len) {
                size_t chunk_len;
 
                remaining = (len - fr_sbuff_used_total(&our_in));
 
                if (!fr_sbuff_extend(&our_in)) break;
 
                chunk_len = fr_sbuff_remaining(&our_in);
                if (chunk_len > remaining) chunk_len = remaining;
 
                FILL_OR_GOTO_DONE(out, &our_in, chunk_len);
        }
 
done:
        *out->p = '\0';
        return fr_sbuff_used_total(&our_in);
}
 
/** Copy exactly len bytes from a sbuff to a sbuff or fail
 *
 * Copy size is limited by available data in sbuff, space in output sbuff, and length.
 *
 * @param[out] out      Where to copy to.
 * @param[in] in        Where to copy from.  Will copy len bytes from current position in buffer.
 * @param[in] len       How many bytes to copy.  If SIZE_MAX the entire buffer will be copied.
 * @return
 *      - 0 no bytes copied, no token found of sufficient length in input buffer.
 *      - >0 the number of bytes copied.
 *      - -1 the copy would not fit in the output buffer.
 */
ssize_t fr_sbuff_out_bstrncpy_exact(fr_sbuff_t *out, fr_sbuff_t *in, size_t len)
{
        fr_sbuff_t              our_in = FR_SBUFF(in);
        size_t                  remaining;
        fr_sbuff_marker_t       m;
 
        CHECK_SBUFF_INIT(in);
 
        fr_sbuff_marker(&m, out);
 
        do {
                size_t chunk_len;
                ssize_t copied;
 
                remaining = (len - fr_sbuff_used_total(&our_in));
                if (remaining && !fr_sbuff_extend(&our_in)) {
                        fr_sbuff_marker_release(&m);
                        return 0;
                }
 
                chunk_len = fr_sbuff_remaining(&our_in);
                if (chunk_len > remaining) chunk_len = remaining;
 
                copied = fr_sbuff_in_bstrncpy(out, our_in.p, chunk_len);
                if (copied < 0) {
                        fr_sbuff_set(out, &m);          /* Reset out */
                        *m.p = '\0';                    /* Re-terminate */
 
                        fr_sbuff_marker_release(&m);
                        return -1;
                }
                fr_sbuff_advance(&our_in, copied);
        } while (fr_sbuff_used_total(&our_in) < len);
 
        fr_sbuff_marker_release(&m);
 
        FR_SBUFF_SET_RETURN(in, &our_in);       /* in was pinned, so this works */
}
 
/** Copy as many allowed characters as possible from a sbuff to a sbuff
 *
 * Copy size is limited by available data in sbuff and output buffer length.
 *
 * As soon as a disallowed character is found the copy is stopped.
 * The input sbuff will be left pointing at the first disallowed character.
 *
 * @param[out] out              Where to copy to.
 * @param[in] in                Where to copy from.  Will copy len bytes from current position in buffer.
 * @param[in] len               How many bytes to copy.  If SIZE_MAX the entire buffer will be copied.
 * @param[in] allowed           Characters to include the copy.
 * @return
 *      - 0 no bytes copied.
 *      - >0 the number of bytes copied.
 */
size_t fr_sbuff_out_bstrncpy_allowed(fr_sbuff_t *out, fr_sbuff_t *in, size_t len,
                                     bool const allowed[static SBUFF_CHAR_CLASS])
{
        fr_sbuff_t      our_in = FR_SBUFF_BIND_CURRENT(in);
 
        CHECK_SBUFF_INIT(in);
 
        while (fr_sbuff_used_total(&our_in) < len) {
                char    *p;
                char    *end;
 
                if (!fr_sbuff_extend(&our_in)) break;
 
                p = fr_sbuff_current(&our_in);
                end = CONSTRAINED_END(&our_in, len, fr_sbuff_used_total(&our_in));
 
                while ((p < end) && allowed[(uint8_t)*p]) p++;
 
                FILL_OR_GOTO_DONE(out, &our_in, p - our_in.p);
 
                if (p != end) break;            /* stopped early, break */
        }
 
done:
        *out->p = '\0';
        return fr_sbuff_used_total(&our_in);
}
 
/** Copy as many allowed characters as possible from a sbuff to a sbuff
 *
 * Copy size is limited by available data in sbuff and output buffer length.
 *
 * As soon as a disallowed character is found the copy is stopped.
 * The input sbuff will be left pointing at the first disallowed character.
 *
 * @param[out] out              Where to copy to.
 * @param[in] in                Where to copy from.  Will copy len bytes from current position in buffer.
 * @param[in] len               How many bytes to copy.  If SIZE_MAX the entire buffer will be copied.
 * @param[in] tt                Token terminals in the encompassing grammar.
 * @param[in] u_rules           If not NULL, ignore characters in the until set when
 *                              prefixed with u_rules->chr. FIXME - Should actually evaluate
 *                              u_rules fully.
 * @return
 *      - 0 no bytes copied.
 *      - >0 the number of bytes copied.
 */
size_t fr_sbuff_out_bstrncpy_until(fr_sbuff_t *out, fr_sbuff_t *in, size_t len,
                                   fr_sbuff_term_t const *tt,
                                   fr_sbuff_unescape_rules_t const *u_rules)
{
        fr_sbuff_t      our_in = FR_SBUFF_BIND_CURRENT(in);
        bool            do_escape = false;              /* Track state across extensions */
 
        uint8_t         idx[SBUFF_CHAR_CLASS];          /* Fast path index */
        size_t          needle_len = 1;
        char            escape_chr = u_rules ? u_rules->chr : '\0';
 
        CHECK_SBUFF_INIT(in);
 
        /*
         *      Initialise the fastpath index and
         *      figure out the longest needle.
         */
        fr_sbuff_terminal_idx_init(&needle_len, idx, tt);
 
        while (fr_sbuff_used_total(&our_in) < len) {
                char    *p;
                char    *end;
 
                if (fr_sbuff_extend_lowat(NULL, &our_in, needle_len) == 0) break;
 
                p = fr_sbuff_current(&our_in);
                end = CONSTRAINED_END(&our_in, len, fr_sbuff_used_total(&our_in));
 
                if (p == end) break;
 
                if (escape_chr == '\0') {
                        while ((p < end) && !fr_sbuff_terminal_search(in, p, idx, tt, needle_len)) p++;
                } else {
                        while (p < end) {
                                if (do_escape) {
                                        do_escape = false;
                                } else if (*p == escape_chr) {
                                        do_escape = true;
                                } else if (fr_sbuff_terminal_search(in, p, idx, tt, needle_len)) {
                                        break;
                                }
                                p++;
                        }
                }
 
                FILL_OR_GOTO_DONE(out, &our_in, p - our_in.p);
 
                if (p != end) break;            /* stopped early, break */
        }
 
done:
        *out->p = '\0';
        return fr_sbuff_used_total(&our_in);
}
 
/** Copy as many allowed characters as possible from a sbuff to a sbuff
 *
 * Copy size is limited by available data in sbuff and output buffer length.
 *
 * As soon as a disallowed character is found the copy is stopped.
 * The input sbuff will be left pointing at the first disallowed character.
 *
 * This de-escapes characters as they're copied out of the sbuff.
 *
 * @param[out] out              Where to copy to.
 * @param[in] in                Where to copy from.  Will copy len bytes from current position in buffer.
 * @param[in] len               How many bytes to copy.  If SIZE_MAX the entire buffer will be copied.
 * @param[in] tt                Token terminal strings in the encompassing grammar.
 * @param[in] u_rules           for processing unescape sequences.
 * @return
 *      - 0 no bytes copied.
 *      - >0 the number of bytes written to out.
 */
size_t fr_sbuff_out_unescape_until(fr_sbuff_t *out, fr_sbuff_t *in, size_t len,
                                   fr_sbuff_term_t const *tt,
                                   fr_sbuff_unescape_rules_t const *u_rules)
{
        fr_sbuff_t                      our_in;
        bool                            do_escape = false;                      /* Track state across extensions */
        fr_sbuff_marker_t               o_s;
        fr_sbuff_marker_t               c_s;
        fr_sbuff_marker_t               end;
 
        uint8_t                         idx[SBUFF_CHAR_CLASS];                  /* Fast path index */
        size_t                          needle_len = 1;
        fr_sbuff_extend_status_t        status = 0;
 
        /*
         *      If we don't need to do unescaping
         *      call a more suitable function.
         */
        if (!u_rules || (u_rules->chr == '\0')) return fr_sbuff_out_bstrncpy_until(out, in, len, tt, u_rules);
 
        CHECK_SBUFF_INIT(in);
 
        our_in = FR_SBUFF(in);
 
        /*
         *      Chunk tracking...
         */
        fr_sbuff_marker(&c_s, &our_in);
        fr_sbuff_marker(&end, &our_in);
        fr_sbuff_marker_update_end(&end, len);
 
        fr_sbuff_marker(&o_s, out);
 
        /*
         *      Initialise the fastpath index and
         *      figure out the longest needle.
         */
        fr_sbuff_terminal_idx_init(&needle_len, idx, tt);
 
        /*
         *      ...while we have remaining data
         */
        while (fr_sbuff_extend_lowat(&status, &our_in, needle_len) > 0) {
                if (fr_sbuff_was_extended(status)) fr_sbuff_marker_update_end(&end, len);
                if (fr_sbuff_diff(&our_in, &end) >= 0) break;   /* Reached the end */
 
                if (do_escape) {
                        do_escape = false;
 
                        /*
                         *      Check for \x<hex><hex>
                         */
                        if (u_rules->do_hex && fr_sbuff_is_char(&our_in, 'x')) {
                                uint8_t                 escape;
                                fr_sbuff_marker_t       m;
 
                                /*
                                 *      Any leading 'x' and subsequente hex digits have to fit within "len".
                                 *      We therefore check if there's enough room before trying to parse
                                 *      hexits.  If there's insufficient room, it's not a valid hex sequence.
                                 */
                                if ((len < 3) || (fr_sbuff_used_total(&our_in) > (len - 3))) goto check_subs;
 
                                fr_sbuff_marker(&m, &our_in);           /* allow for backtrack */
                                fr_sbuff_advance(&our_in, 1);           /* skip over the 'x' */
 
                                if (fr_sbuff_out_uint8_hex(NULL, &escape, &our_in, false) != 2) {
                                        fr_sbuff_set(&our_in, &m);      /* backtrack */
                                        fr_sbuff_marker_release(&m);
                                        goto check_subs;                /* allow sub for \x */
                                }
 
                                if (fr_sbuff_in_char(out, escape) <= 0) {
                                        fr_sbuff_set(&our_in, &m);      /* backtrack */
                                        fr_sbuff_marker_release(&m);
                                        break;
                                }
                                fr_sbuff_marker_release(&m);
                                fr_sbuff_set(&c_s, &our_in);
                                continue;
                        }
 
                        /*
                         *      Check for <oct><oct><oct>
                         */
                        if (u_rules->do_oct && fr_sbuff_is_digit(&our_in)) {
                                uint8_t                 escape;
                                fr_sbuff_marker_t       m;
 
                                /*
                                 *      The octal digits have to fit, too.  See 'x' above.
                                 */
                                if ((len < 3) || (fr_sbuff_used_total(&our_in) > (len - 3))) goto check_subs;
 
                                fr_sbuff_marker(&m, &our_in);           /* allow for backtrack */
 
                                if (fr_sbuff_out_uint8_oct(NULL, &escape, &our_in, false) != 3) {
                                        fr_sbuff_set(&our_in, &m);      /* backtrack */
                                        fr_sbuff_marker_release(&m);
                                        goto check_subs;                /* allow sub for <oct> */
                                }
 
                                if (fr_sbuff_in_char(out, escape) <= 0) {
                                        fr_sbuff_set(&our_in, &m);      /* backtrack */
                                        fr_sbuff_marker_release(&m);
                                        break;
                                }
                                fr_sbuff_marker_release(&m);
                                fr_sbuff_set(&c_s, &our_in);
                                continue;
                        }
 
                check_subs:
                        /*
                         *      Not a recognised hex or octal escape sequence
                         *      may be a substitution or a sequence that
                         *      should be copied to the output buffer.
                         */
                        {
                                uint8_t c = *fr_sbuff_current(&our_in);
 
                                if (u_rules->subs[c] == '\0') {
                                        if (u_rules->skip[c] == true) goto next;
                                        goto next_esc;
                                }
 
                                /*
                                 *      We already copied everything up
                                 *      to this point, so we can now
                                 *      write the substituted char to
                                 *      the output buffer.
                                 */
                                if (fr_sbuff_in_char(out, u_rules->subs[c]) <= 0) break;
 
                                /*
                                 *      ...and advance past the entire
                                 *      escape seq in the input buffer.
                                 */
                                fr_sbuff_advance(&our_in, 1);
                                fr_sbuff_set(&c_s, &our_in);
                                continue;
                        }
                }
 
        next_esc:
                if (*fr_sbuff_current(&our_in) == u_rules->chr) {
                        /*
                         *      Copy out any data we got before
                         *      we hit the escape char.
                         *
                         *      We need to do this before we
                         *      can write the escape char to
                         *      the output sbuff.
                         */
                        FILL_OR_GOTO_DONE(out, &c_s, fr_sbuff_behind(&c_s));
 
                        do_escape = true;
                        fr_sbuff_advance(&our_in, 1);
                        continue;
                }
 
        next:
                if (tt && fr_sbuff_terminal_search(&our_in, fr_sbuff_current(&our_in), idx, tt, needle_len)) break;
                fr_sbuff_advance(&our_in, 1);
        }
 
        /*
         *      Copy any remaining data over
         */
        FILL_OR_GOTO_DONE(out, &c_s, fr_sbuff_behind(&c_s));
 
done:
        fr_sbuff_set(in, &c_s); /* Only advance by as much as we copied */
        *out->p = '\0';
 
        return fr_sbuff_marker_release_behind(&o_s);
}
 
/** See if the string contains a truth value
 *
 * @param[out] out      Where to write boolean value.
 * @param[in] in        Where to search for a truth value.
 * @return
 *      - >0 the number of bytes consumed.
 *      - -1 no bytes copied, was not a truth value.
 */
fr_slen_t fr_sbuff_out_bool(bool *out, fr_sbuff_t *in)
{
        fr_sbuff_t our_in = FR_SBUFF(in);
 
        static bool const bool_prefix[SBUFF_CHAR_CLASS] = {
                ['t'] = true, ['T'] = true,     /* true */
                ['f'] = true, ['F'] = true,     /* false */
                ['y'] = true, ['Y'] = true,     /* yes */
                ['n'] = true, ['N'] = true,     /* no */
        };
 
        if (fr_sbuff_is_in_charset(&our_in, bool_prefix)) {
                switch (tolower(fr_sbuff_uint8(&our_in, '\0'))) {
                default:
                        break;
 
                case 't':
                        if (fr_sbuff_adv_past_strcase_literal(&our_in, "true")) {
                                *out = true;
                                FR_SBUFF_SET_RETURN(in, &our_in);
                        }
                        break;
 
                case 'f':
                        if (fr_sbuff_adv_past_strcase_literal(&our_in, "false")) {
                                *out = false;
                                FR_SBUFF_SET_RETURN(in, &our_in);
                        }
                        break;
 
                case 'y':
                        if (fr_sbuff_adv_past_strcase_literal(&our_in, "yes")) {
                                *out = true;
                                FR_SBUFF_SET_RETURN(in, &our_in);
                        }
                        break;
 
                case 'n':
                        if (fr_sbuff_adv_past_strcase_literal(&our_in, "no")) {
                                *out = false;
                                FR_SBUFF_SET_RETURN(in, &our_in);
                        }
                        break;
                }
        }
 
        *out = false;   /* Always initialise out */
 
        fr_strerror_const("Not a valid boolean value.  Accepted values are 'yes', 'no', 'true', 'false'");
 
        return -1;
}
 
/** Used to define a number parsing functions for signed integers
 *
 * @param[in] _name     Function suffix.
 * @param[in] _type     Output type.
 * @param[in] _min      value.
 * @param[in] _max      value.
 * @param[in] _max_char Maximum digits that can be used to represent an integer.
 *                      Can't use stringify because of width modifiers like 'u'
 *                      used in <stdint.h>.
 * @param[in] _base     to use.
 */
#define SBUFF_PARSE_INT_DEF(_name, _type, _min, _max, _max_char, _base) \
fr_slen_t fr_sbuff_out_##_name(fr_sbuff_parse_error_t *err, _type *out, fr_sbuff_t *in, bool no_trailing) \
{ \
        char            buff[_max_char + 1]; \
        char            *end, *a_end; \
        size_t          len; \
        long long       num; \
        _type           cast_num; \
        fr_sbuff_t      our_in = FR_SBUFF(in); \
        buff[0] = '\0'; /* clang scan */ \
        len = fr_sbuff_out_bstrncpy(&FR_SBUFF_IN(buff, sizeof(buff)), &our_in, _max_char); \
        if (len == 0) { \
                if (err) *err = (fr_sbuff_remaining(in) == 0) ? FR_SBUFF_PARSE_ERROR_INPUT_EMPTY : FR_SBUFF_PARSE_ERROR_NOT_FOUND; \
                return -1; \
        } \
        errno = 0; /* this is needed as strtoll doesn't reset errno */ \
        num = strtoll(buff, &end, _base); \
        cast_num = (_type)(num); \
        if (end == buff) { \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NOT_FOUND; \
                return -1; \
        } \
        if (num > cast_num) { \
        overflow: \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NUM_OVERFLOW; \
                *out = (_type)(_max); \
                return -1; \
        } \
        if (((errno == EINVAL) && (num == 0)) || ((errno == ERANGE) && (num == LLONG_MAX))) goto overflow; \
        if (num < cast_num) { \
        underflow: \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NUM_UNDERFLOW; \
                *out = (_type)(_min); \
                return -1; \
        } \
        if ((errno == ERANGE) && (num == LLONG_MIN)) goto underflow; \
        if (no_trailing && ((a_end = in->p + (end - buff)) < in->end)) { \
                if (isdigit((uint8_t) *a_end) || (((_base > 10) || ((_base == 0) && (len > 2) && (buff[0] == '0') && (buff[1] == 'x'))) && \
                    ((tolower((uint8_t) *a_end) >= 'a') && (tolower((uint8_t) *a_end) <= 'f')))) { \
                        if (err) *err = FR_SBUFF_PARSE_ERROR_TRAILING; \
                        *out = (_type)(_max); \
                        FR_SBUFF_ERROR_RETURN(&our_in); \
                } \
                *out = cast_num; \
        } else { \
                if (err) *err = FR_SBUFF_PARSE_OK; \
                *out = cast_num; \
        } \
        return fr_sbuff_advance(in, end - buff); /* Advance by the length strtoll gives us */ \
}
 
SBUFF_PARSE_INT_DEF(int8, int8_t, INT8_MIN, INT8_MAX, 4, 0)
SBUFF_PARSE_INT_DEF(int16, int16_t, INT16_MIN, INT16_MAX, 6, 0)
SBUFF_PARSE_INT_DEF(int32, int32_t, INT32_MIN, INT32_MAX, 11, 0)
SBUFF_PARSE_INT_DEF(int64, int64_t, INT64_MIN, INT64_MAX, 20, 0)
SBUFF_PARSE_INT_DEF(ssize, ssize_t, SSIZE_MIN, SSIZE_MAX, 20, 0)
 
/** Used to define a number parsing functions for signed integers
 *
 * @param[in] _name     Function suffix.
 * @param[in] _type     Output type.
 * @param[in] _max      value.
 * @param[in] _max_char Maximum digits that can be used to represent an integer.
 *                      Can't use stringify because of width modifiers like 'u'
 *                      used in <stdint.h>.
 * @param[in] _base     of the number being parsed, 8, 10, 16 etc...
 */
#define SBUFF_PARSE_UINT_DEF(_name, _type, _max, _max_char, _base) \
fr_slen_t fr_sbuff_out_##_name(fr_sbuff_parse_error_t *err, _type *out, fr_sbuff_t *in, bool no_trailing) \
{ \
        char                    buff[_max_char + 1]; \
        char                    *end, *a_end; \
        size_t                  len; \
        unsigned long long      num; \
        _type                   cast_num; \
        fr_sbuff_t              our_in = FR_SBUFF(in); \
        buff[0] = '\0'; /* clang scan */ \
        len = fr_sbuff_out_bstrncpy(&FR_SBUFF_IN(buff, sizeof(buff)), &our_in, _max_char); \
        if (len == 0) { \
                if (err) *err = (fr_sbuff_remaining(in) == 0) ? FR_SBUFF_PARSE_ERROR_INPUT_EMPTY : FR_SBUFF_PARSE_ERROR_NOT_FOUND; \
                return -1; \
        } \
        if (buff[0] == '-') { \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NUM_UNDERFLOW; \
                return -1; \
        } \
        errno = 0; /* this is needed as strtoull doesn't reset errno */ \
        num = strtoull(buff, &end, _base); \
        cast_num = (_type)(num); \
        if (end == buff) { \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NOT_FOUND; \
                return -1; \
        } \
        if (num > cast_num) { \
        overflow: \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NUM_OVERFLOW; \
                *out = (_type)(_max); \
                return -1; \
        } \
        if (((errno == EINVAL) && (num == 0)) || ((errno == ERANGE) && (num == ULLONG_MAX))) goto overflow; \
        if (no_trailing && ((a_end = in->p + (end - buff)) < in->end)) { \
                if (isdigit((uint8_t) *a_end) || (((_base > 10) || ((_base == 0) && (len > 2) && (buff[0] == '0') && (buff[1] == 'x'))) && \
                    ((tolower((uint8_t) *a_end) >= 'a') && (tolower((uint8_t) *a_end) <= 'f')))) { \
                        if (err) *err = FR_SBUFF_PARSE_ERROR_TRAILING; \
                        *out = (_type)(_max); \
                        FR_SBUFF_ERROR_RETURN(&our_in); \
                } \
                if (err) *err = FR_SBUFF_PARSE_OK; \
                *out = cast_num; \
        } else { \
                if (err) *err = FR_SBUFF_PARSE_OK; \
                *out = cast_num; \
        } \
        return fr_sbuff_advance(in, end - buff); /* Advance by the length strtoull gives us */ \
}
 
/* max chars here is the octal string value with prefix */
SBUFF_PARSE_UINT_DEF(uint8, uint8_t, UINT8_MAX, 4, 0)
SBUFF_PARSE_UINT_DEF(uint16, uint16_t, UINT16_MAX, 7, 0)
SBUFF_PARSE_UINT_DEF(uint32, uint32_t, UINT32_MAX, 12, 0)
SBUFF_PARSE_UINT_DEF(uint64, uint64_t, UINT64_MAX, 23, 0)
SBUFF_PARSE_UINT_DEF(size, size_t, SIZE_MAX, 23, 0)
 
SBUFF_PARSE_UINT_DEF(uint8_dec, uint8_t, UINT8_MAX, 3, 0)
SBUFF_PARSE_UINT_DEF(uint16_dec, uint16_t, UINT16_MAX, 4, 0)
SBUFF_PARSE_UINT_DEF(uint32_dec, uint32_t, UINT32_MAX, 10, 0)
SBUFF_PARSE_UINT_DEF(uint64_dec, uint64_t, UINT64_MAX, 19, 0)
SBUFF_PARSE_UINT_DEF(size_dec, size_t, SIZE_MAX, 19, 0)
 
 
SBUFF_PARSE_UINT_DEF(uint8_oct, uint8_t, UINT8_MAX, 3, 8)
SBUFF_PARSE_UINT_DEF(uint16_oct, uint16_t, UINT16_MAX, 6, 8)
SBUFF_PARSE_UINT_DEF(uint32_oct, uint32_t, UINT32_MAX, 11, 8)
SBUFF_PARSE_UINT_DEF(uint64_oct, uint64_t, UINT64_MAX, 22, 8)
SBUFF_PARSE_UINT_DEF(size_oct, size_t, SIZE_MAX, 22, 8)
 
SBUFF_PARSE_UINT_DEF(uint8_hex, uint8_t, UINT8_MAX, 2, 16)
SBUFF_PARSE_UINT_DEF(uint16_hex, uint16_t, UINT16_MAX, 4, 16)
SBUFF_PARSE_UINT_DEF(uint32_hex, uint32_t, UINT32_MAX, 8, 16)
SBUFF_PARSE_UINT_DEF(uint64_hex, uint64_t, UINT64_MAX, 16, 16)
SBUFF_PARSE_UINT_DEF(size_hex, size_t, SIZE_MAX, 22, 16)
 
/** Used to define a number parsing functions for floats
 *
 * @param[in] _name     Function suffix.
 * @param[in] _type     Output type.
 * @param[in] _func     Parsing function to use.
 * @param[in] _max_char Maximum digits that can be used to represent an integer.
 *                      Can't use stringify because of width modifiers like 'u'
 *                      used in <stdint.h>.
 */
#define SBUFF_PARSE_FLOAT_DEF(_name, _type, _func, _max_char) \
fr_slen_t fr_sbuff_out_##_name(fr_sbuff_parse_error_t *err, _type *out, fr_sbuff_t *in, bool no_trailing) \
{ \
        char            buff[_max_char + 1] = ""; \
        char            *end; \
        fr_sbuff_t      our_in = FR_SBUFF(in); \
        size_t          len; \
        _type           res; \
        len = fr_sbuff_out_bstrncpy_allowed(&FR_SBUFF_OUT(buff, sizeof(buff)), &our_in, SIZE_MAX, sbuff_char_class_float); \
        if (len == sizeof(buff)) { \
                if (err) *err = FR_SBUFF_PARSE_ERROR_NOT_FOUND; \
                return -1; \
        } else if (len == 0) { \
                if (err) *err = (fr_sbuff_remaining(in) == 0) ? FR_SBUFF_PARSE_ERROR_INPUT_EMPTY : FR_SBUFF_PARSE_ERROR_NOT_FOUND; \
                return -1; \
        } \
        errno = 0; /* this is needed as parsing functions don't reset errno */ \
        res = _func(buff, &end); \
        if (errno == ERANGE) { \
                if (res > 0) { \
                        if (err) *err = FR_SBUFF_PARSE_ERROR_NUM_OVERFLOW; \
                } else { \
                        if (err) *err = FR_SBUFF_PARSE_ERROR_NUM_UNDERFLOW; \
                } \
                return -1; \
        } \
        if (no_trailing && (*end != '\0')) { \
                if (err) *err = FR_SBUFF_PARSE_ERROR_TRAILING; \
                FR_SBUFF_ERROR_RETURN(&our_in); \
        } \
        *out = res; \
        return fr_sbuff_advance(in, end - buff); \
}
 
SBUFF_PARSE_FLOAT_DEF(float32, float, strtof, 100)
SBUFF_PARSE_FLOAT_DEF(float64, double, strtod, 100)
 
/** Move data from one sbuff to another
 *
 * @note Do not call this function directly use #fr_sbuff_move
 *
 * Both in and out will be advanced by len, with len set to the shortest
 * value between the user specified value, the number of bytes remaining
 * in the input buffer (after extension), and the number of bytes remaining
 * in the output buffer (after extension).
 *
 * @param[in] out       sbuff to copy data to.
 * @param[in] in        sbuff to copy data from.
 * @param[in] len       Maximum length of string to copy.
 * @return The amount of data copied.
 */
size_t _fr_sbuff_move_sbuff_to_sbuff(fr_sbuff_t *out, fr_sbuff_t *in, size_t len)
{
        size_t o_remaining = fr_sbuff_extend_lowat(NULL, out, len);
        size_t i_remaining = fr_sbuff_extend_lowat(NULL, in, len);
        size_t to_copy = len;
        if (to_copy > o_remaining) to_copy = o_remaining;
        if (to_copy > i_remaining) to_copy = i_remaining;
        safecpy(fr_sbuff_current(out), fr_sbuff_end(out), fr_sbuff_current(in), fr_sbuff_current(in) + to_copy);
        return fr_sbuff_advance(out, fr_sbuff_advance(in, to_copy));
}
 
/** Move data from a marker to an sbuff
 *
 * @note Do not call this function directly use #fr_sbuff_move
 *
 * @param[in] out       sbuff to copy data to.
 * @param[in] in        marker to copy data from.
 * @param[in] len       Maximum length of string to copy.
 * @return The amount of data copied.
 */
size_t _fr_sbuff_move_marker_to_sbuff(fr_sbuff_t *out, fr_sbuff_marker_t *in, size_t len)
{
        size_t o_remaining = fr_sbuff_extend_lowat(NULL, out, len);
        size_t i_remaining = fr_sbuff_extend_lowat(NULL, in, len);
        size_t to_copy = len;
        if (to_copy > o_remaining) to_copy = o_remaining;
        if (to_copy > i_remaining) to_copy = i_remaining;
        safecpy(fr_sbuff_current(out), fr_sbuff_end(out), fr_sbuff_current(in), fr_sbuff_current(in) + to_copy);
        return fr_sbuff_advance(out, fr_sbuff_advance(in, to_copy));
}
 
/** Move data from one marker to another
 *
 * @note Do not call this function directly use #fr_sbuff_move
 *
 * @param[in] out       marker to copy data to.
 * @param[in] in        marker to copy data from.
 * @param[in] len       Maximum length of string to copy.
 * @return The amount of data copied.
 */
size_t _fr_sbuff_move_marker_to_marker(fr_sbuff_marker_t *out, fr_sbuff_marker_t *in, size_t len)
{
        size_t o_remaining = fr_sbuff_extend_lowat(NULL, out, len);
        size_t i_remaining = fr_sbuff_extend_lowat(NULL, in, len);
        size_t to_copy = len;
        if (to_copy > o_remaining) to_copy = o_remaining;
        if (to_copy > i_remaining) to_copy = i_remaining;
        safecpy(fr_sbuff_current(out), fr_sbuff_end(out), fr_sbuff_current(in), fr_sbuff_current(in) + to_copy);
        return fr_sbuff_advance(out, fr_sbuff_advance(in, to_copy));
}
 
/** Move data from an sbuff to a marker
 *
 * @note Do not call this function directly use #fr_sbuff_move
 *
 * @param[in] out       marker to copy data to.
 * @param[in] in        sbuff to copy data from.
 * @param[in] len       Maximum length of string to copy.
 * @return The amount of data copied.
 */
size_t _fr_sbuff_move_sbuff_to_marker(fr_sbuff_marker_t *out, fr_sbuff_t *in, size_t len)
{
        size_t o_remaining = fr_sbuff_extend_lowat(NULL, out, len);
        size_t i_remaining = fr_sbuff_extend_lowat(NULL, in, len);
        size_t to_copy = len;
        if (to_copy > o_remaining) to_copy = o_remaining;
        if (to_copy > i_remaining) to_copy = i_remaining;
        safecpy(fr_sbuff_current(out), fr_sbuff_end(out), fr_sbuff_current(in), fr_sbuff_current(in) + to_copy);
        return fr_sbuff_advance(out, fr_sbuff_advance(in, to_copy));
}
 
/** Copy bytes into the sbuff up to the first \0
 *
 * @param[in] sbuff     to copy into.
 * @param[in] str       to copy into buffer.
 * @return
 *      - >= 0 the number of bytes copied into the sbuff.
 *      - -1 the copy would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_strcpy(fr_sbuff_t *sbuff, char const *str)
{
        size_t len;
 
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        len = strlen(str);
        FR_SBUFF_EXTEND_LOWAT_OR_RETURN(sbuff, len);
 
        safecpy(sbuff->p, sbuff->end, str, str + len);
        sbuff->p[len] = '\0';
 
        return fr_sbuff_advance(sbuff, len);
}
 
/** Copy bytes into the sbuff up to the first \0
 *
 * @param[in] sbuff     to copy into.
 * @param[in] str       to copy into buffer.
 * @param[in] len       number of bytes to copy.
 * @return
 *      - >= 0 the number of bytes copied into the sbuff.
 *      - -1 the copy would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_bstrncpy(fr_sbuff_t *sbuff, char const *str, size_t len)
{
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        FR_SBUFF_EXTEND_LOWAT_OR_RETURN(sbuff, len);
 
        safecpy(sbuff->p, sbuff->end, str, str + len);
        sbuff->p[len] = '\0';
 
        return fr_sbuff_advance(sbuff, len);
}
 
/** Copy bytes into the sbuff up to the first \0
 *
 * @param[in] sbuff     to copy into.
 * @param[in] str       talloced buffer to copy into sbuff.
 * @return
 *      - >= 0 the number of bytes copied into the sbuff.
 *      - -1 the copy would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_bstrcpy_buffer(fr_sbuff_t *sbuff, char const *str)
{
        size_t len;
 
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        len = talloc_strlen(str);
 
        FR_SBUFF_EXTEND_LOWAT_OR_RETURN(sbuff, len);
 
        safecpy(sbuff->p, sbuff->end, str, str + len);
        sbuff->p[len] = '\0';
 
        return fr_sbuff_advance(sbuff, len);
}
 
/** Free the scratch buffer used for printf
 *
 */
static int _sbuff_scratch_free(void *arg)
{
        sbuff_scratch_freed = true;
        return talloc_free(arg);
}
 
/** Initialise a thread local scratch context
 *
 * The scratch pool is an optimisation.  When it is unavailable, because
 * thread local allocation has been disabled at shutdown or the pool could
 * not be allocated, out is set to NULL and callers allocate at top level.
 *
 * @param[out] out      the scratch context, or NULL.
 */
static inline CC_HINT(always_inline) void sbuff_scratch_init(TALLOC_CTX **out)
{
        TALLOC_CTX      *scratch;
 
        /*
         *      Once main has signalled shutdown the TLS slot may be a
         *      dangling pointer on threads we don't own; skip the scratch
         *      cache and let callers allocate at top level instead.  The
         *      TLS-local `sbuff_scratch_freed` is left in place for the
         *      per-thread teardown path on FR-managed threads.
         */
        if (sbuff_scratch_freed || fr_atexit_thread_local_alloc_disabled()) {
                *out = NULL;
                return;
        }
 
        scratch = sbuff_scratch;
        if (!scratch) {
                scratch = talloc_pool(NULL, 4096);
                if (unlikely(!scratch)) {
                        *out = NULL;
                        return;
                }
                fr_atexit_thread_local(sbuff_scratch, _sbuff_scratch_free, scratch);
        }
 
        *out = scratch;
}
 
/** Print using a fmt string to an sbuff
 *
 * @param[in] sbuff     to print into.
 * @param[in] fmt       string.
 * @param[in] ap        arguments for format string.
 * @return
 *      - >= 0 the number of bytes printed into the sbuff.  0 if the
 *        formatted string could not be allocated.  Nothing was printed
 *        and the error stack says why.
 *      - -1 the printed output would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_vsprintf(fr_sbuff_t *sbuff, char const *fmt, va_list ap)
{
        TALLOC_CTX      *scratch;
        va_list         ap_p;
        char            *tmp;
        ssize_t         slen;
 
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        sbuff_scratch_init(&scratch);
 
        va_copy(ap_p, ap);
        tmp = fr_vasprintf(scratch, fmt, ap_p);
        va_end(ap_p);
        if (unlikely(!tmp)) {
                fr_strerror_const_push("Failed formatting string, nothing printed");    /* fr_vasprintf sets the cause */
                return 0;
        }
 
        slen = fr_sbuff_in_bstrcpy_buffer(sbuff, tmp);
        talloc_free(tmp);       /* Free the temporary buffer */
 
        return slen;
}
 
/** Print using a fmt string to an sbuff
 *
 * @param[in] sbuff     to print into.
 * @param[in] fmt       string.
 * @param[in] ...       arguments for format string.
 * @return
 *      - >= 0 the number of bytes printed into the sbuff.
 *      - -1 the printed output would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_sprintf(fr_sbuff_t *sbuff, char const *fmt, ...)
{
        va_list         ap;
        ssize_t         slen;
 
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        va_start(ap, fmt);
        slen = fr_sbuff_in_vsprintf(sbuff, fmt, ap);
        va_end(ap);
 
        return slen;
}
 
/** Print an escaped string to an sbuff
 *
 * @param[in] sbuff     to print into.
 * @param[in] in        to escape.
 * @param[in] inlen     of string to escape.
 * @param[in] e_rules   Escaping rules.  Used to escape special characters
 *                      as data is written to the sbuff.  May be NULL.
 * @return
 *      - >= 0 the number of bytes printed into the sbuff.
 *      - -1 the printed output would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_escape(fr_sbuff_t *sbuff, char const *in, size_t inlen, fr_sbuff_escape_rules_t const *e_rules)
{
        char const      *end = in + inlen;
        char const      *p = in;
        fr_sbuff_t      our_sbuff;
 
        /* Significantly quicker if there are no rules */
        if (!e_rules || (e_rules->chr == '\0')) return fr_sbuff_in_bstrncpy(sbuff, in, inlen);
 
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        our_sbuff = FR_SBUFF(sbuff);
        while (p < end) {
                size_t  clen;
                uint8_t c = (uint8_t)*p;
                char    sub;
 
                /*
                 *      We don't support escaping UTF8 sequences
                 *      as they're not used anywhere in our
                 *      grammar.
                 */
                if (e_rules->do_utf8 && ((clen = fr_utf8_char((uint8_t const *)p, end - p)) > 1)) {
                        FR_SBUFF_IN_BSTRNCPY_RETURN(&our_sbuff, p, clen);
                        p += clen;
                        continue;
                }
 
                /*
                 *      Check if there's a special substitution
                 *      like 0x0a -> \n.
                 */
                sub = e_rules->subs[c];
                if (sub != '\0') {
                        FR_SBUFF_IN_CHAR_RETURN(&our_sbuff, e_rules->chr, sub);
                        p++;
                        continue;
                }
 
                /*
                 *      Check if the character is in the range
                 *      we escape.
                 */
                if (e_rules->esc[c]) {
                        /*
                         *      For legacy reasons we prefer
                         *      octal escape sequences.
                         */
                        if (e_rules->do_oct) {
                                FR_SBUFF_IN_SPRINTF_RETURN(&our_sbuff, "%c%03o", e_rules->chr, (uint8_t)*p++);
                                continue;
                        } else if (e_rules->do_hex) {
                                FR_SBUFF_IN_SPRINTF_RETURN(&our_sbuff, "%cx%02x", e_rules->chr, (uint8_t)*p++);
                                continue;
                        }
                }
 
                FR_SBUFF_IN_CHAR_RETURN(&our_sbuff, *p++);
        }
 
        FR_SBUFF_SET_RETURN(sbuff, &our_sbuff);
}
 
/** Walk an input string and report whether fr_sbuff_in_escape() would
 * escape any characters in it.
 *
 *  Mirrors the per-byte decisions of #fr_sbuff_in_escape: a byte
 *  inside a multi-byte UTF-8 sequence (when do_utf8 is set) is passed
 *  through, a byte with a substitution mapping is escaped, and a byte
 *  in the esc[] table is escaped.  If any byte would be escaped, the
 *  function returns false at that byte.  A NULL or chr=='\0' ruleset
 *  is treated as "no escaping": the function always returns true.
 *
 * @param[in] in        to inspect.
 * @param[in] inlen     bytes of `in` to inspect.
 * @param[in] e_rules   escaping rules.  May be NULL.
 * @return
 *      - false at least one byte would be escaped.
 *      - true  no byte would be escaped (the string is already safe).
 */
bool fr_sbuff_in_needs_escaping(char const *in, size_t inlen, fr_sbuff_escape_rules_t const *e_rules)
{
        char const      *end = in + inlen;
        char const      *p = in;
 
        if (!e_rules || !e_rules->chr) return false;
 
        while (p < end) {
                size_t  clen;
                uint8_t c = (uint8_t) *p;
 
                if (e_rules->do_utf8 && ((clen = fr_utf8_char((uint8_t const *) p, end - p)) > 1)) {
                        p += clen;
                        continue;
                }
 
                if (e_rules->subs[c] != '\0') return false;
 
                if (e_rules->esc[c]) return false;
 
                p++;
        }
 
        return true;
}
 
/** Print an escaped string to an sbuff taking a talloced buffer as input
 *
 * @param[in] sbuff     to print into.
 * @param[in] in        to escape.
 * @param[in] e_rules   Escaping rules.  Used to escape special characters
 *                      as data is written to the sbuff.  May be NULL.
 * @return
 *      - >= 0 the number of bytes printed into the sbuff.
 *      - -1 the printed output would not fit in the output buffer.
 */
ssize_t fr_sbuff_in_escape_buffer(fr_sbuff_t *sbuff, char const *in, fr_sbuff_escape_rules_t const *e_rules)
{
        if (unlikely(!in)) return 0;
 
        CHECK_SBUFF_WRITEABLE(sbuff);
 
        return fr_sbuff_in_escape(sbuff, in, talloc_strlen(in), e_rules);
}
 
/** Concat an array of strings (NULL terminated), with a string separator
 *
 * @param[out] out      Where to write the resulting string.
 * @param[in] array     of strings to concat.
 * @param[in] sep       to insert between elements.  May be NULL.
 * @return
 *      - >= 0 on success - length of the string created.
 *      - -1 on failure.
 */
fr_slen_t fr_sbuff_in_array(fr_sbuff_t *out, char const * const *array, char const *sep)
{
        fr_sbuff_t              our_out = FR_SBUFF(out);
        char const * const *    p;
        fr_sbuff_escape_rules_t e_rules = {
                                        .name = __FUNCTION__,
                                        .chr = '\\'
                                };
 
        if (sep) e_rules.subs[(uint8_t)*sep] = *sep;
 
        CHECK_SBUFF_WRITEABLE(out);
 
        for (p = array; *p; p++) {
                if (*p) FR_SBUFF_RETURN(fr_sbuff_in_escape, &our_out, *p, strlen(*p), &e_rules);
 
                if (sep && p[1]) {
                        FR_SBUFF_RETURN(fr_sbuff_in_strcpy, &our_out, sep);
                }
        }
 
        FR_SBUFF_SET_RETURN(out, &our_out);
}
 
/** Return true and advance past the end of the needle if needle occurs next in the sbuff
 *
 * @param[in] sbuff             to search in.
 * @param[in] needle            to search for.
 * @param[in] needle_len        of needle. If SIZE_MAX strlen is used
 *                              to determine length of the needle.
 * @return how many bytes we advanced
 */
size_t fr_sbuff_adv_past_str(fr_sbuff_t *sbuff, char const *needle, size_t needle_len)
{
        char const *found;
 
        CHECK_SBUFF_INIT(sbuff);
 
        if (needle_len == SIZE_MAX) needle_len = strlen(needle);
 
        /*
         *      If there's insufficient bytes in the
         *      buffer currently, try to extend it,
         *      returning if we can't.
         */
        if (fr_sbuff_extend_lowat(NULL, sbuff, needle_len) < needle_len) return 0;
 
        found = memmem(sbuff->p, needle_len, needle, needle_len);       /* sbuff needle_len and needle needle_len ensures match must be next */
        if (!found) return 0;
 
        return fr_sbuff_advance(sbuff, needle_len);
}
 
/** Return true and advance past the end of the needle if needle occurs next in the sbuff
 *
 * This function is similar to fr_sbuff_adv_past_str but is case insensitive.
 *
 * @param[in] sbuff             to search in.
 * @param[in] needle            to search for.
 * @param[in] needle_len        of needle. If SIZE_MAX strlen is used
 *                              to determine length of the needle.
 * @return how many bytes we advanced
 */
size_t fr_sbuff_adv_past_strcase(fr_sbuff_t *sbuff, char const *needle, size_t needle_len)
{
        char const *p, *n_p;
        char const *end;
 
        CHECK_SBUFF_INIT(sbuff);
 
        if (needle_len == SIZE_MAX) needle_len = strlen(needle);
 
        /*
         *      If there's insufficient bytes in the
         *      buffer currently, try to extend it,
         *      returning if we can't.
         */
        if (fr_sbuff_extend_lowat(NULL, sbuff, needle_len) < needle_len) return 0;
 
        p = sbuff->p;
        end = p + needle_len;
 
        for (p = sbuff->p, n_p = needle; p < end; p++, n_p++) {
                if (tolower((uint8_t) *p) != tolower((uint8_t) *n_p)) return 0;
        }
 
        return fr_sbuff_advance(sbuff, needle_len);
}
 
/** Wind position past characters in the allowed set
 *
 * @param[in] sbuff             sbuff to search in.
 * @param[in] len               Maximum amount to advance by. Unconstrained if SIZE_MAX.
 * @param[in] allowed           character set.
 * @param[in] tt                If not NULL, stop if we find a terminal sequence.
 * @return how many bytes we advanced.
 */
size_t fr_sbuff_adv_past_allowed(fr_sbuff_t *sbuff, size_t len, bool
                                 const allowed[static SBUFF_CHAR_CLASS], fr_sbuff_term_t const *tt)
{
        size_t          total = 0;
        char const      *p;
        uint8_t         idx[SBUFF_CHAR_CLASS];  /* Fast path index */
        size_t          needle_len = 0;
 
        CHECK_SBUFF_INIT(sbuff);
 
        if (tt) fr_sbuff_terminal_idx_init(&needle_len, idx, tt);
 
        while (total < len) {
                char *end;
 
                if (!fr_sbuff_extend(sbuff)) break;
 
                end = CONSTRAINED_END(sbuff, len, total);
                p = sbuff->p;
                while ((p < end) && allowed[(uint8_t)*p]) {
                        if (needle_len == 0) {
                                p++;
                                continue;
                        }
 
                       /*
                        *       If this character is allowed, BUT is also listed as a one-character terminal,
                        *       then we still allow it.  This decision implements "greedy" parsing.
                        */
                       if (fr_sbuff_terminal_search(sbuff, p, idx, tt, 1)) {
                               p++;
                               continue;
                       }
 
                       /*
                        *       Otherwise if the next *set* of characters) is not in the terminals, then
                        *       allow the current character.
                        */
                       if (!fr_sbuff_terminal_search(sbuff, p, idx, tt, needle_len)) {
                               p++;
                               continue;
                       }
 
                       /*
                        *       The character is allowed, and is NOT listed as a terminal character by itself.
                        *       However, it is part of a multi-character terminal sequence.  We therefore
                        *       stop.
                        *
                        *       This decision allows us to parse things like "Framed-User", where we might
                        *       normally stop at the "-".  However, we will still stop at "Framed-=User", as
                        *       "-=" may be a terminal sequence.
                        *
                        *       There is no perfect solution here, other than to fix the input grammar so that
                        *       it has no ambiguity.  Since we can't do that, we choose to err on the side of
                        *       allowing the existing grammar, where it makes sense
                        */
                       break;
                }
 
                total += fr_sbuff_set(sbuff, p);
                if (p != end) break;            /* stopped early, break */
        }
 
        return total;
}
 
/** Wind position until we hit a character in the terminal set
 *
 * @param[in] sbuff             sbuff to search in.
 * @param[in] len               Maximum amount to advance by. Unconstrained if SIZE_MAX.
 * @param[in] tt                Token terminals in the encompassing grammar.
 * @param[in] escape_chr        If not '\0', ignore characters in the tt set when
 *                              prefixed with this escape character.
 * @return how many bytes we advanced.
 */
size_t fr_sbuff_adv_until(fr_sbuff_t *sbuff, size_t len, fr_sbuff_term_t const *tt, char escape_chr)
{
        size_t          total = 0;
        char const      *p;
        bool            do_escape = false;              /* Track state across extensions */
 
        uint8_t         idx[SBUFF_CHAR_CLASS];          /* Fast path index */
        size_t          needle_len = 1;
 
        CHECK_SBUFF_INIT(sbuff);
 
        /*
         *      Initialise the fastpath index and
         *      figure out the longest needle.
         */
        fr_sbuff_terminal_idx_init(&needle_len, idx, tt);
 
        while (total < len) {
                char *end;
 
                if (fr_sbuff_extend_lowat(NULL, sbuff, needle_len) == 0) break;
 
                end = CONSTRAINED_END(sbuff, len, total);
                p = sbuff->p;
 
                if (escape_chr == '\0') {
                        while ((p < end) && !fr_sbuff_terminal_search(sbuff, p, idx, tt, needle_len)) p++;
                } else {
                        while (p < end) {
                                if (do_escape) {
                                        do_escape = false;
                                } else if (*p == escape_chr) {
                                        do_escape = true;
                                } else if (fr_sbuff_terminal_search(sbuff, p, idx, tt, needle_len)) {
                                        break;
                                }
                                p++;
                        }
                }
 
                total += fr_sbuff_set(sbuff, p);
                if (p != end) break;    /* stopped early, break */
        }
 
        return total;
}
 
/** Wind position to first instance of specified multibyte utf8 char
 *
 * Only use this function if the search char could be multibyte,
 * as there's a large performance penalty.
 *
 * @param[in,out] sbuff         to search in.
 * @param[in] len               the maximum number of characters to search in sbuff.
 * @param[in] chr               to search for.
 * @return
 *      - NULL, no instances found.
 *      - The position of the first character.
 */
char *fr_sbuff_adv_to_chr_utf8(fr_sbuff_t *sbuff, size_t len, char const *chr)
{
        fr_sbuff_t      our_sbuff = FR_SBUFF(sbuff);
        size_t          total = 0;
        size_t          clen = strlen(chr);
 
        CHECK_SBUFF_INIT(sbuff);
 
        /*
         *      Needle bigger than haystack
         */
        if (len < clen) return NULL;
 
        while (total <= (len - clen)) {
                char const      *found;
                char            *end;
 
                /*
                 *      Ensure we have enough chars to match
                 *      the needle.
                 */
                if (fr_sbuff_extend_lowat(NULL, &our_sbuff, clen) < clen) break;
 
                end = CONSTRAINED_END(&our_sbuff, len, total);
 
                found = fr_utf8_strchr(NULL, our_sbuff.p, end - our_sbuff.p, chr);
                if (found) {
                        (void)fr_sbuff_set(sbuff, found);
                        return sbuff->p;
                }
                total += fr_sbuff_set(&our_sbuff, (end - clen) + 1);
        }
 
        return NULL;
}
 
/** Wind position to first instance of specified char
 *
 * @param[in,out] sbuff         to search in.
 * @param[in] len               Maximum amount to advance by. Unconstrained if SIZE_MAX.
 * @param[in] c                 to search for.
 * @return
 *      - NULL, no instances found.
 *      - The position of the first character.
 */
char *fr_sbuff_adv_to_chr(fr_sbuff_t *sbuff, size_t len, char c)
{
        fr_sbuff_t      our_sbuff = FR_SBUFF(sbuff);
        size_t          total = 0;
 
        CHECK_SBUFF_INIT(sbuff);
 
        while (total < len) {
                char const      *found;
                char            *end;
 
                if (!fr_sbuff_extend(&our_sbuff)) break;
 
                end = CONSTRAINED_END(&our_sbuff, len, total);
                found = memchr(our_sbuff.p, c, end - our_sbuff.p);
                if (found) {
                        (void)fr_sbuff_set(sbuff, found);
                        return sbuff->p;
                }
 
                total += fr_sbuff_set(&our_sbuff, end);
        }
 
        return NULL;
}
 
/** Wind position to the first instance of the specified needle
 *
 * @param[in,out] sbuff         sbuff to search in.
 * @param[in] len               Maximum amount to advance by. Unconstrained if SIZE_MAX.
 * @param[in] needle            to search for.
 * @param[in] needle_len        Length of the needle. SIZE_MAX to used strlen.
 * @return
 *      - NULL, no instances found.
 *      - The position of the first character.
 */
char *fr_sbuff_adv_to_str(fr_sbuff_t *sbuff, size_t len, char const *needle, size_t needle_len)
{
        fr_sbuff_t      our_sbuff = FR_SBUFF(sbuff);
        size_t          total = 0;
 
        CHECK_SBUFF_INIT(sbuff);
 
        if (needle_len == SIZE_MAX) needle_len = strlen(needle);
        if (!needle_len) return NULL;
 
        /*
         *      Needle bigger than haystack
         */
        if (len < needle_len) return NULL;
 
        while (total <= (len - needle_len)) {
                char const      *found;
                char            *end;
 
                /*
                 *      If the needle is longer than
                 *      the remaining buffer, return.
                 */
                if (fr_sbuff_extend_lowat(NULL, &our_sbuff, needle_len) < needle_len) break;
 
                end = CONSTRAINED_END(&our_sbuff, len, total);
                found = memmem(our_sbuff.p, end - our_sbuff.p, needle, needle_len);
                if (found) {
                        (void)fr_sbuff_set(sbuff, found);
                        return sbuff->p;
                }
 
                /*
                 *      Partial needle may be in
                 *      the end of the buffer so
                 *      don't advance too far.
                 */
                total += fr_sbuff_set(&our_sbuff, (end - needle_len) + 1);
        }
 
        return NULL;
}
 
/** Wind position to the first instance of the specified needle
 *
 * @param[in,out] sbuff         sbuff to search in.
 * @param[in] len               Maximum amount to advance by. Unconstrained if SIZE_MAX.
 * @param[in] needle            to search for.
 * @param[in] needle_len        Length of the needle. SIZE_MAX to used strlen.
 * @return
 *      - NULL, no instances found.
 *      - The position of the first character.
 */
char *fr_sbuff_adv_to_strcase(fr_sbuff_t *sbuff, size_t len, char const *needle, size_t needle_len)
{
        fr_sbuff_t      our_sbuff = FR_SBUFF(sbuff);
        size_t          total = 0;
 
        CHECK_SBUFF_INIT(sbuff);
 
        if (needle_len == SIZE_MAX) needle_len = strlen(needle);
        if (!needle_len) return NULL;
 
        /*
         *      Needle bigger than haystack
         */
        if (len < needle_len) return NULL;
 
        while (total <= (len - needle_len)) {
                char *p, *end;
                char const *n_p;
 
                if (fr_sbuff_extend_lowat(NULL, &our_sbuff, needle_len) < needle_len) break;
 
                for (p = our_sbuff.p, n_p = needle, end = our_sbuff.p + needle_len;
                     (p < end) && (tolower((uint8_t) *p) == tolower((uint8_t) *n_p));
                     p++, n_p++);
                if (p == end) {
                        (void)fr_sbuff_set(sbuff, our_sbuff.p);
                        return sbuff->p;
                }
 
                total += fr_sbuff_advance(&our_sbuff, 1);
        }
 
        return NULL;
}
 
/** Return true if the current char matches, and if it does, advance
 *
 * @param[in] sbuff     to search for char in.
 * @param[in] c         char to search for.
 * @return
 *      - true and advance if the next character matches.
 *      - false and don't advance if the next character doesn't match.
 */
bool fr_sbuff_next_if_char(fr_sbuff_t *sbuff, char c)
{
        CHECK_SBUFF_INIT(sbuff);
 
        if (!fr_sbuff_extend(sbuff)) return false;
 
        if (*sbuff->p != c) return false;
 
        fr_sbuff_advance(sbuff, 1);
 
        return true;
}
 
/** Return true and advance if the next char does not match
 *
 * @param[in] sbuff     to search for char in.
 * @param[in] c         char to search for.
 * @return
 *      - true and advance unless the character matches.
 *      - false and don't advance if the next character matches.
 */
bool fr_sbuff_next_unless_char(fr_sbuff_t *sbuff, char c)
{
        CHECK_SBUFF_INIT(sbuff);
 
        if (!fr_sbuff_extend(sbuff)) return false;
 
        if (*sbuff->p == c) return false;
 
        fr_sbuff_advance(sbuff, 1);
 
        return true;
}
 
/** Trim trailing characters from a string we're composing
 *
 * @param[in] sbuff             to trim trailing characters from.
 * @param[in] to_trim           Charset to trim.
 * @return how many chars we removed.
 */
size_t fr_sbuff_trim(fr_sbuff_t *sbuff, bool const to_trim[static SBUFF_CHAR_CLASS])
{
        char    *p = sbuff->p - 1;
        ssize_t slen;
 
        while ((p >= sbuff->start) && to_trim[(uint8_t)*p]) p--;
 
        slen = fr_sbuff_set(sbuff, p + 1);
        if (slen != 0) fr_sbuff_terminate(sbuff);
 
        return slen;
}
 
/** Efficient terminal string search
 *
 * Caller should ensure that a buffer extension of needle_len bytes has been requested
 * before calling this function.
 *
 * @param[in] in        Sbuff to search in.
 * @param[in] tt        Token terminals in the encompassing grammar.
 * @return
 *      - true if found.
 *      - false if not.
 */
bool fr_sbuff_is_terminal(fr_sbuff_t *in, fr_sbuff_term_t const *tt)
{
        uint8_t         idx[SBUFF_CHAR_CLASS];  /* Fast path index */
        size_t          needle_len = 1;
 
        /*
         *      No terminal, check for EOF.
         */
        if (!tt) {
                fr_sbuff_extend_status_t status = 0;
 
                if ((fr_sbuff_extend_lowat(&status, in, 1) == 0) &&
                    (status & FR_SBUFF_FLAG_EXTEND_ERROR) == 0) {
                        return true;
                }
 
                return false;
        }
 
        /*
         *      Initialise the fastpath index and
         *      figure out the longest needle.
         */
        fr_sbuff_terminal_idx_init(&needle_len, idx, tt);
 
        fr_sbuff_extend_lowat(NULL, in, needle_len);
 
        return fr_sbuff_terminal_search(in, in->p, idx, tt, needle_len);
}
 
/** Print a char in a friendly format
 *
 */
static char const *sbuff_print_char(char c)
{
        static bool const unprintables[SBUFF_CHAR_CLASS] = {
                SBUFF_CHAR_UNPRINTABLES_LOW,
                SBUFF_CHAR_UNPRINTABLES_EXTENDED
        };
 
        static _Thread_local char str[10][5];
        static _Thread_local size_t i = 0;
 
        switch (c) {
        case '\a':
                return "\a";
 
        case '\b':
                return "\b";
 
        case '\n':
                return "\n";
 
        case '\r':
                return "\r";
 
        case '\t':
                return "\t";
 
        case '\f':
                return "\f";
 
        case '\v':
                return "\v";
 
        default:
                if (i >= NUM_ELEMENTS(str)) i = 0;
 
                if (unprintables[(uint8_t)c]) {
                        snprintf(str[i], sizeof(str[i]), "\\x%02x", (uint8_t) c);
                        return str[i++];
                }
 
                str[i][0] = c;
                str[i][1] = '\0';
                return str[i++];
        }
}
 
void fr_sbuff_unescape_debug(FILE *fp, fr_sbuff_unescape_rules_t const *escapes)
{
        int i;
 
        fprintf(fp, "Escape rules %s (%p)\n", escapes->name, escapes);
        fprintf(fp, "chr     : %c\n", escapes->chr ? escapes->chr : ' ');
        fprintf(fp, "do_hex  : %s\n", escapes->do_hex ? "yes" : "no");
        fprintf(fp, "do_oct  : %s\n", escapes->do_oct ? "yes" : "no");
 
        fprintf(fp, "substitutions:\n");
        for (i = 0; i < SBUFF_CHAR_CLASS; i++) {
                if (escapes->subs[i]) FR_FAULT_LOG("\t%s -> %s\n",
                                                   sbuff_print_char((char)i),
                                                   sbuff_print_char((char)escapes->subs[i]));
        }
        fprintf(fp, "skips:\n");
        for (i = 0; i < SBUFF_CHAR_CLASS; i++) {
                if (escapes->skip[i]) fprintf(fp, "\t%s\n", sbuff_print_char((char)i));
        }
}
 
void fr_sbuff_terminal_debug(FILE *fp, fr_sbuff_term_t const *tt)
{
        size_t i;
 
        fprintf(fp, "Terminal count %zu\n", tt->len);
 
        for (i = 0; i < tt->len; i++) fprintf(fp, "\t\"%s\" (%zu)\n", tt->elem[i].str, tt->elem[i].len);
}
 
void fr_sbuff_parse_rules_debug(FILE *fp, fr_sbuff_parse_rules_t const *p_rules)
{
        fprintf(fp, "Parse rules %p\n", p_rules);
 
        FR_FAULT_LOG("Escapes - ");
        if (p_rules->escapes) {
                fr_sbuff_unescape_debug(fp, p_rules->escapes);
        } else {
                fprintf(fp, "<none>\n");
        }
 
        FR_FAULT_LOG("Terminals - ");
        if (p_rules->terminals) {
                fr_sbuff_terminal_debug(fp, p_rules->terminals);
        } else {
                fprintf(fp, "<none>\n");
        }
}
 
/** Concat an array of strings (not NULL terminated), with a string separator
 *
 * @param[out] out      Where to write the resulting string.
 * @param[in] array     of strings to concat.
 * @param[in] sep       to insert between elements.  May be NULL.
 * @return
 *      - >= 0 on success - length of the string created.
 *      - -1 on failure.
 */
fr_slen_t fr_sbuff_array_concat(fr_sbuff_t *out, char const * const *array, char const *sep)
{
        fr_sbuff_t              our_out = FR_SBUFF(out);
        size_t                  len = talloc_array_length(array);
        char const * const *    p;
        char const * const *    end;
        fr_sbuff_escape_rules_t e_rules = {
                                        .name = __FUNCTION__,
                                        .chr = '\\'
                                };
 
        if (sep) e_rules.subs[(uint8_t)*sep] = *sep;
 
        for (p = array, end = array + len;
             (p < end);
             p++) {
                if (*p) FR_SBUFF_RETURN(fr_sbuff_in_escape, &our_out, *p, strlen(*p), &e_rules);
 
                if (sep && ((p + 1) < end)) {
                        FR_SBUFF_RETURN(fr_sbuff_in_strcpy, &our_out, sep);
                }
        }
 
        FR_SBUFF_SET_RETURN(out, &our_out);
}