bytea.c

Go to the documentation of this file.

/*-------------------------------------------------------------------------
 *
 * bytea.c
 *    Functions for the bytea type.
 *
 * Portions Copyright (c) 2025-2026, PostgreSQL Global Development Group
 *
 *
 * IDENTIFICATION
 *    src/backend/utils/adt/bytea.c
 *
 *-------------------------------------------------------------------------
 */
 
#include "postgres.h"
 
#include "access/detoast.h"
#include "common/hashfn.h"
#include "common/int.h"
#include "fmgr.h"
#include "lib/hyperloglog.h"
#include "libpq/pqformat.h"
#include "port/pg_bitutils.h"
#include "port/pg_bswap.h"
#include "utils/builtins.h"
#include "utils/bytea.h"
#include "utils/fmgrprotos.h"
#include "utils/guc.h"
#include "utils/memutils.h"
#include "utils/sortsupport.h"
#include "varatt.h"
 
/* GUC variable */
int         bytea_output = BYTEA_OUTPUT_HEX;
 
static bytea *bytea_catenate(bytea *t1, bytea *t2);
static bytea *bytea_substring(Datum str, int S, int L,
                              bool length_not_specified);
static bytea *bytea_overlay(bytea *t1, bytea *t2, int sp, int sl);
 
typedef struct
{
    bool        abbreviate;     /* Should we abbreviate keys? */
    hyperLogLogState abbr_card; /* Abbreviated key cardinality state */
    hyperLogLogState full_card; /* Full key cardinality state */
    double      prop_card;      /* Required cardinality proportion */
} ByteaSortSupport;
 
/* Static function declarations for sort support */
static int  byteafastcmp(Datum x, Datum y, SortSupport ssup);
static Datum bytea_abbrev_convert(Datum original, SortSupport ssup);
static bool bytea_abbrev_abort(int memtupcount, SortSupport ssup);
 
/*
 * bytea_catenate
 *  Guts of byteacat(), broken out so it can be used by other functions
 *
 * Arguments can be in short-header form, but not compressed or out-of-line
 */
static bytea *
bytea_catenate(bytea *t1, bytea *t2)
{
    bytea      *result;
    int         len1,
                len2,
                len;
    char       *ptr;
 
    len1 = VARSIZE_ANY_EXHDR(t1);
    len2 = VARSIZE_ANY_EXHDR(t2);
 
    /* paranoia ... probably should throw error instead? */
    if (len1 < 0)
        len1 = 0;
    if (len2 < 0)
        len2 = 0;
 
    len = len1 + len2 + VARHDRSZ;
    result = (bytea *) palloc(len);
 
    /* Set size of result string... */
    SET_VARSIZE(result, len);
 
    /* Fill data field of result string... */
    ptr = VARDATA(result);
    if (len1 > 0)
        memcpy(ptr, VARDATA_ANY(t1), len1);
    if (len2 > 0)
        memcpy(ptr + len1, VARDATA_ANY(t2), len2);
 
    return result;
}
 
#define PG_STR_GET_BYTEA(str_) \
    DatumGetByteaPP(DirectFunctionCall1(byteain, CStringGetDatum(str_)))
 
static bytea *
bytea_substring(Datum str,
                int S,
                int L,
                bool length_not_specified)
{
    int32       S1;             /* adjusted start position */
    int32       L1;             /* adjusted substring length */
    int32       E;              /* end position */
 
    /*
     * The logic here should generally match text_substring().
     */
    S1 = Max(S, 1);
 
    if (length_not_specified)
    {
        /*
         * Not passed a length - DatumGetByteaPSlice() grabs everything to the
         * end of the string if we pass it a negative value for length.
         */
        L1 = -1;
    }
    else if (L < 0)
    {
        /* SQL99 says to throw an error for E < S, i.e., negative length */
        ereport(ERROR,
                (errcode(ERRCODE_SUBSTRING_ERROR),
                 errmsg("negative substring length not allowed")));
        L1 = -1;                /* silence stupider compilers */
    }
    else if (pg_add_s32_overflow(S, L, &E))
    {
        /*
         * L could be large enough for S + L to overflow, in which case the
         * substring must run to end of string.
         */
        L1 = -1;
    }
    else
    {
        /*
         * A zero or negative value for the end position can happen if the
         * start was negative or one. SQL99 says to return a zero-length
         * string.
         */
        if (E < 1)
            return PG_STR_GET_BYTEA("");
 
        L1 = E - S1;
    }
 
    /*
     * If the start position is past the end of the string, SQL99 says to
     * return a zero-length string -- DatumGetByteaPSlice() will do that for
     * us.  We need only convert S1 to zero-based starting position.
     */
    return DatumGetByteaPSlice(str, S1 - 1, L1);
}
 
static bytea *
bytea_overlay(bytea *t1, bytea *t2, int sp, int sl)
{
    bytea      *result;
    bytea      *s1;
    bytea      *s2;
    int         sp_pl_sl;
 
    /*
     * Check for possible integer-overflow cases.  For negative sp, throw a
     * "substring length" error because that's what should be expected
     * according to the spec's definition of OVERLAY().
     */
    if (sp <= 0)
        ereport(ERROR,
                (errcode(ERRCODE_SUBSTRING_ERROR),
                 errmsg("negative substring length not allowed")));
    if (pg_add_s32_overflow(sp, sl, &sp_pl_sl))
        ereport(ERROR,
                (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
                 errmsg("integer out of range")));
 
    s1 = bytea_substring(PointerGetDatum(t1), 1, sp - 1, false);
    s2 = bytea_substring(PointerGetDatum(t1), sp_pl_sl, -1, true);
    result = bytea_catenate(s1, t2);
    result = bytea_catenate(result, s2);
 
    return result;
}
 
/*****************************************************************************
 *   USER I/O ROUTINES                                                       *
 *****************************************************************************/
 
#define VAL(CH)         ((CH) - '0')
#define DIG(VAL)        ((VAL) + '0')
 
/*
 *      byteain         - converts from printable representation of byte array
 *
 *      Non-printable characters must be passed as '\nnn' (octal) and are
 *      converted to internal form.  '\' must be passed as '\\'.
 */
Datum
byteain(PG_FUNCTION_ARGS)
{
    char       *inputText = PG_GETARG_CSTRING(0);
    Node       *escontext = fcinfo->context;
    size_t      len = strlen(inputText);
    size_t      bc;
    char       *tp;
    char       *rp;
    bytea      *result;
 
    /* Recognize hex input */
    if (inputText[0] == '\\' && inputText[1] == 'x')
    {
        bc = (len - 2) / 2 + VARHDRSZ;  /* maximum possible length */
        result = palloc(bc);
        bc = hex_decode_safe(inputText + 2, len - 2, VARDATA(result),
                             escontext);
        SET_VARSIZE(result, bc + VARHDRSZ); /* actual length */
 
        PG_RETURN_BYTEA_P(result);
    }
 
    /* Else, it's the traditional escaped style */
    result = (bytea *) palloc(len + VARHDRSZ);  /* maximum possible length */
 
    tp = inputText;
    rp = VARDATA(result);
    while (*tp != '\0')
    {
        if (tp[0] != '\\')
            *rp++ = *tp++;
        else if ((tp[1] >= '0' && tp[1] <= '3') &&
                 (tp[2] >= '0' && tp[2] <= '7') &&
                 (tp[3] >= '0' && tp[3] <= '7'))
        {
            int         v;
 
            v = VAL(tp[1]);
            v <<= 3;
            v += VAL(tp[2]);
            v <<= 3;
            *rp++ = v + VAL(tp[3]);
 
            tp += 4;
        }
        else if (tp[1] == '\\')
        {
            *rp++ = '\\';
            tp += 2;
        }
        else
        {
            /*
             * one backslash, not followed by another or ### valid octal
             */
            ereturn(escontext, (Datum) 0,
                    (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
                     errmsg("invalid input syntax for type %s", "bytea")));
        }
    }
 
    bc = rp - VARDATA(result);  /* actual length */
    SET_VARSIZE(result, bc + VARHDRSZ);
 
    PG_RETURN_BYTEA_P(result);
}
 
/*
 *      byteaout        - converts to printable representation of byte array
 *
 *      In the traditional escaped format, non-printable characters are
 *      printed as '\nnn' (octal) and '\' as '\\'.
 */
Datum
byteaout(PG_FUNCTION_ARGS)
{
    bytea      *vlena = PG_GETARG_BYTEA_PP(0);
    char       *result;
    char       *rp;
 
    if (bytea_output == BYTEA_OUTPUT_HEX)
    {
        /* Print hex format */
        rp = result = palloc(VARSIZE_ANY_EXHDR(vlena) * 2 + 2 + 1);
        *rp++ = '\\';
        *rp++ = 'x';
        rp += hex_encode(VARDATA_ANY(vlena), VARSIZE_ANY_EXHDR(vlena), rp);
    }
    else if (bytea_output == BYTEA_OUTPUT_ESCAPE)
    {
        /* Print traditional escaped format */
        char       *vp;
        uint64      len;
        int         i;
 
        len = 1;                /* empty string has 1 char */
        vp = VARDATA_ANY(vlena);
        for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
        {
            if (*vp == '\\')
                len += 2;
            else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
                len += 4;
            else
                len++;
        }
 
        /*
         * In principle len can't overflow uint32 if the input fit in 1GB, but
         * for safety let's check rather than relying on palloc's internal
         * check.
         */
        if (len > MaxAllocSize)
            ereport(ERROR,
                    (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
                     errmsg_internal("result of bytea output conversion is too large")));
        rp = result = (char *) palloc(len);
 
        vp = VARDATA_ANY(vlena);
        for (i = VARSIZE_ANY_EXHDR(vlena); i != 0; i--, vp++)
        {
            if (*vp == '\\')
            {
                *rp++ = '\\';
                *rp++ = '\\';
            }
            else if ((unsigned char) *vp < 0x20 || (unsigned char) *vp > 0x7e)
            {
                int         val;    /* holds unprintable chars */
 
                val = *vp;
                rp[0] = '\\';
                rp[3] = DIG(val & 07);
                val >>= 3;
                rp[2] = DIG(val & 07);
                val >>= 3;
                rp[1] = DIG(val & 03);
                rp += 4;
            }
            else
                *rp++ = *vp;
        }
    }
    else
    {
        elog(ERROR, "unrecognized \"bytea_output\" setting: %d",
             bytea_output);
        rp = result = NULL;     /* keep compiler quiet */
    }
    *rp = '\0';
    PG_RETURN_CSTRING(result);
}
 
/*
 *      bytearecv           - converts external binary format to bytea
 */
Datum
bytearecv(PG_FUNCTION_ARGS)
{
    StringInfo  buf = (StringInfo) PG_GETARG_POINTER(0);
    bytea      *result;
    int         nbytes;
 
    nbytes = buf->len - buf->cursor;
    result = (bytea *) palloc(nbytes + VARHDRSZ);
    SET_VARSIZE(result, nbytes + VARHDRSZ);
    pq_copymsgbytes(buf, VARDATA(result), nbytes);
    PG_RETURN_BYTEA_P(result);
}
 
/*
 *      byteasend           - converts bytea to binary format
 *
 * This is a special case: just copy the input...
 */
Datum
byteasend(PG_FUNCTION_ARGS)
{
    bytea      *vlena = PG_GETARG_BYTEA_P_COPY(0);
 
    PG_RETURN_BYTEA_P(vlena);
}
 
Datum
bytea_string_agg_transfn(PG_FUNCTION_ARGS)
{
    StringInfo  state;
 
    state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
 
    /* Append the value unless null, preceding it with the delimiter. */
    if (!PG_ARGISNULL(1))
    {
        bytea      *value = PG_GETARG_BYTEA_PP(1);
        bool        isfirst = false;
 
        /*
         * You might think we can just throw away the first delimiter, however
         * we must keep it as we may be a parallel worker doing partial
         * aggregation building a state to send to the main process.  We need
         * to keep the delimiter of every aggregation so that the combine
         * function can properly join up the strings of two separately
         * partially aggregated results.  The first delimiter is only stripped
         * off in the final function.  To know how much to strip off the front
         * of the string, we store the length of the first delimiter in the
         * StringInfo's cursor field, which we don't otherwise need here.
         */
        if (state == NULL)
        {
            MemoryContext aggcontext;
            MemoryContext oldcontext;
 
            if (!AggCheckCallContext(fcinfo, &aggcontext))
            {
                /* cannot be called directly because of internal-type argument */
                elog(ERROR, "bytea_string_agg_transfn called in non-aggregate context");
            }
 
            /*
             * Create state in aggregate context.  It'll stay there across
             * subsequent calls.
             */
            oldcontext = MemoryContextSwitchTo(aggcontext);
            state = makeStringInfo();
            MemoryContextSwitchTo(oldcontext);
 
            isfirst = true;
        }
 
        if (!PG_ARGISNULL(2))
        {
            bytea      *delim = PG_GETARG_BYTEA_PP(2);
 
            appendBinaryStringInfo(state, VARDATA_ANY(delim),
                                   VARSIZE_ANY_EXHDR(delim));
            if (isfirst)
                state->cursor = VARSIZE_ANY_EXHDR(delim);
        }
 
        appendBinaryStringInfo(state, VARDATA_ANY(value),
                               VARSIZE_ANY_EXHDR(value));
    }
 
    /*
     * The transition type for string_agg() is declared to be "internal",
     * which is a pass-by-value type the same size as a pointer.
     */
    if (state)
        PG_RETURN_POINTER(state);
    PG_RETURN_NULL();
}
 
Datum
bytea_string_agg_finalfn(PG_FUNCTION_ARGS)
{
    StringInfo  state;
 
    /* cannot be called directly because of internal-type argument */
    Assert(AggCheckCallContext(fcinfo, NULL));
 
    state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
 
    if (state != NULL)
    {
        /* As per comment in transfn, strip data before the cursor position */
        bytea      *result;
        int         strippedlen = state->len - state->cursor;
 
        result = (bytea *) palloc(strippedlen + VARHDRSZ);
        SET_VARSIZE(result, strippedlen + VARHDRSZ);
        memcpy(VARDATA(result), &state->data[state->cursor], strippedlen);
        PG_RETURN_BYTEA_P(result);
    }
    else
        PG_RETURN_NULL();
}
 
/*-------------------------------------------------------------
 * byteaoctetlen
 *
 * get the number of bytes contained in an instance of type 'bytea'
 *-------------------------------------------------------------
 */
Datum
byteaoctetlen(PG_FUNCTION_ARGS)
{
    Datum       str = PG_GETARG_DATUM(0);
 
    /* We need not detoast the input at all */
    PG_RETURN_INT32(toast_raw_datum_size(str) - VARHDRSZ);
}
 
/*
 * byteacat -
 *    takes two bytea* and returns a bytea* that is the concatenation of
 *    the two.
 *
 * Cloned from textcat and modified as required.
 */
Datum
byteacat(PG_FUNCTION_ARGS)
{
    bytea      *t1 = PG_GETARG_BYTEA_PP(0);
    bytea      *t2 = PG_GETARG_BYTEA_PP(1);
 
    PG_RETURN_BYTEA_P(bytea_catenate(t1, t2));
}
 
/*
 * byteaoverlay
 *  Replace specified substring of first string with second
 *
 * The SQL standard defines OVERLAY() in terms of substring and concatenation.
 * This code is a direct implementation of what the standard says.
 */
Datum
byteaoverlay(PG_FUNCTION_ARGS)
{
    bytea      *t1 = PG_GETARG_BYTEA_PP(0);
    bytea      *t2 = PG_GETARG_BYTEA_PP(1);
    int         sp = PG_GETARG_INT32(2);    /* substring start position */
    int         sl = PG_GETARG_INT32(3);    /* substring length */
 
    PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
}
 
Datum
byteaoverlay_no_len(PG_FUNCTION_ARGS)
{
    bytea      *t1 = PG_GETARG_BYTEA_PP(0);
    bytea      *t2 = PG_GETARG_BYTEA_PP(1);
    int         sp = PG_GETARG_INT32(2);    /* substring start position */
    int         sl;
 
    sl = VARSIZE_ANY_EXHDR(t2); /* defaults to length(t2) */
    PG_RETURN_BYTEA_P(bytea_overlay(t1, t2, sp, sl));
}
 
/*
 * bytea_substr()
 * Return a substring starting at the specified position.
 * Cloned from text_substr and modified as required.
 *
 * Input:
 *  - string
 *  - starting position (is one-based)
 *  - string length (optional)
 *
 * If the starting position is zero or less, then return from the start of the string
 * adjusting the length to be consistent with the "negative start" per SQL.
 * If the length is less than zero, an ERROR is thrown. If no third argument
 * (length) is provided, the length to the end of the string is assumed.
 */
Datum
bytea_substr(PG_FUNCTION_ARGS)
{
    PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
                                      PG_GETARG_INT32(1),
                                      PG_GETARG_INT32(2),
                                      false));
}
 
/*
 * bytea_substr_no_len -
 *    Wrapper to avoid opr_sanity failure due to
 *    one function accepting a different number of args.
 */
Datum
bytea_substr_no_len(PG_FUNCTION_ARGS)
{
    PG_RETURN_BYTEA_P(bytea_substring(PG_GETARG_DATUM(0),
                                      PG_GETARG_INT32(1),
                                      -1,
                                      true));
}
 
/*
 * bit_count
 */
Datum
bytea_bit_count(PG_FUNCTION_ARGS)
{
    bytea      *t1 = PG_GETARG_BYTEA_PP(0);
 
    PG_RETURN_INT64(pg_popcount(VARDATA_ANY(t1), VARSIZE_ANY_EXHDR(t1)));
}
 
/*
 * byteapos -
 *    Return the position of the specified substring.
 *    Implements the SQL POSITION() function.
 * Cloned from textpos and modified as required.
 */
Datum
byteapos(PG_FUNCTION_ARGS)
{
    bytea      *t1 = PG_GETARG_BYTEA_PP(0);
    bytea      *t2 = PG_GETARG_BYTEA_PP(1);
    int         pos;
    int         px,
                p;
    int         len1,
                len2;
    char       *p1,
               *p2;
 
    len1 = VARSIZE_ANY_EXHDR(t1);
    len2 = VARSIZE_ANY_EXHDR(t2);
 
    if (len2 <= 0)
        PG_RETURN_INT32(1);     /* result for empty pattern */
 
    p1 = VARDATA_ANY(t1);
    p2 = VARDATA_ANY(t2);
 
    pos = 0;
    px = (len1 - len2);
    for (p = 0; p <= px; p++)
    {
        if ((*p2 == *p1) && (memcmp(p1, p2, len2) == 0))
        {
            pos = p + 1;
            break;
        };
        p1++;
    };
 
    PG_RETURN_INT32(pos);
}
 
/*-------------------------------------------------------------
 * byteaGetByte
 *
 * this routine treats "bytea" as an array of bytes.
 * It returns the Nth byte (a number between 0 and 255).
 *-------------------------------------------------------------
 */
Datum
byteaGetByte(PG_FUNCTION_ARGS)
{
    bytea      *v = PG_GETARG_BYTEA_PP(0);
    int32       n = PG_GETARG_INT32(1);
    int         len;
    int         byte;
 
    len = VARSIZE_ANY_EXHDR(v);
 
    if (n < 0 || n >= len)
        ereport(ERROR,
                (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
                 errmsg("index %d out of valid range, 0..%d",
                        n, len - 1)));
 
    byte = ((unsigned char *) VARDATA_ANY(v))[n];
 
    PG_RETURN_INT32(byte);
}
 
/*-------------------------------------------------------------
 * byteaGetBit
 *
 * This routine treats a "bytea" type like an array of bits.
 * It returns the value of the Nth bit (0 or 1).
 *
 *-------------------------------------------------------------
 */
Datum
byteaGetBit(PG_FUNCTION_ARGS)
{
    bytea      *v = PG_GETARG_BYTEA_PP(0);
    int64       n = PG_GETARG_INT64(1);
    int         byteNo,
                bitNo;
    int         len;
    int         byte;
 
    len = VARSIZE_ANY_EXHDR(v);
 
    if (n < 0 || n >= (int64) len * 8)
        ereport(ERROR,
                (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
                 errmsg("index %" PRId64 " out of valid range, 0..%" PRId64,
                        n, (int64) len * 8 - 1)));
 
    /* n/8 is now known < len, so safe to cast to int */
    byteNo = (int) (n / 8);
    bitNo = (int) (n % 8);
 
    byte = ((unsigned char *) VARDATA_ANY(v))[byteNo];
 
    if (byte & (1 << bitNo))
        PG_RETURN_INT32(1);
    else
        PG_RETURN_INT32(0);
}
 
/*-------------------------------------------------------------
 * byteaSetByte
 *
 * Given an instance of type 'bytea' creates a new one with
 * the Nth byte set to the given value.
 *
 *-------------------------------------------------------------
 */
Datum
byteaSetByte(PG_FUNCTION_ARGS)
{
    bytea      *res = PG_GETARG_BYTEA_P_COPY(0);
    int32       n = PG_GETARG_INT32(1);
    int32       newByte = PG_GETARG_INT32(2);
    int         len;
 
    len = VARSIZE(res) - VARHDRSZ;
 
    if (n < 0 || n >= len)
        ereport(ERROR,
                (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
                 errmsg("index %d out of valid range, 0..%d",
                        n, len - 1)));
 
    /*
     * Now set the byte.
     */
    ((unsigned char *) VARDATA(res))[n] = newByte;
 
    PG_RETURN_BYTEA_P(res);
}
 
/*-------------------------------------------------------------
 * byteaSetBit
 *
 * Given an instance of type 'bytea' creates a new one with
 * the Nth bit set to the given value.
 *
 *-------------------------------------------------------------
 */
Datum
byteaSetBit(PG_FUNCTION_ARGS)
{
    bytea      *res = PG_GETARG_BYTEA_P_COPY(0);
    int64       n = PG_GETARG_INT64(1);
    int32       newBit = PG_GETARG_INT32(2);
    int         len;
    int         oldByte,
                newByte;
    int         byteNo,
                bitNo;
 
    len = VARSIZE(res) - VARHDRSZ;
 
    if (n < 0 || n >= (int64) len * 8)
        ereport(ERROR,
                (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
                 errmsg("index %" PRId64 " out of valid range, 0..%" PRId64,
                        n, (int64) len * 8 - 1)));
 
    /* n/8 is now known < len, so safe to cast to int */
    byteNo = (int) (n / 8);
    bitNo = (int) (n % 8);
 
    /*
     * sanity check!
     */
    if (newBit != 0 && newBit != 1)
        ereport(ERROR,
                (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
                 errmsg("new bit must be 0 or 1")));
 
    /*
     * Update the byte.
     */
    oldByte = ((unsigned char *) VARDATA(res))[byteNo];
 
    if (newBit == 0)
        newByte = oldByte & (~(1 << bitNo));
    else
        newByte = oldByte | (1 << bitNo);
 
    ((unsigned char *) VARDATA(res))[byteNo] = newByte;
 
    PG_RETURN_BYTEA_P(res);
}
 
/*
 * Return reversed bytea
 */
Datum
bytea_reverse(PG_FUNCTION_ARGS)
{
    bytea      *v = PG_GETARG_BYTEA_PP(0);
    const char *p = VARDATA_ANY(v);
    int         len = VARSIZE_ANY_EXHDR(v);
    const char *endp = p + len;
    bytea      *result = palloc(len + VARHDRSZ);
    char       *dst = (char *) VARDATA(result) + len;
 
    SET_VARSIZE(result, len + VARHDRSZ);
 
    while (p < endp)
        *(--dst) = *p++;
 
    PG_RETURN_BYTEA_P(result);
}
 
 
/*****************************************************************************
 *  Comparison Functions used for bytea
 *
 * Note: btree indexes need these routines not to leak memory; therefore,
 * be careful to free working copies of toasted datums.  Most places don't
 * need to be so careful.
 *****************************************************************************/
 
Datum
byteaeq(PG_FUNCTION_ARGS)
{
    Datum       arg1 = PG_GETARG_DATUM(0);
    Datum       arg2 = PG_GETARG_DATUM(1);
    bool        result;
    Size        len1,
                len2;
 
    /*
     * We can use a fast path for unequal lengths, which might save us from
     * having to detoast one or both values.
     */
    len1 = toast_raw_datum_size(arg1);
    len2 = toast_raw_datum_size(arg2);
    if (len1 != len2)
        result = false;
    else
    {
        bytea      *barg1 = DatumGetByteaPP(arg1);
        bytea      *barg2 = DatumGetByteaPP(arg2);
 
        result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
                         len1 - VARHDRSZ) == 0);
 
        PG_FREE_IF_COPY(barg1, 0);
        PG_FREE_IF_COPY(barg2, 1);
    }
 
    PG_RETURN_BOOL(result);
}
 
Datum
byteane(PG_FUNCTION_ARGS)
{
    Datum       arg1 = PG_GETARG_DATUM(0);
    Datum       arg2 = PG_GETARG_DATUM(1);
    bool        result;
    Size        len1,
                len2;
 
    /*
     * We can use a fast path for unequal lengths, which might save us from
     * having to detoast one or both values.
     */
    len1 = toast_raw_datum_size(arg1);
    len2 = toast_raw_datum_size(arg2);
    if (len1 != len2)
        result = true;
    else
    {
        bytea      *barg1 = DatumGetByteaPP(arg1);
        bytea      *barg2 = DatumGetByteaPP(arg2);
 
        result = (memcmp(VARDATA_ANY(barg1), VARDATA_ANY(barg2),
                         len1 - VARHDRSZ) != 0);
 
        PG_FREE_IF_COPY(barg1, 0);
        PG_FREE_IF_COPY(barg2, 1);
    }
 
    PG_RETURN_BOOL(result);
}
 
Datum
bytealt(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
 
    PG_FREE_IF_COPY(arg1, 0);
    PG_FREE_IF_COPY(arg2, 1);
 
    PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 < len2)));
}
 
Datum
byteale(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
 
    PG_FREE_IF_COPY(arg1, 0);
    PG_FREE_IF_COPY(arg2, 1);
 
    PG_RETURN_BOOL((cmp < 0) || ((cmp == 0) && (len1 <= len2)));
}
 
Datum
byteagt(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
 
    PG_FREE_IF_COPY(arg1, 0);
    PG_FREE_IF_COPY(arg2, 1);
 
    PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 > len2)));
}
 
Datum
byteage(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
 
    PG_FREE_IF_COPY(arg1, 0);
    PG_FREE_IF_COPY(arg2, 1);
 
    PG_RETURN_BOOL((cmp > 0) || ((cmp == 0) && (len1 >= len2)));
}
 
Datum
byteacmp(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
    if ((cmp == 0) && (len1 != len2))
        cmp = (len1 < len2) ? -1 : 1;
 
    PG_FREE_IF_COPY(arg1, 0);
    PG_FREE_IF_COPY(arg2, 1);
 
    PG_RETURN_INT32(cmp);
}
 
Datum
bytea_larger(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    bytea      *result;
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
    result = ((cmp > 0) || ((cmp == 0) && (len1 > len2)) ? arg1 : arg2);
 
    PG_RETURN_BYTEA_P(result);
}
 
Datum
bytea_smaller(PG_FUNCTION_ARGS)
{
    bytea      *arg1 = PG_GETARG_BYTEA_PP(0);
    bytea      *arg2 = PG_GETARG_BYTEA_PP(1);
    bytea      *result;
    int         len1,
                len2;
    int         cmp;
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    cmp = memcmp(VARDATA_ANY(arg1), VARDATA_ANY(arg2), Min(len1, len2));
    result = ((cmp < 0) || ((cmp == 0) && (len1 < len2)) ? arg1 : arg2);
 
    PG_RETURN_BYTEA_P(result);
}
 
/*
 * sortsupport comparison func
 */
static int
byteafastcmp(Datum x, Datum y, SortSupport ssup)
{
    bytea      *arg1 = DatumGetByteaPP(x);
    bytea      *arg2 = DatumGetByteaPP(y);
    char       *a1p,
               *a2p;
    int         len1,
                len2,
                result;
 
    a1p = VARDATA_ANY(arg1);
    a2p = VARDATA_ANY(arg2);
 
    len1 = VARSIZE_ANY_EXHDR(arg1);
    len2 = VARSIZE_ANY_EXHDR(arg2);
 
    result = memcmp(a1p, a2p, Min(len1, len2));
    if ((result == 0) && (len1 != len2))
        result = (len1 < len2) ? -1 : 1;
 
    /* We can't afford to leak memory here. */
    if (PointerGetDatum(arg1) != x)
        pfree(arg1);
    if (PointerGetDatum(arg2) != y)
        pfree(arg2);
 
    return result;
}
 
/*
 * Conversion routine for sortsupport.  Converts original to abbreviated key
 * representation.  Our encoding strategy is simple -- pack the first 8 bytes
 * of the bytea data into a Datum (on little-endian machines, the bytes are
 * stored in reverse order), and treat it as an unsigned integer.
 */
static Datum
bytea_abbrev_convert(Datum original, SortSupport ssup)
{
    const size_t max_prefix_bytes = sizeof(Datum);
    ByteaSortSupport *bss = (ByteaSortSupport *) ssup->ssup_extra;
    bytea      *authoritative = DatumGetByteaPP(original);
    char       *authoritative_data = VARDATA_ANY(authoritative);
    Datum       res;
    char       *pres;
    int         len;
    uint32      hash;
 
    pres = (char *) &res;
 
    /* memset(), so any non-overwritten bytes are NUL */
    memset(pres, 0, max_prefix_bytes);
    len = VARSIZE_ANY_EXHDR(authoritative);
 
    /*
     * Short byteas will have terminating NUL bytes in the abbreviated datum.
     * Abbreviated comparison need not make a distinction between these NUL
     * bytes, and NUL bytes representing actual NULs in the authoritative
     * representation.
     *
     * Hopefully a comparison at or past one abbreviated key's terminating NUL
     * byte will resolve the comparison without consulting the authoritative
     * representation; specifically, some later non-NUL byte in the longer
     * bytea can resolve the comparison against a subsequent terminating NUL
     * in the shorter bytea.  There will usually be what is effectively a
     * "length-wise" resolution there and then.
     *
     * If that doesn't work out -- if all bytes in the longer bytea positioned
     * at or past the offset of the smaller bytea (first) terminating NUL are
     * actually representative of NUL bytes in the authoritative binary bytea
     * (perhaps with some *terminating* NUL bytes towards the end of the
     * longer bytea iff it happens to still be small) -- then an authoritative
     * tie-breaker will happen, and do the right thing: explicitly consider
     * bytea length.
     */
    memcpy(pres, authoritative_data, Min(len, max_prefix_bytes));
 
    /*
     * Maintain approximate cardinality of both abbreviated keys and original,
     * authoritative keys using HyperLogLog.  Used as cheap insurance against
     * the worst case, where we do many string abbreviations for no saving in
     * full memcmp()-based comparisons.  These statistics are used by
     * bytea_abbrev_abort().
     *
     * First, Hash key proper, or a significant fraction of it.  Mix in length
     * in order to compensate for cases where differences are past
     * PG_CACHE_LINE_SIZE bytes, so as to limit the overhead of hashing.
     */
    hash = DatumGetUInt32(hash_any((unsigned char *) authoritative_data,
                                   Min(len, PG_CACHE_LINE_SIZE)));
 
    if (len > PG_CACHE_LINE_SIZE)
        hash ^= DatumGetUInt32(hash_uint32((uint32) len));
 
    addHyperLogLog(&bss->full_card, hash);
 
    /* Hash abbreviated key */
    {
        uint32      tmp;
 
        tmp = DatumGetUInt32(res) ^ (uint32) (DatumGetUInt64(res) >> 32);
        hash = DatumGetUInt32(hash_uint32(tmp));
    }
 
    addHyperLogLog(&bss->abbr_card, hash);
 
    /*
     * Byteswap on little-endian machines.
     *
     * This is needed so that ssup_datum_unsigned_cmp() works correctly on all
     * platforms.
     */
    res = DatumBigEndianToNative(res);
 
    /* Don't leak memory here */
    if (PointerGetDatum(authoritative) != original)
        pfree(authoritative);
 
    return res;
}
 
/*
 * Callback for estimating effectiveness of abbreviated key optimization, using
 * heuristic rules.  Returns value indicating if the abbreviation optimization
 * should be aborted, based on its projected effectiveness.
 *
 * This is based on varstr_abbrev_abort(), but some comments have been elided
 * for brevity. See there for more details.
 */
static bool
bytea_abbrev_abort(int memtupcount, SortSupport ssup)
{
    ByteaSortSupport *bss = (ByteaSortSupport *) ssup->ssup_extra;
    double      abbrev_distinct,
                key_distinct;
 
    Assert(ssup->abbreviate);
 
    /* Have a little patience */
    if (memtupcount < 100)
        return false;
 
    abbrev_distinct = estimateHyperLogLog(&bss->abbr_card);
    key_distinct = estimateHyperLogLog(&bss->full_card);
 
    /*
     * Clamp cardinality estimates to at least one distinct value.  While
     * NULLs are generally disregarded, if only NULL values were seen so far,
     * that might misrepresent costs if we failed to clamp.
     */
    if (abbrev_distinct < 1.0)
        abbrev_distinct = 1.0;
 
    if (key_distinct < 1.0)
        key_distinct = 1.0;
 
    if (trace_sort)
    {
        double      norm_abbrev_card = abbrev_distinct / (double) memtupcount;
 
        elog(LOG, "bytea_abbrev: abbrev_distinct after %d: %f "
             "(key_distinct: %f, norm_abbrev_card: %f, prop_card: %f)",
             memtupcount, abbrev_distinct, key_distinct, norm_abbrev_card,
             bss->prop_card);
    }
 
    /*
     * If the number of distinct abbreviated keys approximately matches the
     * number of distinct original keys, continue with abbreviation.
     */
    if (abbrev_distinct > key_distinct * bss->prop_card)
    {
        /*
         * Decay required cardinality aggressively after 10,000 tuples.
         */
        if (memtupcount > 10000)
            bss->prop_card *= 0.65;
 
        return false;
    }
 
    /*
     * Abort abbreviation strategy.
     */
    if (trace_sort)
        elog(LOG, "bytea_abbrev: aborted abbreviation at %d "
             "(abbrev_distinct: %f, key_distinct: %f, prop_card: %f)",
             memtupcount, abbrev_distinct, key_distinct, bss->prop_card);
 
    return true;
}
 
Datum
bytea_sortsupport(PG_FUNCTION_ARGS)
{
    SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
    MemoryContext oldcontext;
 
    oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
 
    ssup->comparator = byteafastcmp;
 
    /*
     * Set up abbreviation support if requested.
     */
    if (ssup->abbreviate)
    {
        ByteaSortSupport *bss;
 
        bss = palloc_object(ByteaSortSupport);
        bss->abbreviate = true;
        bss->prop_card = 0.20;
        initHyperLogLog(&bss->abbr_card, 10);
        initHyperLogLog(&bss->full_card, 10);
 
        ssup->ssup_extra = bss;
        ssup->abbrev_full_comparator = ssup->comparator;
        ssup->comparator = ssup_datum_unsigned_cmp;
        ssup->abbrev_converter = bytea_abbrev_convert;
        ssup->abbrev_abort = bytea_abbrev_abort;
    }
 
    MemoryContextSwitchTo(oldcontext);
 
    PG_RETURN_VOID();
}
 
/* Cast bytea -> int2 */
Datum
bytea_int2(PG_FUNCTION_ARGS)
{
    bytea      *v = PG_GETARG_BYTEA_PP(0);
    int         len = VARSIZE_ANY_EXHDR(v);
    uint16      result;
 
    /* Check that the byte array is not too long */
    if (len > sizeof(result))
        ereport(ERROR,
                errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
                errmsg("smallint out of range"));
 
    /* Convert it to an integer; most significant bytes come first */
    result = 0;
    for (int i = 0; i < len; i++)
    {
        result <<= BITS_PER_BYTE;
        result |= ((unsigned char *) VARDATA_ANY(v))[i];
    }
 
    PG_RETURN_INT16(result);
}
 
/* Cast bytea -> int4 */
Datum
bytea_int4(PG_FUNCTION_ARGS)
{
    bytea      *v = PG_GETARG_BYTEA_PP(0);
    int         len = VARSIZE_ANY_EXHDR(v);
    uint32      result;
 
    /* Check that the byte array is not too long */
    if (len > sizeof(result))
        ereport(ERROR,
                errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
                errmsg("integer out of range"));
 
    /* Convert it to an integer; most significant bytes come first */
    result = 0;
    for (int i = 0; i < len; i++)
    {
        result <<= BITS_PER_BYTE;
        result |= ((unsigned char *) VARDATA_ANY(v))[i];
    }
 
    PG_RETURN_INT32(result);
}
 
/* Cast bytea -> int8 */
Datum
bytea_int8(PG_FUNCTION_ARGS)
{
    bytea      *v = PG_GETARG_BYTEA_PP(0);
    int         len = VARSIZE_ANY_EXHDR(v);
    uint64      result;
 
    /* Check that the byte array is not too long */
    if (len > sizeof(result))
        ereport(ERROR,
                errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
                errmsg("bigint out of range"));
 
    /* Convert it to an integer; most significant bytes come first */
    result = 0;
    for (int i = 0; i < len; i++)
    {
        result <<= BITS_PER_BYTE;
        result |= ((unsigned char *) VARDATA_ANY(v))[i];
    }
 
    PG_RETURN_INT64(result);
}
 
/* Cast int2 -> bytea; can just use int2send() */
Datum
int2_bytea(PG_FUNCTION_ARGS)
{
    return int2send(fcinfo);
}
 
/* Cast int4 -> bytea; can just use int4send() */
Datum
int4_bytea(PG_FUNCTION_ARGS)
{
    return int4send(fcinfo);
}
 
/* Cast int8 -> bytea; can just use int8send() */
Datum
int8_bytea(PG_FUNCTION_ARGS)
{
    return int8send(fcinfo);
}