From c363ac4ce91fa72880d94e96084670e134f98f06 Mon Sep 17 00:00:00 2001 From: sohardforaname Date: Sun, 10 May 2026 21:42:17 +0800 Subject: [PATCH 1/2] numeric: make NumericVar storage states explicit and use them in core paths Clarify the in-memory storage model of NumericVar by making buffer ownership and mutability states explicit, and reorganize the related helper layer around that model. This introduces explicit storage states for empty, borrowed, owned-inline, and owned-heap representations, together with borrow-kind tracking for borrowed digit views. It also adds BufferedNumericVar as the inline-home wrapper used by the new model. The patch then connects the new helper semantics to core comparison, hashing, and basic arithmetic paths, while keeping higher-level numeric algorithms structurally close to the existing implementation. The goal is not to alter SQL-visible numeric behavior, but to make the existing internal contract around NumericVar storage easier to understand, document, and maintain. --- src/backend/utils/adt/numeric.c | 1197 ++++++++++++++++++++++++------- 1 file changed, 949 insertions(+), 248 deletions(-) diff --git a/src/backend/utils/adt/numeric.c b/src/backend/utils/adt/numeric.c index cb23dfe9b95..f85abf12881 100644 --- a/src/backend/utils/adt/numeric.c +++ b/src/backend/utils/adt/numeric.c @@ -104,6 +104,7 @@ typedef int16 NumericDigit; #endif #define NBASE_SQR (NBASE * NBASE) +#define NUMERIC_INLINE_BUFFER_SIZE 128 /* * The Numeric type as stored on disk. @@ -261,66 +262,165 @@ struct NumericData #define NUMERIC_WEIGHT_MAX PG_INT16_MAX /* ---------- - * NumericVar is the format we use for arithmetic. The digit-array part - * is the same as the NumericData storage format, but the header is more - * complex. - * - * The value represented by a NumericVar is determined by the sign, weight, - * ndigits, and digits[] array. If it is a "special" value (NaN or Inf) - * then only the sign field matters; ndigits should be zero, and the weight - * and dscale fields are ignored. - * - * Note: the first digit of a NumericVar's value is assumed to be multiplied - * by NBASE ** weight. Another way to say it is that there are weight+1 - * digits before the decimal point. It is possible to have weight < 0. - * - * buf points at the physical start of the palloc'd digit buffer for the - * NumericVar. digits points at the first digit in actual use (the one - * with the specified weight). We normally leave an unused digit or two - * (preset to zeroes) between buf and digits, so that there is room to store - * a carry out of the top digit without reallocating space. We just need to - * decrement digits (and increment weight) to make room for the carry digit. - * (There is no such extra space in a numeric value stored in the database, - * only in a NumericVar in memory.) - * - * If buf is NULL then the digit buffer isn't actually palloc'd and should - * not be freed --- see the constants below for an example. - * - * dscale, or display scale, is the nominal precision expressed as number - * of digits after the decimal point (it must always be >= 0 at present). - * dscale may be more than the number of physically stored fractional digits, - * implying that we have suppressed storage of significant trailing zeroes. - * It should never be less than the number of stored digits, since that would - * imply hiding digits that are present. NOTE that dscale is always expressed - * in *decimal* digits, and so it may correspond to a fractional number of - * base-NBASE digits --- divide by DEC_DIGITS to convert to NBASE digits. - * - * rscale, or result scale, is the target precision for a computation. - * Like dscale it is expressed as number of *decimal* digits after the decimal - * point, and is always >= 0 at present. - * Note that rscale is not stored in variables --- it's figured on-the-fly - * from the dscales of the inputs. - * - * While we consistently use "weight" to refer to the base-NBASE weight of - * a numeric value, it is convenient in some scale-related calculations to - * make use of the base-10 weight (ie, the approximate log10 of the value). - * To avoid confusion, such a decimal-units weight is called a "dweight". - * - * NB: All the variable-level functions are written in a style that makes it - * possible to give one and the same variable as argument and destination. - * This is feasible because the digit buffer is separate from the variable. + * NumericVar is the in-memory working representation of a numeric value. + * + * A NumericVar always stores the logical numeric metadata (ndigits, weight, + * sign, dscale), but the physical storage backing digits[] is stateful. + * + * There are four storage states: + * + * 1. NUMVAR_EMPTY + * The variable carries no digits storage and no digit payload. + * + * 2. NUMVAR_BORROWED + * digits points to storage owned by some other representation. + * In this state the current NumericVar does not own writable digit + * storage, and buf is NULL. + * + * Borrowed storage is further classified by borrow_kind: + * + * - NUMVAR_BORROW_EXTERNAL + * digits points into external numeric storage, such as a detoasted + * Numeric datum. This storage is read-only from the NumericVar's + * perspective and cannot be upgraded in-place to writable inline + * storage. + * + * - NUMVAR_BORROW_INLINE + * digits points into numeric storage currently materialized inside the + * inline buffer of an associated BufferedNumericVar. This is still a + * read-only borrowed view, but because the underlying memory belongs to + * the same buffered home, it may later be converted into writable inline + * working storage. + * + * 3. NUMVAR_OWNED_INLINE + * buf points to the inline buffer associated with this NumericVar, and + * digits points somewhere within that owned writable storage. + * + * 4. NUMVAR_OWNED_HEAP + * buf points to palloc'd writable storage owned by this NumericVar, and + * digits points somewhere within that owned writable storage. + * + * In owned states, buf is the base of the current writable digit storage and + * digits points to the logical first digit. As in the historical NumericVar + * representation, owned storage is initially allocated with a spare leading + * digit so that carry propagation can enlarge the value without immediate + * reallocation. Stored numeric datums do not have this extra working space; + * it exists only in the in-memory working representation. Therefore, a + * borrowed view of a stored/detoasted Numeric datum must not be treated as a + * writable work buffer without an explicit state transition. + * + * Note, however, that digits is not required to remain equal to buf + 1 for + * the lifetime of an owned NumericVar. Subsequent normalization steps such as + * stripping leading zeroes may advance digits within the owned buffer, as long + * as the active digit range remains within the storage owned by buf. + * + * inline_buf, when non-NULL, identifies the fixed-size inline home provided by + * BufferedNumericVar. It does not by itself imply that the variable is + * currently in NUMVAR_OWNED_INLINE state. For example, a NumericVar may be in + * NUMVAR_BORROWED or NUMVAR_OWNED_HEAP state while still retaining an inline + * home for possible future reuse. + * + * capacity is the physical capacity of the current owned buffer, including the + * spare leading digit. It is meaningful only in owned states, and is zero in + * empty or borrowed states. + * + * As with the traditional implementation, variable-level functions should be + * written so that a NumericVar may safely be used as both input and + * destination. State transitions must preserve that property. + * + * NumericVar state invariants: + * + * NUMVAR_EMPTY: + * buf == NULL + * digits == NULL + * borrow_kind == NUMVAR_BORROW_NONE + * capacity == 0 + * + * NUMVAR_BORROWED: + * buf == NULL + * borrow_kind != NUMVAR_BORROW_NONE + * capacity == 0 + * digits points to borrowed storage, unless sign is special + * + * NUMVAR_OWNED_INLINE: + * buf == inline_buf + * buf != NULL + * digits != NULL + * borrow_kind == NUMVAR_BORROW_NONE + * capacity == NUMERIC_INLINE_BUFFER_SIZE + * the active digit range [digits, digits + ndigits) lies within + * the owned storage range [buf, buf + capacity) + * + * NUMVAR_OWNED_HEAP: + * buf != NULL + * digits != NULL + * borrow_kind == NUMVAR_BORROW_NONE + * capacity > 0 + * the active digit range [digits, digits + ndigits) lies within + * the owned storage range [buf, buf + capacity) + * + * Special values are represented through sign, not through a separate storage + * state. For NaN and infinities, only sign is semantically meaningful. * ---------- */ + +typedef enum NUMERIC_VAR_STATUS +{ + NUMVAR_EMPTY, + NUMVAR_BORROWED, + NUMVAR_OWNED_INLINE, + NUMVAR_OWNED_HEAP +} NUMERIC_VAR_STATUS; + +typedef enum NUMERIC_BORROW_KIND +{ + NUMVAR_BORROW_NONE, + NUMVAR_BORROW_EXTERNAL, + NUMVAR_BORROW_INLINE +} NUMERIC_BORROW_KIND; + typedef struct NumericVar { - int ndigits; /* # of digits in digits[] - can be 0! */ + /* logical numeric metadata */ + int ndigits; /* # of digits in digits[]; may be 0 */ int weight; /* weight of first digit */ int sign; /* NUMERIC_POS, _NEG, _NAN, _PINF, or _NINF */ int dscale; /* display scale */ - NumericDigit *buf; /* start of palloc'd space for digits[] */ - NumericDigit *digits; /* base-NBASE digits */ + + /* storage state metadata */ + NUMERIC_VAR_STATUS state; /* current storage state */ + NUMERIC_BORROW_KIND borrow_kind; /* source of borrowed digits, if + * borrowed */ + int capacity; /* current owned buffer capacity, spare digit + * included */ + + /* current digit storage */ + NumericDigit *buf; /* base of current owned writable storage, or + * NULL */ + NumericDigit *digits; /* logical first base-NBASE digit, or NULL */ + + /* optional fixed-size inline home */ + NumericDigit *inline_buf; /* inline buffer of associated + * BufferedNumericVar, or NULL */ } NumericVar; +#define NUMERIC_VARLENA_SIZE (VARHDRSZ / sizeof(NumericDigit) + \ + (sizeof(uint16) + sizeof(int16)) / sizeof(NumericDigit) + \ + NUMERIC_INLINE_BUFFER_SIZE + DIV_GUARD_DIGITS + 4) + +typedef struct BufferedNumericVar +{ + NumericVar var; + int64 magic; /* used for align with 8-byte */ + NumericDigit buffer[NUMERIC_VARLENA_SIZE]; +} BufferedNumericVar; + + +#define BUFFERED_NUMERIC_VAR_MAGIC 0x4E554D455249435F /* 'NUMERIC_MAGIC' */ + +#define NUMERICVAR_IS_SPECIAL(v) \ + ((v)->sign == NUMERIC_NAN || (v)->sign == NUMERIC_PINF || (v)->sign == NUMERIC_NINF) + /* ---------- * Data for generate_series @@ -415,18 +515,18 @@ typedef struct NumericSumAccum */ static const NumericDigit const_zero_data[1] = {0}; static const NumericVar const_zero = -{0, 0, NUMERIC_POS, 0, NULL, (NumericDigit *) const_zero_data}; +{0, 0, NUMERIC_POS, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, (NumericDigit *) const_zero_data, NULL}; static const NumericDigit const_one_data[1] = {1}; static const NumericVar const_one = -{1, 0, NUMERIC_POS, 0, NULL, (NumericDigit *) const_one_data}; +{1, 0, NUMERIC_POS, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, (NumericDigit *) const_one_data, NULL}; static const NumericVar const_minus_one = -{1, 0, NUMERIC_NEG, 0, NULL, (NumericDigit *) const_one_data}; +{1, 0, NUMERIC_NEG, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, (NumericDigit *) const_one_data, NULL}; static const NumericDigit const_two_data[1] = {2}; static const NumericVar const_two = -{1, 0, NUMERIC_POS, 0, NULL, (NumericDigit *) const_two_data}; +{1, 0, NUMERIC_POS, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, (NumericDigit *) const_two_data, NULL}; #if DEC_DIGITS == 4 static const NumericDigit const_zero_point_nine_data[1] = {9000}; @@ -436,7 +536,7 @@ static const NumericDigit const_zero_point_nine_data[1] = {90}; static const NumericDigit const_zero_point_nine_data[1] = {9}; #endif static const NumericVar const_zero_point_nine = -{1, -1, NUMERIC_POS, 1, NULL, (NumericDigit *) const_zero_point_nine_data}; +{1, -1, NUMERIC_POS, 1, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, (NumericDigit *) const_zero_point_nine_data, NULL}; #if DEC_DIGITS == 4 static const NumericDigit const_one_point_one_data[2] = {1, 1000}; @@ -446,28 +546,31 @@ static const NumericDigit const_one_point_one_data[2] = {1, 10}; static const NumericDigit const_one_point_one_data[2] = {1, 1}; #endif static const NumericVar const_one_point_one = -{2, 0, NUMERIC_POS, 1, NULL, (NumericDigit *) const_one_point_one_data}; +{2, 0, NUMERIC_POS, 1, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, (NumericDigit *) const_one_point_one_data, NULL}; static const NumericVar const_nan = -{0, 0, NUMERIC_NAN, 0, NULL, NULL}; +{0, 0, NUMERIC_NAN, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, NULL, NULL}; static const NumericVar const_pinf = -{0, 0, NUMERIC_PINF, 0, NULL, NULL}; +{0, 0, NUMERIC_PINF, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, NULL, NULL}; static const NumericVar const_ninf = -{0, 0, NUMERIC_NINF, 0, NULL, NULL}; +{0, 0, NUMERIC_NINF, 0, NUMVAR_BORROWED, NUMVAR_BORROW_EXTERNAL, 0, NULL, NULL, NULL}; #if DEC_DIGITS == 4 static const int round_powers[4] = {0, 1000, 100, 10}; #endif - /* ---------- * Local functions * ---------- */ #ifdef NUMERIC_DEBUG +static inline int check_digit_overflow(const NumericVar *var); +static inline void check_numericvar_state(const NumericVar *var, const char *where); +#define CHECK_NUMERICVAR(v) check_numericvar_state((v), __func__) +#define CHECK_NUMERICVAR_AT(v, where) check_numericvar_state((v), (where)) static void dump_numeric(const char *str, Numeric num); static void dump_var(const char *str, NumericVar *var); #else @@ -475,15 +578,65 @@ static void dump_var(const char *str, NumericVar *var); #define dump_var(s,v) #endif -#define digitbuf_alloc(ndigits) \ - ((NumericDigit *) palloc((ndigits) * sizeof(NumericDigit))) -#define digitbuf_free(buf) \ - do { \ - if ((buf) != NULL) \ - pfree(buf); \ - } while (0) +/* + * digitbuf_alloc_raw() - + * + * Allocate raw heap-backed digit storage with the specified physical + * capacity. + * + * The capacity is measured in NumericDigit units and includes any spare + * leading digit space required by the caller. + * + * This is a low-level storage primitive. It does not inspect or update any + * NumericVar state, and it does not by itself establish a complete working + * NumericVar representation. + */ +static inline NumericDigit * +digitbuf_alloc_raw(int sz) +{ + return (NumericDigit *) palloc(sizeof(NumericDigit) * sz); +} -#define init_var(v) memset(v, 0, sizeof(NumericVar)) +/* + * init_var() - + * + * Initialize a NumericVar in NUMVAR_EMPTY storage state, with no active digit + * storage, no borrowed digit view, and no current digit payload. + * + * This establishes a clean starting point for later transitions into + * borrowed or owned states. + */ +static inline void +init_var(NumericVar *v) +{ + memset(v, 0, sizeof(NumericVar)); +} + +/* + * init_buffered_var() - + * + * Initialize a BufferedNumericVar so that its embedded NumericVar starts with + * writable inline working storage backed by the object's fixed-size inline + * buffer. + * + * The inline buffer is installed as the current owned storage, and the digit + * pointer is initialized to the normal working position after the spare + * leading digit. + */ +static void +init_buffered_var(BufferedNumericVar * bvar) +{ + bvar->magic = BUFFERED_NUMERIC_VAR_MAGIC; + init_var(&bvar->var); + + bvar->var.inline_buf = bvar->buffer; + bvar->var.capacity = NUMERIC_INLINE_BUFFER_SIZE; + bvar->var.state = NUMVAR_OWNED_INLINE; + bvar->var.borrow_kind = NUMVAR_BORROW_NONE; + bvar->var.buf = bvar->buffer; + bvar->var.buf[0] = 0; + bvar->var.digits = bvar->buffer + 1; +} #define NUMERIC_DIGITS(num) (NUMERIC_HEADER_IS_SHORT(num) ? \ (num)->choice.n_short.n_data : (num)->choice.n_long.n_data) @@ -494,8 +647,27 @@ static void dump_var(const char *str, NumericVar *var); (weight) <= NUMERIC_SHORT_WEIGHT_MAX && \ (weight) >= NUMERIC_SHORT_WEIGHT_MIN) +static void digitbuf_free(NumericVar *var); + +static NumericDigit *digitbuf_get(NumericVar *var, + int require_ndigits); + static void alloc_var(NumericVar *var, int ndigits); static void free_var(NumericVar *var); + +static bool result_aliases_input(const NumericVar *result, + const NumericVar *a, + const NumericVar *b); + +static NumericVar *prepare_result_var(NumericVar *result, + const NumericVar *a, + const NumericVar *b, + int res_ndigits, + BufferedNumericVar * tmp); + +static void finish_result_var(NumericVar *work_result, + NumericVar *result); + static void zero_var(NumericVar *var); static bool set_var_from_str(const char *str, const char *cp, @@ -507,7 +679,10 @@ static bool set_var_from_non_decimal_integer_str(const char *str, const char **endptr, Node *escontext); static void set_var_from_num(Numeric num, NumericVar *dest); -static void init_var_from_num(Numeric num, NumericVar *dest); +static Numeric init_var_from_num(Numeric num, NumericVar *dest); +static Numeric init_buffered_var_from_num(Numeric src, BufferedNumericVar * bvar); +static varlena *numeric_detoast_attr_to_buffer(varlena *attr, unsigned char *buffer, + size_t buffer_size); static void set_var_from_var(const NumericVar *value, NumericVar *dest); static char *get_str_from_var(const NumericVar *var); static char *get_str_from_var_sci(const NumericVar *var, int rscale); @@ -543,6 +718,7 @@ static int cmp_var_common(const NumericDigit *var1digits, int var1ndigits, int var1weight, int var1sign, const NumericDigit *var2digits, int var2ndigits, int var2weight, int var2sign); + static void add_var(const NumericVar *var1, const NumericVar *var2, NumericVar *result); static void sub_var(const NumericVar *var1, const NumericVar *var2, @@ -2416,8 +2592,8 @@ numeric_abbrev_convert_var(const NumericVar *var, NumericSortSupport *nss) Datum numeric_cmp(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); int result; result = cmp_numerics(num1, num2); @@ -2428,12 +2604,11 @@ numeric_cmp(PG_FUNCTION_ARGS) PG_RETURN_INT32(result); } - Datum numeric_eq(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); bool result; result = cmp_numerics(num1, num2) == 0; @@ -2447,8 +2622,8 @@ numeric_eq(PG_FUNCTION_ARGS) Datum numeric_ne(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); bool result; result = cmp_numerics(num1, num2) != 0; @@ -2462,8 +2637,8 @@ numeric_ne(PG_FUNCTION_ARGS) Datum numeric_gt(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); bool result; result = cmp_numerics(num1, num2) > 0; @@ -2477,8 +2652,8 @@ numeric_gt(PG_FUNCTION_ARGS) Datum numeric_ge(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); bool result; result = cmp_numerics(num1, num2) >= 0; @@ -2492,8 +2667,8 @@ numeric_ge(PG_FUNCTION_ARGS) Datum numeric_lt(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); bool result; result = cmp_numerics(num1, num2) < 0; @@ -2507,8 +2682,8 @@ numeric_lt(PG_FUNCTION_ARGS) Datum numeric_le(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); bool result; result = cmp_numerics(num1, num2) <= 0; @@ -2523,6 +2698,11 @@ static int cmp_numerics(Numeric num1, Numeric num2) { int result; + BufferedNumericVar var1; + BufferedNumericVar var2; + + num1 = init_buffered_var_from_num(num1, &var1); + num2 = init_buffered_var_from_num(num2, &var2); /* * We consider all NANs to be equal and larger than any non-NAN (including @@ -2714,7 +2894,7 @@ in_range_numeric_numeric(PG_FUNCTION_ARGS) Datum hash_numeric(PG_FUNCTION_ARGS) { - Numeric key = PG_GETARG_NUMERIC(0); + Numeric key = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); Datum digit_hash; Datum result; int weight; @@ -2723,6 +2903,9 @@ hash_numeric(PG_FUNCTION_ARGS) int i; int hash_len; NumericDigit *digits; + BufferedNumericVar var; + + key = init_buffered_var_from_num(key, &var); /* If it's NaN or infinity, don't try to hash the rest of the fields */ if (NUMERIC_IS_SPECIAL(key)) @@ -2865,12 +3048,16 @@ hash_numeric_extended(PG_FUNCTION_ARGS) Datum numeric_add(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); + Numeric res; res = numeric_add_safe(num1, num2, NULL); + PG_FREE_IF_COPY(num1, 0); + PG_FREE_IF_COPY(num2, 1); + PG_RETURN_NUMERIC(res); } @@ -2882,11 +3069,15 @@ numeric_add(PG_FUNCTION_ARGS) Numeric numeric_add_safe(Numeric num1, Numeric num2, Node *escontext) { - NumericVar arg1; - NumericVar arg2; - NumericVar result; + BufferedNumericVar arg1; + BufferedNumericVar arg2; + + BufferedNumericVar result; Numeric res; + num1 = init_buffered_var_from_num(num1, &arg1); + num2 = init_buffered_var_from_num(num2, &arg2); + /* * Handle NaN and infinities */ @@ -2915,18 +3106,15 @@ numeric_add_safe(Numeric num1, Numeric num2, Node *escontext) return make_result(&const_ninf); } - /* - * Unpack the values, let add_var() compute the result and return it. - */ - init_var_from_num(num1, &arg1); - init_var_from_num(num2, &arg2); + init_buffered_var(&result); - init_var(&result); - add_var(&arg1, &arg2, &result); + add_var(&arg1.var, &arg2.var, &result.var); - res = make_result_safe(&result, escontext); + res = make_result_safe(&result.var, escontext); - free_var(&result); + free_var(&result.var); + free_var(&arg2.var); + free_var(&arg1.var); return res; } @@ -2940,12 +3128,16 @@ numeric_add_safe(Numeric num1, Numeric num2, Node *escontext) Datum numeric_sub(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); + Numeric res; res = numeric_sub_safe(num1, num2, NULL); + PG_FREE_IF_COPY(num1, 0); + PG_FREE_IF_COPY(num2, 1); + PG_RETURN_NUMERIC(res); } @@ -2958,11 +3150,17 @@ numeric_sub(PG_FUNCTION_ARGS) Numeric numeric_sub_safe(Numeric num1, Numeric num2, Node *escontext) { - NumericVar arg1; - NumericVar arg2; - NumericVar result; + BufferedNumericVar arg1; + BufferedNumericVar arg2; + BufferedNumericVar result; Numeric res; + /* + * Unpack the values, let sub_var() compute the result and return it. + */ + num1 = init_buffered_var_from_num(num1, &arg1); + num2 = init_buffered_var_from_num(num2, &arg2); + /* * Handle NaN and infinities */ @@ -2991,18 +3189,14 @@ numeric_sub_safe(Numeric num1, Numeric num2, Node *escontext) return make_result(&const_pinf); } - /* - * Unpack the values, let sub_var() compute the result and return it. - */ - init_var_from_num(num1, &arg1); - init_var_from_num(num2, &arg2); + init_buffered_var(&result); + sub_var(&arg1.var, &arg2.var, &result.var); - init_var(&result); - sub_var(&arg1, &arg2, &result); + res = make_result_safe(&result.var, escontext); - res = make_result_safe(&result, escontext); - - free_var(&result); + free_var(&result.var); + free_var(&arg2.var); + free_var(&arg1.var); return res; } @@ -3016,12 +3210,16 @@ numeric_sub_safe(Numeric num1, Numeric num2, Node *escontext) Datum numeric_mul(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); + Numeric res; res = numeric_mul_safe(num1, num2, fcinfo->context); + PG_FREE_IF_COPY(num1, 0); + PG_FREE_IF_COPY(num2, 1); + if (unlikely(SOFT_ERROR_OCCURRED(fcinfo->context))) PG_RETURN_NULL(); @@ -3037,11 +3235,25 @@ numeric_mul(PG_FUNCTION_ARGS) Numeric numeric_mul_safe(Numeric num1, Numeric num2, Node *escontext) { - NumericVar arg1; - NumericVar arg2; - NumericVar result; + BufferedNumericVar arg1; + BufferedNumericVar arg2; + BufferedNumericVar result; Numeric res; + /* + * Unpack the values, let mul_var() compute the result and return it. + * Unlike add_var() and sub_var(), mul_var() will round its result. In the + * case of numeric_mul(), which is invoked for the * operator on numerics, + * we request exact representation for the product (rscale = sum(dscale of + * arg1, dscale of arg2)). If the exact result has more digits after the + * decimal point than can be stored in a numeric, we round it. Rounding + * after computing the exact result ensures that the final result is + * correctly rounded (rounding in mul_var() using a truncated product + * would not guarantee this). + */ + num1 = init_buffered_var_from_num(num1, &arg1); + num2 = init_buffered_var_from_num(num2, &arg2); + /* * Handle NaN and infinities */ @@ -3102,29 +3314,17 @@ numeric_mul_safe(Numeric num1, Numeric num2, Node *escontext) Assert(false); } - /* - * Unpack the values, let mul_var() compute the result and return it. - * Unlike add_var() and sub_var(), mul_var() will round its result. In the - * case of numeric_mul(), which is invoked for the * operator on numerics, - * we request exact representation for the product (rscale = sum(dscale of - * arg1, dscale of arg2)). If the exact result has more digits after the - * decimal point than can be stored in a numeric, we round it. Rounding - * after computing the exact result ensures that the final result is - * correctly rounded (rounding in mul_var() using a truncated product - * would not guarantee this). - */ - init_var_from_num(num1, &arg1); - init_var_from_num(num2, &arg2); + init_buffered_var(&result); + mul_var(&arg1.var, &arg2.var, &result.var, arg1.var.dscale + arg2.var.dscale); - init_var(&result); - mul_var(&arg1, &arg2, &result, arg1.dscale + arg2.dscale); + if (result.var.dscale > NUMERIC_DSCALE_MAX) + round_var(&result.var, NUMERIC_DSCALE_MAX); - if (result.dscale > NUMERIC_DSCALE_MAX) - round_var(&result, NUMERIC_DSCALE_MAX); + res = make_result_safe(&result.var, escontext); - res = make_result_safe(&result, escontext); - - free_var(&result); + free_var(&result.var); + free_var(&arg2.var); + free_var(&arg1.var); return res; } @@ -3138,12 +3338,16 @@ numeric_mul_safe(Numeric num1, Numeric num2, Node *escontext) Datum numeric_div(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); + Numeric res; res = numeric_div_safe(num1, num2, NULL); + PG_FREE_IF_COPY(num1, 0); + PG_FREE_IF_COPY(num2, 1); + PG_RETURN_NUMERIC(res); } @@ -3156,12 +3360,18 @@ numeric_div(PG_FUNCTION_ARGS) Numeric numeric_div_safe(Numeric num1, Numeric num2, Node *escontext) { - NumericVar arg1; - NumericVar arg2; - NumericVar result; + BufferedNumericVar arg1; + BufferedNumericVar arg2; + BufferedNumericVar result; Numeric res; int rscale; + /* + * Unpack the arguments + */ + num1 = init_buffered_var_from_num(num1, &arg1); + num2 = init_buffered_var_from_num(num2, &arg2); + /* * Handle NaN and infinities */ @@ -3209,31 +3419,27 @@ numeric_div_safe(Numeric num1, Numeric num2, Node *escontext) return make_result(&const_zero); } - /* - * Unpack the arguments - */ - init_var_from_num(num1, &arg1); - init_var_from_num(num2, &arg2); - - init_var(&result); + init_buffered_var(&result); /* * Select scale for division result */ - rscale = select_div_scale(&arg1, &arg2); + rscale = select_div_scale(&arg1.var, &arg2.var); /* Check for division by zero */ - if (arg2.ndigits == 0 || arg2.digits[0] == 0) + if (arg2.var.ndigits == 0 || arg2.var.digits[0] == 0) goto division_by_zero; /* * Do the divide and return the result */ - div_var(&arg1, &arg2, &result, rscale, true, true); + div_var(&arg1.var, &arg2.var, &result.var, rscale, true, true); - res = make_result_safe(&result, escontext); + res = make_result_safe(&result.var, escontext); - free_var(&result); + free_var(&result.var); + free_var(&arg2.var); + free_var(&arg1.var); return res; @@ -3450,17 +3656,17 @@ numeric_inc(PG_FUNCTION_ARGS) Datum numeric_smaller(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); /* * Use cmp_numerics so that this will agree with the comparison operators, * particularly as regards comparisons involving NaN. */ if (cmp_numerics(num1, num2) < 0) - PG_RETURN_NUMERIC(num1); + PG_RETURN_DATUM(PG_GETARG_DATUM(0)); else - PG_RETURN_NUMERIC(num2); + PG_RETURN_DATUM(PG_GETARG_DATUM(1)); } @@ -3472,17 +3678,17 @@ numeric_smaller(PG_FUNCTION_ARGS) Datum numeric_larger(PG_FUNCTION_ARGS) { - Numeric num1 = PG_GETARG_NUMERIC(0); - Numeric num2 = PG_GETARG_NUMERIC(1); + Numeric num1 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(0)); + Numeric num2 = (Numeric) PG_DETOAST_DATUM_PACKED(PG_GETARG_DATUM(1)); /* * Use cmp_numerics so that this will agree with the comparison operators, * particularly as regards comparisons involving NaN. */ if (cmp_numerics(num1, num2) > 0) - PG_RETURN_NUMERIC(num1); + PG_RETURN_DATUM(PG_GETARG_DATUM(0)); else - PG_RETURN_NUMERIC(num2); + PG_RETURN_DATUM(PG_GETARG_DATUM(1)); } @@ -4854,7 +5060,6 @@ do_numeric_accum(NumericAggState *state, Numeric newval) /* The rest of this needs to work in the aggregate context */ old_context = MemoryContextSwitchTo(state->agg_context); - state->N++; /* Accumulate sums */ @@ -4862,7 +5067,6 @@ do_numeric_accum(NumericAggState *state, Numeric newval) if (state->calcSumX2) accum_sum_add(&(state->sumX2), &X2); - MemoryContextSwitchTo(old_context); } @@ -4932,7 +5136,6 @@ do_numeric_discard(NumericAggState *state, Numeric newval) return false; } } - /* if we need X^2, calculate that in short-lived context */ if (state->calcSumX2) { @@ -4942,7 +5145,6 @@ do_numeric_discard(NumericAggState *state, Numeric newval) /* The rest of this needs to work in the aggregate context */ old_context = MemoryContextSwitchTo(state->agg_context); - if (state->N-- > 1) { /* Negate X, to subtract it from the sum */ @@ -4960,14 +5162,11 @@ do_numeric_discard(NumericAggState *state, Numeric newval) { /* Zero the sums */ Assert(state->N == 0); - accum_sum_reset(&state->sumX); if (state->calcSumX2) accum_sum_reset(&state->sumX2); } - MemoryContextSwitchTo(old_context); - return true; } @@ -6689,6 +6888,80 @@ dump_var(const char *str, NumericVar *var) printf("\n"); } + + +static inline int +check_digit_overflow(const NumericVar *var) +{ + int i; + + for (i = 0; i < var->ndigits; i++) + { + if (var->digits[i] >= NBASE) + { + return 1; + } + } + return 0; +} + + +/* + * check_numericvar_state() - + * + * Validate internal NumericVar storage-state invariants. + * + * This is primarily intended as a debugging aid for future changes to the + * NumericVar storage-state machinery. + */ +static inline void +check_numericvar_state(const NumericVar *var, const char *where) +{ + Assert(var != NULL); + Assert(var->ndigits >= 0); + + if (var->state != NUMVAR_BORROWED) + Assert(var->borrow_kind == NUMVAR_BORROW_NONE); + + switch (var->state) + { + case NUMVAR_EMPTY: + Assert(var->buf == NULL); + Assert(var->digits == NULL); + Assert(var->capacity == 0); + break; + + case NUMVAR_BORROWED: + Assert(var->buf == NULL); + Assert(var->capacity == 0); + Assert(var->borrow_kind != NUMVAR_BORROW_NONE); + + if (!NUMERICVAR_IS_SPECIAL(var)) + Assert(var->digits != NULL); + break; + + case NUMVAR_OWNED_INLINE: + Assert(var->inline_buf != NULL); + Assert(var->buf == var->inline_buf); + Assert(var->digits != NULL); + Assert(var->capacity == NUMERIC_INLINE_BUFFER_SIZE); + Assert(var->digits >= var->buf); + Assert(var->digits + var->ndigits <= var->buf + var->capacity); + break; + + case NUMVAR_OWNED_HEAP: + Assert(var->buf != NULL); + Assert(var->digits != NULL); + Assert(var->capacity > 0); + Assert(var->digits >= var->buf); + Assert(var->digits + var->ndigits <= var->buf + var->capacity); + break; + + default: + elog(ERROR, "unexpected NumericVar state %d at %s", + (int) var->state, where); + } +} #endif /* NUMERIC_DEBUG */ @@ -6703,35 +6976,296 @@ dump_var(const char *str, NumericVar *var) * ---------------------------------------------------------------------- */ +/* + * digitbuf_get() - + * + * Ensure that a NumericVar has owned writable digit storage with at least the + * requested physical capacity, including any spare leading digit required by + * the caller. + * + * Regardless of the variable's current state, this function returns the base + * of writable owned storage and leaves the variable in either + * NUMVAR_OWNED_INLINE or NUMVAR_OWNED_HEAP state. + * + * This function is concerned only with storage acquisition and state + * transition. Callers remain responsible for maintaining the logical numeric + * fields and for copying digit contents when required. + */ +static inline NumericDigit * +digitbuf_get(NumericVar *var, int required_capacity) +{ + switch (var->state) + { + case NUMVAR_EMPTY: + Assert(var->borrow_kind == NUMVAR_BORROW_NONE); + if (required_capacity <= NUMERIC_INLINE_BUFFER_SIZE && var->inline_buf) + { + var->buf = var->inline_buf; + var->capacity = NUMERIC_INLINE_BUFFER_SIZE; + var->state = NUMVAR_OWNED_INLINE; + } + else + { + var->buf = digitbuf_alloc_raw(required_capacity); + var->capacity = required_capacity; + var->state = NUMVAR_OWNED_HEAP; + } + break; + case NUMVAR_BORROWED: + if (var->borrow_kind == NUMVAR_BORROW_EXTERNAL) + { + if (required_capacity <= NUMERIC_INLINE_BUFFER_SIZE && var->inline_buf != NULL) + { + var->buf = var->inline_buf; + var->capacity = NUMERIC_INLINE_BUFFER_SIZE; + var->state = NUMVAR_OWNED_INLINE; + } + else + { + var->buf = digitbuf_alloc_raw(required_capacity); + var->capacity = required_capacity; + var->state = NUMVAR_OWNED_HEAP; + } + } + else + { + Assert(var->inline_buf != NULL); + Assert(var->borrow_kind == NUMVAR_BORROW_INLINE); + if (required_capacity <= NUMERIC_INLINE_BUFFER_SIZE) + { + /* + * NUMVAR_BORROW_INLINE is currently used only for the + * special case where the inline buffer of a + * BufferedNumericVar holds a detoasted Numeric datum. + * + * In this case, the caller guarantees that the borrowed + * buffered representation does not need to be preserved + * as an independent read-only view. Therefore, when the + * requested capacity fits in the inline home, we may + * convert that storage to NUMVAR_OWNED_INLINE by + * resetting buf to the beginning of the inline buffer and + * reestablishing the normal NumericVar working-buffer + * layout there. + */ + var->buf = var->inline_buf; + var->capacity = NUMERIC_INLINE_BUFFER_SIZE; + var->state = NUMVAR_OWNED_INLINE; + } + else + { + var->buf = digitbuf_alloc_raw(required_capacity); + var->capacity = required_capacity; + var->state = NUMVAR_OWNED_HEAP; + } + } + break; + case NUMVAR_OWNED_INLINE: + Assert(var->borrow_kind == NUMVAR_BORROW_NONE); + Assert(var->inline_buf != NULL); + if (required_capacity > var->capacity) + { + var->buf = digitbuf_alloc_raw(required_capacity); + var->capacity = required_capacity; + var->state = NUMVAR_OWNED_HEAP; + } + break; + case NUMVAR_OWNED_HEAP: + Assert(var->borrow_kind == NUMVAR_BORROW_NONE); + if (required_capacity > var->capacity) + { + pfree(var->buf); + var->buf = digitbuf_alloc_raw(required_capacity); + var->capacity = required_capacity; + } + break; + default: + Assert(false); + return NULL; /* silence compiler warning */ + } + + var->borrow_kind = NUMVAR_BORROW_NONE; + var->digits = var->buf + 1; + + return var->buf; +} + + +/* + * digitbuf_free() - + * + * Release the current active digit storage or borrowed digit view of a + * NumericVar and return it to NUMVAR_EMPTY storage state. + * + * Heap-backed owned storage is pfree'd. Inline owned storage is simply + * detached as the active buffer and remains available as the variable's + * inline home. Borrowed storage is not freed, but any active borrowed view + * is dropped. + * + * This function resets only storage-state fields. It does not by itself + * establish a meaningful numeric value. + */ +static inline void +digitbuf_free(NumericVar *var) +{ + switch (var->state) + { + case NUMVAR_OWNED_HEAP: + pfree(var->buf); + pg_fallthrough; + case NUMVAR_EMPTY: + case NUMVAR_BORROWED: + case NUMVAR_OWNED_INLINE: + var->buf = NULL; + var->capacity = 0; + var->digits = NULL; + var->state = NUMVAR_EMPTY; + var->borrow_kind = NUMVAR_BORROW_NONE; + break; + default: + Assert(false); + } +} + + +/* + * result_aliases_input() - + * + * Return true if the destination variable cannot be used directly as the + * output of an arithmetic operation because it aliases one of the inputs. + * + * In addition to direct pointer equality, this treats result as aliasing if + * its current digit pointer falls within the active digit range of either + * input variable. + */ +static bool +result_aliases_input(const NumericVar *result, + const NumericVar *a, + const NumericVar *b) +{ + if (result == a || result == b) + return true; + + /* + * Be conservative: if result currently points into either input digit + * range, treat it as aliasing as well. + */ + if (result->digits != NULL) + { + if (a->digits != NULL && + result->digits >= a->digits && + result->digits < a->digits + a->ndigits) + return true; + + if (b->digits != NULL && + result->digits >= b->digits && + result->digits < b->digits + b->ndigits) + return true; + } + + return false; +} + + +/* + * prepare_result_var() - + * + * Prepare a writable result variable for an arithmetic operation. + * + * If the caller's destination does not alias either input, writable storage + * is allocated directly in that destination and the destination is returned. + * + * If the destination aliases an input, a temporary BufferedNumericVar is + * initialized and allocated instead, so that the operation may read its + * inputs without destroying their digit storage. + */ +static NumericVar * +prepare_result_var(NumericVar *result, + const NumericVar *a, + const NumericVar *b, + int res_ndigits, + BufferedNumericVar * tmp) +{ + if (result_aliases_input(result, a, b)) + { + init_buffered_var(tmp); + + alloc_var(&tmp->var, res_ndigits); + + return &tmp->var; + } + + alloc_var(result, res_ndigits); + + return result; +} + + +/* + * finish_result_var() - + * + * Finalize a result prepared by prepare_result_var(). + * + * If the arithmetic result was computed directly in the caller's destination, + * this function does nothing. + * + * If the result was computed in a temporary variable, copy that value back + * into the caller's destination and release the temporary storage. + */ +static void +finish_result_var(NumericVar *work_result, NumericVar *result) +{ + if (work_result != result) + { + set_var_from_var(work_result, result); + free_var(work_result); + } +} + /* * alloc_var() - * - * Allocate a digit buffer of ndigits digits (plus a spare digit for rounding) + * Allocate owned writable digit storage for a NumericVar large enough for the + * requested number of logical digits plus one spare leading digit. + * + * The variable is left in an owned writable state with basic working metadata + * initialized, but callers remain responsible for constructing the final + * numeric value. */ static void alloc_var(NumericVar *var, int ndigits) { - digitbuf_free(var->buf); - var->buf = digitbuf_alloc(ndigits + 1); - var->buf[0] = 0; /* spare digit for rounding */ - var->digits = var->buf + 1; + digitbuf_free(var); + + digitbuf_get(var, ndigits + 1); + var->ndigits = ndigits; + var->weight = 0; + var->sign = NUMERIC_POS; + var->dscale = 0; + var->buf[0] = 0; /* ensure spare digit is zeroed */ } /* * free_var() - * - * Return the digit buffer of a variable to the free pool + * Release the current active storage of a NumericVar and reset its numeric + * metadata to an empty working state. + * + * Unlike digitbuf_free(), this is a variable-level cleanup helper: it clears + * both storage-state and value-level fields. */ static void free_var(NumericVar *var) { - digitbuf_free(var->buf); - var->buf = NULL; - var->digits = NULL; - var->sign = NUMERIC_NAN; + /* Release current active storage, if any. */ + digitbuf_free(var); + + var->ndigits = 0; + var->weight = 0; + var->sign = NUMERIC_POS; + var->dscale = 0; } @@ -6744,7 +7278,7 @@ free_var(NumericVar *var) static void zero_var(NumericVar *var) { - digitbuf_free(var->buf); + free_var(var); var->buf = NULL; var->digits = NULL; var->ndigits = 0; @@ -6923,9 +7457,11 @@ set_var_from_str(const char *str, const char *cp, ndigits = (ddigits + offset + DEC_DIGITS - 1) / DEC_DIGITS; alloc_var(dest, ndigits); + dest->ndigits = ndigits; dest->sign = sign; dest->weight = weight; dest->dscale = dscale; + dest->digits = dest->buf + 1; i = DEC_DIGITS - offset; digits = dest->digits; @@ -7184,6 +7720,8 @@ set_var_from_num(Numeric num, NumericVar *dest) ndigits = NUMERIC_NDIGITS(num); alloc_var(dest, ndigits); + dest->ndigits = ndigits; + dest->digits = dest->buf + 1; dest->weight = NUMERIC_WEIGHT(num); dest->sign = NUMERIC_SIGN(num); @@ -7196,50 +7734,187 @@ set_var_from_num(Numeric num, NumericVar *dest) /* * init_var_from_num() - * - * Initialize a variable from packed db format. The digits array is not - * copied, which saves some cycles when the resulting var is not modified. - * Also, there's no need to call free_var(), as long as you don't assign any - * other value to it (with set_var_* functions, or by using the var as the - * destination of a function like add_var()) + * Initialize a NumericVar from an input Numeric datum, detoasting the datum + * first if necessary. * - * CAUTION: Do not modify the digits buffer of a var initialized with this - * function, e.g by calling round_var() or trunc_var(), as the changes will - * propagate to the original Numeric! It's OK to use it as the destination - * argument of one of the calculational functions, though. + * For finite values, the resulting NumericVar is a borrowed read-only view of + * externally owned digit storage, with borrow kind + * NUMVAR_BORROW_EXTERNAL. + * + * For special values, no digit storage is borrowed; the resulting NumericVar + * remains in NUMVAR_EMPTY storage state, and only sign is semantically + * meaningful. + * + * The returned Numeric pointer is the detoasted representation actually used + * to initialize the borrowed view. */ -static void +static Numeric init_var_from_num(Numeric num, NumericVar *dest) { + /* ensure the numeric is always detoasted */ + num = (Numeric) PG_DETOAST_DATUM(NumericGetDatum(num)); + + init_var(dest); + dest->ndigits = NUMERIC_NDIGITS(num); dest->weight = NUMERIC_WEIGHT(num); dest->sign = NUMERIC_SIGN(num); dest->dscale = NUMERIC_DSCALE(num); - dest->digits = NUMERIC_DIGITS(num); - dest->buf = NULL; /* digits array is not palloc'd */ + + if (NUMERIC_IS_SPECIAL(num)) + { + dest->state = NUMVAR_EMPTY; + dest->borrow_kind = NUMVAR_BORROW_NONE; + } + else + { + dest->state = NUMVAR_BORROWED; + dest->borrow_kind = NUMVAR_BORROW_EXTERNAL; + + dest->digits = NUMERIC_DIGITS(num); + } + + return num; +} + + +/* + * init_buffered_var_from_num() - + * + * Initialize a BufferedNumericVar from an input Numeric datum, using the + * object's inline buffer as detoast workspace when possible. + * + * For finite values, the resulting NumericVar is a borrowed read-only view. + * If the detoasted representation is materialized in the inline buffer, the + * borrow kind is recorded as NUMVAR_BORROW_INLINE; otherwise it is recorded + * as NUMVAR_BORROW_EXTERNAL. + * + * For special values, no digit storage is borrowed; the resulting NumericVar + * remains in NUMVAR_EMPTY storage state, and only sign is semantically + * meaningful. + * + * The returned Numeric pointer is the representation actually used to + * initialize the borrowed view. + */ +static Numeric +init_buffered_var_from_num(Numeric num, BufferedNumericVar * dest) +{ + num = (Numeric) numeric_detoast_attr_to_buffer((varlena *) (num), + (unsigned char *) dest->buffer, + sizeof(dest->buffer)); + + if (NUMERIC_IS_SPECIAL(num)) + { + dest->magic = BUFFERED_NUMERIC_VAR_MAGIC; + init_var(&dest->var); + dest->var.inline_buf = dest->buffer; + dest->var.sign = NUMERIC_SIGN(num); + return num; + } + + dest->magic = BUFFERED_NUMERIC_VAR_MAGIC; + init_var(&dest->var); + dest->var.inline_buf = dest->buffer; + + dest->var.state = NUMVAR_BORROWED; + dest->var.ndigits = NUMERIC_NDIGITS(num); + dest->var.weight = NUMERIC_WEIGHT(num); + dest->var.sign = NUMERIC_SIGN(num); + dest->var.dscale = NUMERIC_DSCALE(num); + dest->var.digits = NUMERIC_DIGITS(num); + + if ((unsigned char *) num == (unsigned char *) dest->buffer) + dest->var.borrow_kind = NUMVAR_BORROW_INLINE; + else + dest->var.borrow_kind = NUMVAR_BORROW_EXTERNAL; + + return num; +} + + +/* + * numeric_detoast_attr_to_buffer() - + * + * Convert a varlena datum to a detoasted representation, preferably using a + * caller-provided buffer when that is sufficient. + * + * This helper is intended for numeric paths. If the input datum is in + * short-header form, it is expanded to ordinary 4-byte-header form, using + * the supplied buffer when possible and falling back to palloc'd storage + * otherwise. + * + * If the input datum is not in short-header form, this helper falls back to + * detoast_attr(). + * + * The returned pointer may therefore refer either to the caller-provided + * buffer or to freshly allocated storage. + */ +static varlena * +numeric_detoast_attr_to_buffer(varlena *attr, unsigned char *buffer, + size_t buffer_size) +{ + Size data_size; + Size new_size; + varlena *new_attr; + + if (!VARATT_IS_SHORT(attr)) + return pg_detoast_datum(attr); + + /* + * This is a short-header varlena --- convert to 4-byte header format + */ + data_size = VARSIZE_SHORT(attr) - VARHDRSZ_SHORT; + new_size = data_size + VARHDRSZ; + + if (unlikely(buffer_size < new_size)) + new_attr = (varlena *) palloc(new_size); + else + new_attr = (varlena *) buffer; + SET_VARSIZE(new_attr, new_size); + memcpy(VARDATA(new_attr), VARDATA_SHORT(attr), data_size); + attr = new_attr; + + return attr; } /* * set_var_from_var() - * - * Copy one variable into another + * Copy one NumericVar into another. + * + * The destination receives the same numeric value as the source, but this + * function always constructs a destination representation that is valid for + * the destination's own storage state. It does not steal storage from the + * source. + * + * This function must remain safe when source and destination are the same + * variable. */ static void set_var_from_var(const NumericVar *value, NumericVar *dest) { - NumericDigit *newbuf; + if (value == dest) + return; - newbuf = digitbuf_alloc(value->ndigits + 1); - newbuf[0] = 0; /* spare digit for rounding */ - if (value->ndigits > 0) /* else value->digits might be null */ - memcpy(newbuf + 1, value->digits, - value->ndigits * sizeof(NumericDigit)); + if (NUMERICVAR_IS_SPECIAL(value)) + { + free_var(dest); + dest->sign = value->sign; + + return; + } - digitbuf_free(dest->buf); + /* ndigits is set in alloc_var */ + alloc_var(dest, value->ndigits); - memmove(dest, value, sizeof(NumericVar)); - dest->buf = newbuf; - dest->digits = newbuf + 1; + dest->weight = value->weight; + dest->dscale = value->dscale; + dest->sign = value->sign; + + memcpy(dest->digits, + value->digits, + value->ndigits * sizeof(NumericDigit)); } @@ -8600,6 +9275,7 @@ mul_var(const NumericVar *var1, const NumericVar *var2, NumericVar *result, */ result->weight = res_weight; result->sign = res_sign; + result->ndigits = res_ndigits; /* Round to target rscale (and set result->dscale) */ round_var(result, rscale); @@ -8631,6 +9307,8 @@ mul_var_short(const NumericVar *var1, const NumericVar *var2, NumericDigit *res_digits; uint32 carry = 0; uint32 term; + BufferedNumericVar tmp; + NumericVar *res; /* Check preconditions */ Assert(var1ndigits >= 1); @@ -8651,8 +9329,9 @@ mul_var_short(const NumericVar *var1, const NumericVar *var2, res_weight = var1->weight + var2->weight + 1; res_ndigits = var1ndigits + var2ndigits; - /* Allocate result digit array */ - res_buf = digitbuf_alloc(res_ndigits + 1); + res = prepare_result_var(result, var1, var2, res_ndigits, &tmp); + + res_buf = res->buf; res_buf[0] = 0; /* spare digit for later rounding */ res_digits = res_buf + 1; @@ -8872,16 +9551,17 @@ mul_var_short(const NumericVar *var1, const NumericVar *var2, } /* Store the product in result */ - digitbuf_free(result->buf); - result->ndigits = res_ndigits; - result->buf = res_buf; - result->digits = res_digits; - result->weight = res_weight; - result->sign = res_sign; - result->dscale = var1->dscale + var2->dscale; + + res->ndigits = res_ndigits; + res->digits = res_digits; + res->weight = res_weight; + res->sign = res_sign; + res->dscale = var1->dscale + var2->dscale; /* Strip leading and trailing zeroes */ - strip_var(result); + strip_var(res); + + finish_result_var(res, result); } @@ -9457,6 +10137,8 @@ div_var_int(const NumericVar *var, int ival, int ival_weight, NumericDigit *res_digits; uint32 divisor; int i; + BufferedNumericVar tmp; + NumericVar *res; /* Guard against division by zero */ if (ival == 0) @@ -9490,7 +10172,9 @@ div_var_int(const NumericVar *var, int ival, int ival_weight, if (round) res_ndigits++; - res_buf = digitbuf_alloc(res_ndigits + 1); + res = prepare_result_var(result, var, var, res_ndigits, &tmp); + + res_buf = res->buf; res_buf[0] = 0; /* spare digit for later rounding */ res_digits = res_buf + 1; @@ -9532,12 +10216,14 @@ div_var_int(const NumericVar *var, int ival, int ival_weight, } /* Store the quotient in result */ - digitbuf_free(result->buf); - result->ndigits = res_ndigits; - result->buf = res_buf; - result->digits = res_digits; - result->weight = res_weight; - result->sign = res_sign; + + res->ndigits = res_ndigits; + res->digits = res_digits; + res->weight = res_weight; + res->sign = res_sign; + res->dscale = rscale; + + finish_result_var(res, result); /* Round or truncate to target rscale (and set result->dscale) */ if (round) @@ -9573,6 +10259,8 @@ div_var_int64(const NumericVar *var, int64 ival, int ival_weight, NumericDigit *res_digits; uint64 divisor; int i; + BufferedNumericVar tmp; + NumericVar *res; /* Guard against division by zero */ if (ival == 0) @@ -9606,7 +10294,9 @@ div_var_int64(const NumericVar *var, int64 ival, int ival_weight, if (round) res_ndigits++; - res_buf = digitbuf_alloc(res_ndigits + 1); + res = prepare_result_var(result, var, var, res_ndigits, &tmp); + + res_buf = res->buf; res_buf[0] = 0; /* spare digit for later rounding */ res_digits = res_buf + 1; @@ -9648,12 +10338,13 @@ div_var_int64(const NumericVar *var, int64 ival, int ival_weight, } /* Store the quotient in result */ - digitbuf_free(result->buf); - result->ndigits = res_ndigits; - result->buf = res_buf; - result->digits = res_digits; - result->weight = res_weight; - result->sign = res_sign; + res->ndigits = res_ndigits; + res->digits = res_digits; + res->weight = res_weight; + res->sign = res_sign; + res->dscale = rscale; + + finish_result_var(res, result); /* Round or truncate to target rscale (and set result->dscale) */ if (round) @@ -11494,6 +12185,8 @@ add_abs(const NumericVar *var1, const NumericVar *var2, NumericVar *result) i1, i2; int carry = 0; + BufferedNumericVar tmp; + NumericVar *res; /* copy these values into local vars for speed in inner loop */ int var1ndigits = var1->ndigits; @@ -11514,7 +12207,9 @@ add_abs(const NumericVar *var1, const NumericVar *var2, NumericVar *result) if (res_ndigits <= 0) res_ndigits = 1; - res_buf = digitbuf_alloc(res_ndigits + 1); + res = prepare_result_var(result, var1, var2, res_ndigits, &tmp); + + res_buf = res->buf; res_buf[0] = 0; /* spare digit for later rounding */ res_digits = res_buf + 1; @@ -11543,15 +12238,15 @@ add_abs(const NumericVar *var1, const NumericVar *var2, NumericVar *result) Assert(carry == 0); /* else we failed to allow for carry out */ - digitbuf_free(result->buf); - result->ndigits = res_ndigits; - result->buf = res_buf; - result->digits = res_digits; - result->weight = res_weight; - result->dscale = res_dscale; + res->ndigits = res_ndigits; + res->digits = res_digits; + res->weight = res_weight; + res->dscale = res_dscale; /* Remove leading/trailing zeroes */ - strip_var(result); + strip_var(res); + + finish_result_var(res, result); } @@ -11579,6 +12274,8 @@ sub_abs(const NumericVar *var1, const NumericVar *var2, NumericVar *result) i1, i2; int borrow = 0; + BufferedNumericVar tmp; + NumericVar *res; /* copy these values into local vars for speed in inner loop */ int var1ndigits = var1->ndigits; @@ -11599,7 +12296,9 @@ sub_abs(const NumericVar *var1, const NumericVar *var2, NumericVar *result) if (res_ndigits <= 0) res_ndigits = 1; - res_buf = digitbuf_alloc(res_ndigits + 1); + res = prepare_result_var(result, var1, var2, res_ndigits, &tmp); + + res_buf = res->buf; res_buf[0] = 0; /* spare digit for later rounding */ res_digits = res_buf + 1; @@ -11628,15 +12327,15 @@ sub_abs(const NumericVar *var1, const NumericVar *var2, NumericVar *result) Assert(borrow == 0); /* else caller gave us var1 < var2 */ - digitbuf_free(result->buf); - result->ndigits = res_ndigits; - result->buf = res_buf; - result->digits = res_digits; - result->weight = res_weight; - result->dscale = res_dscale; + res->ndigits = res_ndigits; + res->digits = res_digits; + res->weight = res_weight; + res->dscale = res_dscale; /* Remove leading/trailing zeroes */ - strip_var(result); + strip_var(res); + + finish_result_var(res, result); } /* @@ -12099,7 +12798,9 @@ accum_sum_final(NumericSumAccum *accum, NumericVar *result) /* Create NumericVars representing the positive and negative sums */ init_var(&pos_var); + alloc_var(&pos_var, accum->ndigits); init_var(&neg_var); + alloc_var(&neg_var, accum->ndigits); pos_var.ndigits = neg_var.ndigits = accum->ndigits; pos_var.weight = neg_var.weight = accum->weight; @@ -12107,8 +12808,8 @@ accum_sum_final(NumericSumAccum *accum, NumericVar *result) pos_var.sign = NUMERIC_POS; neg_var.sign = NUMERIC_NEG; - pos_var.buf = pos_var.digits = digitbuf_alloc(accum->ndigits); - neg_var.buf = neg_var.digits = digitbuf_alloc(accum->ndigits); + pos_var.buf = pos_var.digits = digitbuf_alloc_raw(accum->ndigits); + neg_var.buf = neg_var.digits = digitbuf_alloc_raw(accum->ndigits); for (i = 0; i < accum->ndigits; i++) { -- 2.43.0