updated code generation for compound-columns, initial support for scratchspace

engine/types.py

@@ -305,7 +305,7 @@ opor = OperatorBase('or', 2, logical, cname = '||', sqlname = ' OR ', call = bin
 opxor = OperatorBase('xor', 2, logical, cname = '^', sqlname = ' XOR ', call = binary_op_behavior)
 opgt = OperatorBase('gt', 2, logical, cname = '>', sqlname = '>', call = binary_op_behavior)
 oplt = OperatorBase('lt', 2, logical, cname = '<', sqlname = '<', call = binary_op_behavior)
-opge = OperatorBase('gte', 2, logical, cname = '>=', sqlname = '>=', call = binary_op_behavior)
+opgte = OperatorBase('gte', 2, logical, cname = '>=', sqlname = '>=', call = binary_op_behavior)
 oplte = OperatorBase('lte', 2, logical, cname = '<=', sqlname = '<=', call = binary_op_behavior)
 opneq = OperatorBase('neq', 2, logical, cname = '!=', sqlname = '!=', call = binary_op_behavior)
 opeq = OperatorBase('eq', 2, logical, cname = '==', sqlname = '=', call = binary_op_behavior)
@@ -355,19 +355,27 @@ fnpow = OperatorBase('pow', 2, lambda *_ : DoubleT, cname = 'pow', sqlname = 'PO
 # type collections
 def _op_make_dict(*items : OperatorBase):
     return { i.name: i for i in items}
+#binary op
 builtin_binary_arith = _op_make_dict(opadd, opdiv, opmul, opsub, opmod)
 builtin_binary_logical = _op_make_dict(opand, opor, opxor, opgt, oplt, 
-                                       opge, oplte, opneq, opeq)
+                                       opgte, oplte, opneq, opeq)
+builtin_binary_ops = {**builtin_binary_arith, **builtin_binary_logical}
+#unary op
 builtin_unary_logical = _op_make_dict(opnot)
 builtin_unary_arith = _op_make_dict(opneg)
 builtin_unary_special = _op_make_dict(spnull, opdistinct)
+# functions 
 builtin_cstdlib = _op_make_dict(fnsqrt, fnlog, fnsin, fncos, fntan, fnpow)
-builtin_func = _op_make_dict(fnmax, fnmin, fnsum, fnavg, fnmaxs, 
-                             fnmins, fndeltas, fnratios, fnlast,
-                             fnfirst, fnsums, fnavgs, fncnt, 
-                             fnpack, fntrunc, fnprev, fnnext, 
-                             fnvar, fnvars, fnstd, fnstds)
+builtin_aggfunc = _op_make_dict(fnmax, fnmin, fnsum, fnavg, 
+                                fnlast, fnfirst, fncnt, fnvar, fnstd)
+builtin_vecfunc = _op_make_dict(fnmaxs, 
+                             fnmins, fndeltas, fnratios, fnsums, fnavgs, 
+                             fnpack, fntrunc, fnprev, fnnext, fnvars, fnstds)
+builtin_vecfunc = {**builtin_vecfunc, **builtin_cstdlib}
+builtin_func = {**builtin_vecfunc, **builtin_aggfunc}
+
 user_module_func = {}
+
 builtin_operators : Dict[str, OperatorBase] = {**builtin_binary_arith, **builtin_binary_logical, 
     **builtin_unary_arith, **builtin_unary_logical, **builtin_unary_special, **builtin_func, **builtin_cstdlib, 
     **user_module_func}

reconstruct/ast.py

@@ -604,9 +604,10 @@ class groupby_c(ast_node):
         self.arr_len = 'arrlen_' + base62uuid(3)
         self.arr_values = 'arrvals_' + base62uuid(3)
         
+        self.context.emitc(f'auto {self.arr_len} = {self.group}.size();')
+        self.context.emitc(f'auto {self.arr_values} = {self.group}.values();')
+        
         if len(tovec_columns):
-            self.context.emitc(f'auto {self.arr_len} = {self.group}.size();')
-            self.context.emitc(f'auto {self.arr_values} = {self.group}.values();')
             preproc_scanner = scan(self, self.arr_len)
             preproc_scanner_it = preproc_scanner.it_var
             for c in tovec_columns:
@@ -674,11 +675,8 @@ class groupby_c(ast_node):
                     gscanner.add(f'{ex.eval(c_code = True, y=get_var_names, materialize_builtin = materialize_builtin)};\n')
                     continue
             if col_tovec[i]:
-                if ex.opname == 'avgs':
-                    patch_expr = get_var_names_ex(ex)
-                    patch_expr = patch_expr[:patch_expr.rindex(')')]
-                    patch_expr += ', ' + f'{ce[0]}[{gscanner.it_var}]' + ')'
-                    gscanner.add(f'{patch_expr};\n')
+                if ex.remake_binary(f'{ce[0]}[{gscanner.it_var}]'):
+                    gscanner.add(f'{get_var_names_ex(ex)};\n')
                 else:
                     gscanner.add(f'{ce[0]}[{gscanner.it_var}] = {get_var_names_ex(ex)};\n')
             else:

reconstruct/expr.py

@@ -367,6 +367,19 @@ class expr(ast_node):
             self.curr_code += c.codegen(delegate)
         return self.curr_code
     
+    def remake_binary(self, ret_expr):
+        if self.root:
+            self.oldsql = self.sql
+            if (self.opname in builtin_binary_ops):
+                patched_opname = 'aqop_' + self.opname
+                self.sql = (f'{patched_opname}({self.children[0].sql}, '
+                            f'{self.children[1].sql}, {ret_expr})')
+                return True
+            elif self.opname in builtin_vecfunc:
+                self.sql = self.sql[:self.sql.rindex(')')]
+                self.sql += ', ' + ret_expr + ')'
+                return True
+        return False
     def __str__(self):
         return self.sql
     def __repr__(self):

server/aggregations.h

@@ -1,5 +1,6 @@
 #pragma once
 #include "types.h"
+#include "gc.h"
 #include <utility>
 #include <limits>
 #include <deque>
@@ -29,14 +30,19 @@ double avg(const VT<T>& v) {
 	return (sum<T>(v) / static_cast<double>(v.size));
 }
 
+template<class T, template<typename ...> class VT, class Ret>
+void sqrt(const VT<T>& v, Ret& ret) {
+	for (uint32_t i = 0; i < v.size; ++i) 
+		ret[i] = sqrt(v[i]);
+}
+
 template<class T, template<typename ...> class VT>
 VT<double> sqrt(const VT<T>& v) {
 	VT<double> ret(v.size);
-	for (uint32_t i = 0; i < v.size; ++i) {
-		ret[i] = sqrt(v[i]);
-	}
+	sqrt(v, ret);
 	return ret;
 }
+
 template <class T>
 T truncate(const T& v, const uint32_t precision) {
 	auto multiplier = pow(10, precision);
@@ -73,109 +79,153 @@ T min(const VT<T>& v) {
 		min_v = min_v < _v ? min_v : _v;
 	return min_v;
 }
-template<class T, template<typename ...> class VT>
-decayed_t<VT, T> mins(const VT<T>& arr) {
+
+// simplify this using a template std::binary_function<T, T, bool> = std::less;
+template<class T, template<typename ...> class VT, class Ret>
+void mins(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	std::deque<std::pair<T, uint32_t>> cache;
-	decayed_t<VT, T> ret(len);
 	T min = std::numeric_limits<T>::max();
 	for (int i = 0; i < len; ++i) {
 		if (arr[i] < min)
 			min = arr[i];
 		ret[i] = min;
 	}
+}
+
+template<class T, template<typename ...> class VT>
+decayed_t<VT, T> mins(const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	mins(arr, ret);
 	return ret;
 }
-template<class T, template<typename ...> class VT>
-decayed_t<VT, T> maxs(const VT<T>& arr) {
+
+template<class T, template<typename ...> class VT, class Ret>
+void maxs(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, T> ret(len);
 	T max = std::numeric_limits<T>::min();
 	for (int i = 0; i < len; ++i) {
 		if (arr[i] > max)
 			max = arr[i];
 		ret[i] = max;
 	}
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, T> minw(uint32_t w, const VT<T>& arr) {
+decayed_t<VT, T> maxs(const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	maxs(arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void minw(uint32_t w, const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, T> ret(len);
 	std::deque<std::pair<T, uint32_t>> cache;
 	for (int i = 0; i < len; ++i) {
 		if (!cache.empty() && cache.front().second == i - w) cache.pop_front();
+		
 		while (!cache.empty() && cache.back().first > arr[i]) cache.pop_back();
 		cache.push_back({ arr[i], i });
 		ret[i] = cache.front().first;
 	}
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, T> maxw(uint32_t w, const VT<T>& arr) {
+decayed_t<VT, T> minw(uint32_t w, const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	minw(w, arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void maxw(uint32_t w, const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, T> ret(len);
 	std::deque<std::pair<T, uint32_t>> cache;
 	for (int i = 0; i < len; ++i) {
 		if (!cache.empty() && cache.front().second == i - w) cache.pop_front();
-		while (!cache.empty() && cache.back().first > arr[i]) cache.pop_back();
+		while (!cache.empty() && cache.back().first < arr[i]) cache.pop_back();
 		cache.push_back({ arr[i], i });
-		arr[i] = cache.front().first;
+		ret[i] = cache.front().first;
 	}
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetFPType<T>> ratiow(uint32_t w, const VT<T>& arr) {
+inline decayed_t<VT, T> maxw(uint32_t w, const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	maxw(w, arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void ratiow(uint32_t w, const VT<T>& arr, Ret& ret) {
 	typedef std::decay_t<types::GetFPType<T>> FPType;
 	uint32_t len = arr.size;
 	if (arr.size <= w) 
 		len = 1;
 	w = w > len ? len : w;
-	decayed_t<VT, FPType> ret(arr.size);
 	ret[0] = 0;
 	for (uint32_t i = 0; i < w; ++i) 
 		ret[i] = arr[i] / (FPType)arr[0];
 	for (uint32_t i = w; i < arr.size; ++i) 
 		ret[i] = arr[i] / (FPType) arr[i - w];
+}
+
+template<class T, template<typename ...> class VT>
+inline decayed_t<VT, types::GetFPType<T>> ratiow(uint32_t w, const VT<T>& arr) {
+	typedef std::decay_t<types::GetFPType<T>> FPType;
+	decayed_t<VT, FPType> ret(arr.size);
+	ratiow(w, arr, ret);
 	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetFPType<T>> ratios(const VT<T>& arr) { 
+inline decayed_t<VT, types::GetFPType<T>> ratios(const VT<T>& arr) { 
 	return ratiow(1, arr);
 }
 
-template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetLongType<T>> sums(const VT<T>& arr) {
+template<class T, template<typename ...> class VT, class Ret>
+inline void ratios(const VT<T>& arr, Ret& ret) { 
+	return ratiow(1, arr, ret);
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void sums(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, types::GetLongType<T>> ret(len);
 	uint32_t i = 0;
 	if (len) ret[i++] = arr[0];
 	for (; i < len; ++i)
 		ret[i] = ret[i - 1] + arr[i];
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetFPType<types::GetLongType<T>>> avgs(const VT<T>& arr) {
+inline decayed_t<VT, types::GetLongType<T>> sums(const VT<T>& arr) {
+	decayed_t<VT, types::GetLongType<T>> ret(arr.size);
+	sums(arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void avgs(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
 	typedef types::GetFPType<types::GetLongType<T>> FPType;
-	decayed_t<VT, FPType> ret(len);
 	uint32_t i = 0;
 	types::GetLongType<T> s;
 	if (len) s = ret[i++] = arr[0];
 	for (; i < len; ++i)
 		ret[i] = (s += arr[i]) / (FPType)(i + 1);
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetLongType<T>> sumw(uint32_t w, const VT<T>& arr) {
+inline decayed_t<VT, types::GetFPType<types::GetLongType<T>>> avgs(const VT<T>& arr) {
+	typedef types::GetFPType<types::GetLongType<T>> FPType;
+	decayed_t<VT, FPType> ret(arr.size);
+	avgs(arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void sumw(uint32_t w, const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, types::GetLongType<T>> ret(len);
 	uint32_t i = 0;
 	w = w > len ? len : w;
 	if (len) ret[i++] = arr[0];
@@ -183,11 +233,17 @@ decayed_t<VT, types::GetLongType<T>> sumw(uint32_t w, const VT<T>& arr) {
 		ret[i] = ret[i - 1] + arr[i];
 	for (; i < len; ++i)
 		ret[i] = ret[i - 1] + arr[i] - arr[i - w];
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-void avgw(uint32_t w, const VT<T>& arr, decayed_t<vector_type, types::GetFPType<types::GetLongType<T>>>& ret) {
+decayed_t<VT, types::GetLongType<T>> sumw(uint32_t w, const VT<T>& arr) {
+	decayed_t<VT, types::GetLongType<T>> ret(arr.size);
+	sumw(w, arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void avgw(uint32_t w, const VT<T>& arr, Ret& ret) {
 	typedef types::GetFPType<types::GetLongType<T>> FPType;
 	const uint32_t& len = arr.size;
 	uint32_t i = 0;
@@ -201,26 +257,19 @@ void avgw(uint32_t w, const VT<T>& arr, decayed_t<vector_type, types::GetFPType<
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetFPType<types::GetLongType<T>>> avgw(uint32_t w, const VT<T>& arr) {
+inline decayed_t<VT, types::GetFPType<types::GetLongType<T>>> avgw(uint32_t w, const VT<T>& arr) {
 	typedef types::GetFPType<types::GetLongType<T>> FPType;
 	const uint32_t& len = arr.size;
 	decayed_t<VT, FPType> ret(len);
-	uint32_t i = 0;
-	types::GetLongType<T> s{};
-	w = w > len ? len : w;
-	if (len) s = ret[i++] = arr[0];
-	for (; i < w; ++i)
-		ret[i] = (s += arr[i]) / (FPType)(i + 1);
-	for (; i < len; ++i)
-		ret[i] = ret[i - 1] + (arr[i] - arr[i - w]) / (FPType)w;
+	avgw(w, arr, ret);
 	return ret;
 }
 
-template<class T, template<typename ...> class VT, bool sd = false>
-decayed_t<VT, types::GetFPType<types::GetLongType<T>>> varw(uint32_t w, const VT<T>& arr) {
+template<class T, template<typename ...> class VT, class Ret, bool sd = false>
+void varw(uint32_t w, const VT<T>& arr, 
+	Ret& ret) {
 	using FPType = types::GetFPType<types::GetLongType<T>>;
 	const uint32_t& len = arr.size;
-	decayed_t<VT, FPType> ret(len);
 	uint32_t i = 0;
 	types::GetLongType<T> s{};
 	w = w > len ? len : w;
@@ -252,7 +301,14 @@ decayed_t<VT, types::GetFPType<types::GetLongType<T>>> varw(uint32_t w, const VT
 	if constexpr(sd) 
 		if(i)
 			ret[i-1] = sqrt(ret[i-1]);
-	
+}
+
+
+template<class T, template<typename ...> class VT, bool sd = false>
+inline decayed_t<VT, types::GetFPType<types::GetLongType<T>>> varw(uint32_t w, const VT<T>& arr) {
+	using FPType = types::GetFPType<types::GetLongType<T>>;
+	decayed_t<VT, FPType> ret(arr.size);
+	varw<T, VT, decayed_t<VT, types::GetFPType<types::GetLongType<T>>>, sd>(w, arr, ret);
 	return ret;
 }
 
@@ -274,11 +330,10 @@ types::GetFPType<types::GetLongType<decays<T>>> var(const VT<T>& arr) {
 	return (ssq - s * s / (FPType)(len + 1)) / (FPType)(len + 1);
 }
 
-template<class T, template<typename ...> class VT, bool sd = false>
-decayed_t<VT, types::GetFPType<types::GetLongType<T>>> vars(const VT<T>& arr) {
+template<class T, template<typename ...> class VT, class Ret, bool sd = false>
+void vars(const VT<T>& arr, Ret& ret) {
 	typedef types::GetFPType<types::GetLongType<T>> FPType;
 	const uint32_t& len = arr.size;
-	decayed_t<VT, FPType> ret(len);
 	uint32_t i = 0;
 	types::GetLongType<T> s{};
 	FPType MnX{};
@@ -298,51 +353,88 @@ decayed_t<VT, types::GetFPType<types::GetLongType<T>>> vars(const VT<T>& arr) {
 		ret[i] = MnX / (FPType)(i + 1);
 		if constexpr(sd) ret[i] = sqrt(ret[i]);
 	}
+}
+
+template<class T, template<typename ...> class VT, bool sd = false>
+inline decayed_t<VT, types::GetFPType<types::GetLongType<T>>> vars(const VT<T>& arr) {
+	typedef types::GetFPType<types::GetLongType<T>> FPType;
+	decayed_t<VT, FPType> ret(arr.size);
+	vars<T, VT, decayed_t<VT, types::GetFPType<types::GetLongType<T>>>, sd>(arr, ret);
 	return ret;
 }
+
 template<class T, template<typename ...> class VT>
-types::GetFPType<types::GetLongType<decays<T>>> stddev(const VT<T>& arr) {
+inline types::GetFPType<types::GetLongType<decays<T>>> stddev(const VT<T>& arr) {
 	return sqrt(var(arr));
 }
+
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetFPType<types::GetLongType<T>>> stddevs(const VT<T>& arr) {
+inline decayed_t<VT, types::GetFPType<types::GetLongType<T>>> stddevs(const VT<T>& arr) {
 	return vars<T, VT, true>(arr);
 }
+
 template<class T, template<typename ...> class VT>
-decayed_t<VT, types::GetFPType<types::GetLongType<T>>> stddevw(uint32_t w, const VT<T>& arr) {
+inline decayed_t<VT, types::GetFPType<types::GetLongType<T>>> stddevw(uint32_t w, const VT<T>& arr) {
 	return varw<T, VT, true>(w, arr);
 }
+
+template<class T, template<typename ...> class VT, class Ret>
+inline auto stddevs(const VT<T>& arr, Ret& ret) {
+	return vars<T, VT, Ret, true>(arr, ret);
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+inline auto stddevw(uint32_t w, const VT<T>& arr, Ret& ret) {
+	return varw<T, VT, Ret, true>(w, arr, ret);
+}
+
+
 // use getSignedType
-template<class T, template<typename ...> class VT>
-decayed_t<VT, T> deltas(const VT<T>& arr) {
+template<class T, template<typename ...> class VT, class Ret>
+void deltas(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, T> ret(len);
 	uint32_t i = 0;
 	if (len) ret[i++] = 0;
 	for (; i < len; ++i)
 		ret[i] = arr[i] - arr[i - 1];
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, T> prev(const VT<T>& arr) {
+inline decayed_t<VT, T> deltas(const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	deltas(arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void prev(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, T> ret(len);
 	uint32_t i = 0;
 	if (len) ret[i++] = arr[0];
 	for (; i < len; ++i)
 		ret[i] = arr[i - 1];
-	return ret;
 }
 
 template<class T, template<typename ...> class VT>
-decayed_t<VT, T> aggnext(const VT<T>& arr) {
+inline decayed_t<VT, T> prev(const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	prev(arr, ret);
+	return ret;
+}
+
+template<class T, template<typename ...> class VT, class Ret>
+void aggnext(const VT<T>& arr, Ret& ret) {
 	const uint32_t& len = arr.size;
-	decayed_t<VT, T> ret(len);
 	uint32_t i = 1;
 	for (; i < len; ++i)
 		ret[i - 1] = arr[i];
 	if (len > 0) ret[len - 1] = arr[len - 1];
+}
+
+template<class T, template<typename ...> class VT>
+inline decayed_t<VT, T> aggnext(const VT<T>& arr) {
+	decayed_t<VT, T> ret(arr.size);
+	aggnext(arr, ret);
 	return ret;
 }
 
@@ -360,8 +452,6 @@ T first(const VT<T>& arr) {
 	return arr[0];
 }
 
-
-
 #define __DEFAULT_AGGREGATE_FUNCTION__(NAME, RET) \
 template <class T> constexpr T NAME(const T& v) { return RET; }
 

server/gc.h

@@ -1,4 +1,39 @@
 #pragma once
+
+class ScratchSpace {
+public:
+	void* ret;
+	char* scratchspace;
+	size_t ptr;
+	size_t cnt;
+	size_t capacity;
+	size_t initial_capacity;
+	void* temp_memory_fractions;
+
+	//uint8_t status; 
+	// record maximum size
+	constexpr static uint8_t Grow = 0x1;
+	// no worry about overflow
+	constexpr static uint8_t Use = 0x0; 
+
+	void init(size_t initial_capacity);
+
+	// apply for memory
+	void* alloc(uint32_t sz);
+
+	void register_ret(void* ret);
+
+	// reorganize memory space
+	void release();
+
+	// reset status of the scratch space 
+	void reset();
+
+	// reset scratch space to initial capacity.
+	void cleanup();
+};
+
+
 #ifndef __AQ_USE_THREADEDGC__
 #include <atomic>
 class GC {
@@ -18,7 +53,7 @@ private:;
 	std::atomic<uint64_t> current_size;
 	volatile bool lock;
 	using gc_deallocator_t = void (*)(void*);
-
+	ScratchSpace scratch;
 	// maybe use volatile std::thread::id instead
 protected:
 	void acquire_lock();
@@ -29,28 +64,36 @@ protected:
 	void terminate_daemon();
 
 public:
-	void reg(void* v, uint32_t sz = 1, 
+	void reg(void* v, uint32_t sz = 0xffffffff, 
 			void(*f)(void*) = free
 		);
 
+	uint32_t get_threshold() const {
+		return threshould;
+	}
+
 	GC(
 		uint64_t max_size = 0xfffffff, uint32_t max_slots = 4096, 
 		uint32_t interval = 10000, uint32_t forced_clean = 1000000,
-		uint32_t threshould = 64 //one seconds
+		uint32_t threshould = 64, //one seconds
+		uint32_t scratch_sz = 0x1000000 // 16 MB
 	) : max_size(max_size), max_slots(max_slots), 
 		interval(interval), forced_clean(forced_clean), 
 		threshould(threshould) {
 
 		start_deamon();
 		GC::gc_handle = this;
+		this->scratch.init(scratch_sz);
 	} // 256 MB
 
 	~GC(){
 		terminate_daemon();
+		scratch.cleanup();
 	}
+
 	static GC* gc_handle;
 	template <class T>
-	constexpr static inline gc_deallocator_t _delete(T*){
+	static inline gc_deallocator_t _delete(T*) {
 		return [](void* v){
 			delete (T*)v;
 		};

server/libaquery.cpp

@@ -452,6 +452,9 @@ void GC::reg(void* v, uint32_t sz, void(*f)(void*)) { //~ 40ns expected v. free
 		f(v);
 		return;
 	}
+	else if (sz == 0xffffffff)
+		sz = this->threshould;
+	
 	auto _q = static_cast<memoryqueue_t>(q);
 	while(lock);
 	++alive_cnt;
@@ -466,6 +469,71 @@ void GC::reg(void* v, uint32_t sz, void(*f)(void*)) { //~ 40ns expected v. free
 
 inline GC* GC::gc_handle = nullptr;
 
+void ScratchSpace::init(size_t initial_capacity) {
+	ret = nullptr;
+	scratchspace = static_cast<char*>(malloc(initial_capacity));
+	ptr = cnt = 0;
+	capacity = initial_capacity;
+	this->initial_capacity = initial_capacity;
+	temp_memory_fractions = new vector_type<void*>();
+}
+
+inline void* ScratchSpace::alloc(uint32_t sz){
+	this->cnt += sz; // major cost
+    ptr = this->cnt;
+	if (this->cnt > capacity) {
+		[[unlikely]] 
+		capacity = this->cnt + (capacity >> 1);
+		auto vec_tmpmem_fractions = static_cast<vector_type<char *>*>(temp_memory_fractions);
+		vec_tmpmem_fractions->emplace_back(scratchspace);
+		scratchspace = static_cast<char*>(malloc(capacity));
+		ptr = 0;
+	}
+	return scratchspace + ptr;
+}
+
+inline void ScratchSpace::register_ret(void* ret){
+	this->ret = ret;
+}
+
+inline void ScratchSpace::release(){
+	ptr = cnt = 0;
+	ret = nullptr;
+	auto vec_tmpmem_fractions = 
+		static_cast<vector_type<void*>*>(temp_memory_fractions);
+	if (vec_tmpmem_fractions->size) {
+    //[[unlikely]]
+		for(auto& mem : *vec_tmpmem_fractions){
+			 free(mem);
+			//GC::gc_handle->reg(mem);
+		}
+		vec_tmpmem_fractions->clear();
+	}
+}
+
+inline void ScratchSpace::reset() {
+	this->release();
+	if (capacity != initial_capacity){
+		capacity = initial_capacity;
+		scratchspace = static_cast<char*>(realloc(scratchspace, capacity));
+	}
+}
+
+void ScratchSpace::cleanup(){
+	auto vec_tmpmem_fractions = 
+		static_cast<vector_type<void*>*>(temp_memory_fractions);
+	if (vec_tmpmem_fractions->size) {
+		for(auto& mem : *vec_tmpmem_fractions){
+			free(mem); 
+			//GC::gc_handle->reg(mem);
+		}
+		vec_tmpmem_fractions->clear();
+	}
+	delete vec_tmpmem_fractions;
+	free(this->scratchspace);
+}
+
+
 #include "dragonbox/dragonbox_to_chars.hpp" 
 
 

server/table.h

@@ -36,7 +36,8 @@ struct ColRef_cstorage {
 	int ty; // what if enum is not int?
 };
 
-template <template <class...> class VT, class T>
+template <template <class...> class VT, class T, 
+	std::enable_if_t<std::is_base_of_v<vector_base<T>, VT<T>>>* = nullptr>
 std::ostream& operator<<(std::ostream& os, const VT<T>& v)
 {
 	v.out();
@@ -143,7 +144,7 @@ public:
 		vector_type<_Ty>::operator=(vt);
 		return *this;
 	}
-	ColRef<_Ty>& operator =(ColRef<_Ty>&& vt) {
+	ColRef<_Ty>& operator =(ColRef<_Ty>&& vt) noexcept {
 		vector_type<_Ty>::operator=(std::move(vt));
 		return *this;
 	
@@ -475,7 +476,7 @@ struct TableInfo {
 
 		std::string printf_string =
 			generate_printf_string<typename std::tuple_element<cols, tuple_type>::type ...>(sep, end);
-		puts(printf_string.c_str());
+		// puts(printf_string.c_str());
 		std::string header_string = std::string();
 		constexpr static int a_cols[] = { cols... };
 		if (fp == nullptr){
@@ -659,11 +660,11 @@ struct TableView {
 };
 
 template <class T>
-constexpr static inline bool is_vector(const ColRef<T>&) {
+constexpr static bool is_vector(const ColRef<T>&) {
 	return true;
 }
 template <class T>
-constexpr static inline bool is_vector(const vector_type<T>&) {
+constexpr static bool is_vector(const vector_type<T>&) {
 	return true;
 }
 
@@ -913,6 +914,42 @@ VT<bool> operator >(const T2& lhs, const VT<T1>& rhs) {
 	return ret;
 }
 
+#define _AQ_OP_(x) __AQ_OP__##x
+#define __AQ_OP__add +
+#define __AQ_OP__minus -
+#define __AQ_OP__div *
+#define __AQ_OP__mul /
+#define __AQ_OP__and &
+#define __AQ_OP__or |
+#define __AQ_OP__xor ^
+#define __AQ_OP__gt >
+#define __AQ_OP__lt <
+#define __AQ_OP__gte >=
+#define __AQ_OP__lte <=
+#define __AQ_OP__eq ==
+#define __AQ_OP__neq !=
+
+#define __D_AQOP(x) \
+template <class T1, class T2, template<typename> class VT, class Ret>\
+void aqop_##x (const VT<T1>& lhs, const VT<T2>& rhs, Ret& ret){\
+	for (uint32_t i = 0; i < ret.size; ++i)\
+		ret[i] = lhs[i] _AQ_OP_(x) rhs[i];\
+}
+
+__D_AQOP(add)
+__D_AQOP(minus)
+__D_AQOP(div)
+__D_AQOP(mul)
+__D_AQOP(and)
+__D_AQOP(or)
+__D_AQOP(xor)
+__D_AQOP(gt)
+__D_AQOP(lt)
+__D_AQOP(gte)
+__D_AQOP(lte)
+__D_AQOP(eq)
+__D_AQOP(neq)
+
 
 template <class ...Types>
 void print(const TableInfo<Types...>& v, const char* delimiter = " ", const char* endline = "\n") {

server/vector_type.hpp

@@ -187,15 +187,15 @@ public:
 		grow<false>(sz);
 	}
 
-	void emplace_back(const _Ty& _val) {
+	inline void emplace_back(const _Ty& _val) {
 		grow();
 		container[size++] = _val;
 	}
-	void emplace_back(_Ty& _val) {
+	inline void emplace_back(_Ty& _val) {
 		grow();
 		container[size++] = std::move(_val);
 	}
-	void emplace_back(_Ty&& _val) {
+	inline void emplace_back(_Ty&& _val) {
 		grow();
 		container[size++] = std::move(_val);
 	}
@@ -212,10 +212,10 @@ public:
 		return _it;
 	}
 
-	iterator_t begin() const {
+	inline iterator_t begin() const {
 		return container;
 	}
-	iterator_t end() const {
+	inline iterator_t end() const {
 		return container + size;
 	}
 
@@ -229,7 +229,7 @@ public:
 		return container[_i];
 	}
 
-	void shrink_to_fit() {
+	inline void shrink_to_fit() {
 		if (size && capacity != size) {
 			capacity = size;
 			_Ty* _container = (_Ty*)malloc(sizeof(_Ty) * size);
@@ -239,13 +239,17 @@ public:
 		}
 	}
 
-	_Ty& back() {
+	inline void clear() {
+		this->size = 0;
+	}
+
+	inline _Ty& back() {
 		return container[size - 1];
 	}
-	void qpop() {
+	inline void qpop() {
 		size = size ? size - 1 : size;
 	}
-	void pop_resize() {
+	inline void pop_resize() {
 		if (size) {
 			--size;
 			if (capacity > (size << 1))
@@ -258,7 +262,7 @@ public:
 			}
 		}
 	}
-	_Ty pop() {
+	inline _Ty pop() {
 		return container[--size];
 	}
 	void merge(vector_type<_Ty>& _other) {
@@ -368,7 +372,7 @@ public:
 #define Ops(o, x) \
 	template<typename T>\
 	vector_type<typename types::Coercion<_Ty, T>::type> operator o (const vector_type<T>& r) const {\
-		/*[[likely]] if (r.size == size) {*/\
+		/*if (r.size == size) { [[likely]] */\
 			return x(r);\
 		/*}*/\
 	}
@@ -376,7 +380,7 @@ public:
 #define Opseq(o, x) \
 	template<typename T>\
 	vector_type<typename types::Coercion<_Ty, T>::type> operator o##= (const vector_type<T>& r) {\
-		/*[[likely]] if (r.size == size) {*/\
+		/*if (r.size == size) { [[likely]] */\
 			return x##eq(r);\
 		/*}*/\
 	}