pg 1.0.0 → 1.5.9

data/ext/pg_text_encoder.c

@@ -1,6 +1,6 @@
 /*
  * pg_text_encoder.c - PG::TextEncoder module
- * $Id: pg_text_encoder.c,v e61a06f1f5ed 2015/12/25 21:14:21 lars $
+ * $Id$
  *
  */
 
@@ -41,7 +41,7 @@
 
 
 #include "pg.h"
-#include "util.h"
+#include "pg_util.h"
 #ifdef HAVE_INTTYPES_H
 #include <inttypes.h>
 #endif
@@ -50,6 +50,9 @@
 VALUE rb_mPG_TextEncoder;
 static ID s_id_encode;
 static ID s_id_to_i;
+static ID s_id_to_s;
+static ID s_cBigDecimal;
+static VALUE s_str_F;
 
 static int pg_text_enc_integer(t_pg_coder *this, VALUE value, char *out, VALUE *intermediate, int enc_idx);
 
@@ -116,6 +119,10 @@ pg_text_enc_boolean(t_pg_coder *this, VALUE value, char *out, VALUE *intermediat
 int
 pg_coder_enc_to_s(t_pg_coder *this, VALUE value, char *out, VALUE *intermediate, int enc_idx)
 {
+	/* Attention:
+	 * In contrast to all other encoders, the "this" pointer of this encoder can be NULL.
+	 * This is because it is used as a fall-back if no encoder is defined.
+	 */
 	VALUE str = rb_obj_as_string(value);
 	if( ENCODING_GET(str) == enc_idx ){
 		*intermediate = str;
@@ -125,11 +132,29 @@ pg_coder_enc_to_s(t_pg_coder *this, VALUE value, char *out, VALUE *intermediate,
 	return -1;
 }
 
+static int
+count_leading_zero_bits(unsigned long long x)
+{
+#if defined(__GNUC__) || defined(__clang__)
+	return __builtin_clzll(x);
+#elif defined(_MSC_VER)
+	DWORD r = 0;
+	_BitScanForward64(&r, x);
+	return (int)r;
+#else
+	unsigned int a;
+	for(a=0; a < sizeof(unsigned long long) * 8; a++){
+		if( x & (1 << (sizeof(unsigned long long) * 8 - 1))) return a;
+		x <<= 1;
+	}
+	return a;
+#endif
+}
 
 /*
  * Document-class: PG::TextEncoder::Integer < PG::SimpleEncoder
  *
- * This is the encoder class for the PostgreSQL int types.
+ * This is the encoder class for the PostgreSQL integer types.
  *
  * Non-Integer values are expected to have method +to_i+ defined.
  *
@@ -144,20 +169,23 @@ pg_text_enc_integer(t_pg_coder *this, VALUE value, char *out, VALUE *intermediat
 			char *start = out;
 			int len;
 			int neg = 0;
-			long long ll = NUM2LL(*intermediate);
+			long long sll = NUM2LL(*intermediate);
+			unsigned long long ll;
 
-			if (ll < 0) {
-				/* We don't expect problems with the most negative integer not being representable
-				 * as a positive integer, because Fixnum is only up to 63 bits.
+			if (sll < 0) {
+				/* Avoid problems with the most negative integer not being representable
+				 * as a positive integer, by using unsigned long long for encoding.
 				 */
-				ll = -ll;
+				ll = -sll;
 				neg = 1;
+			} else {
+				ll = sll;
 			}
 
 			/* Compute the result string backwards. */
 			do {
-				long long remainder;
-				long long oldval = ll;
+				unsigned long long remainder;
+				unsigned long long oldval = ll;
 
 				ll /= 10;
 				remainder = oldval - ll * 10;
@@ -167,7 +195,7 @@ pg_text_enc_integer(t_pg_coder *this, VALUE value, char *out, VALUE *intermediat
 			if (neg)
 				*out++ = '-';
 
-			len = out - start;
+			len = (int)(out - start);
 
 			/* Reverse string. */
 			out--;
@@ -184,45 +212,17 @@ pg_text_enc_integer(t_pg_coder *this, VALUE value, char *out, VALUE *intermediat
 	}else{
 		*intermediate = pg_obj_to_i(value);
 		if(TYPE(*intermediate) == T_FIXNUM){
-			int len;
 			long long sll = NUM2LL(*intermediate);
-			long long ll = sll < 0 ? -sll : sll;
-			if( ll < 100000000 ){
-				if( ll < 10000 ){
-					if( ll < 100 ){
-						len = ll < 10 ? 1 : 2;
-					}else{
-						len = ll < 1000 ? 3 : 4;
-					}
-				}else{
-					if( ll < 1000000 ){
-						len = ll < 100000 ? 5 : 6;
-					}else{
-						len = ll < 10000000 ? 7 : 8;
-					}
-				}
-			}else{
-				if( ll < 1000000000000LL ){
-					if( ll < 10000000000LL ){
-						len = ll < 1000000000LL ? 9 : 10;
-					}else{
-						len = ll < 100000000000LL ? 11 : 12;
-					}
-				}else{
-					if( ll < 100000000000000LL ){
-						len = ll < 10000000000000LL ? 13 : 14;
-					}else{
-						return pg_coder_enc_to_s(this, *intermediate, NULL, intermediate, enc_idx);
-					}
-				}
-			}
-			return sll < 0 ? len+1 : len;
+			unsigned long long ll = sll < 0 ? -sll : sll;
+			int len = (sizeof(unsigned long long) * 8 - count_leading_zero_bits(ll)) / 3;
+			return sll < 0 ? len+2 : len+1;
 		}else{
 			return pg_coder_enc_to_s(this, *intermediate, NULL, intermediate, enc_idx);
 		}
 	}
 }
 
+#define MAX_DOUBLE_DIGITS 16
 
 /*
  * Document-class: PG::TextEncoder::Float < PG::SimpleEncoder
@@ -235,6 +235,12 @@ pg_text_enc_float(t_pg_coder *conv, VALUE value, char *out, VALUE *intermediate,
 {
 	if(out){
 		double dvalue = NUM2DBL(value);
+		int len = 0;
+		int neg = 0;
+		int exp2i, exp10i, i;
+		unsigned long long ll, remainder, oldval;
+		VALUE intermediate;
+
 		/* Cast to the same strings as value.to_s . */
 		if( isinf(dvalue) ){
 			if( dvalue < 0 ){
@@ -248,12 +254,145 @@ pg_text_enc_float(t_pg_coder *conv, VALUE value, char *out, VALUE *intermediate,
 			memcpy( out, "NaN", 3);
 			return 3;
 		}
-		return sprintf( out, "%.16E", dvalue);
+
+		/*
+		 * The following computation is roughly a conversion kind of
+		 *   sprintf( out, "%.16E", dvalue);
+		 */
+
+		/* write the algebraic sign */
+		if( dvalue < 0 ) {
+			dvalue = -dvalue;
+			*out++ = '-';
+			neg++;
+		}
+
+		/* retrieve the power of 2 exponent */
+		frexp(dvalue, &exp2i);
+		/* compute the power of 10 exponent */
+		exp10i = (int)floor(exp2i * 0.30102999566398114); /* Math.log(2)/Math.log(10) */
+		/* move the decimal point, so that we get an integer of MAX_DOUBLE_DIGITS decimal digits */
+		ll = (unsigned long long)(dvalue * pow(10, MAX_DOUBLE_DIGITS - 1 - exp10i) + 0.5);
+
+		/* avoid leading zeros due to inaccuracy of deriving exp10i from exp2i */
+		/* otherwise we would print "09.0" instead of "9.0" */
+		if( ll < 1000000000000000 ){ /* pow(10, MAX_DOUBLE_DIGITS-1) */
+			exp10i--;
+			ll *= 10;
+		}
+
+		if( exp10i <= -5 || exp10i >= 15 ) {
+			/* Write the float in exponent format (1.23e45) */
+
+			/* write fraction digits from right to left */
+			for( i = MAX_DOUBLE_DIGITS; i > 1; i--){
+				oldval = ll;
+				ll /= 10;
+				remainder = oldval - ll * 10;
+				/* omit trailing zeros */
+				if(remainder != 0 || len ) {
+					out[i] = '0' + remainder;
+					len++;
+				}
+			}
+
+			/* write decimal point */
+			if( len ){
+				out[1] = '.';
+				len++;
+			}
+
+			/* write remaining single digit left to the decimal point */
+			oldval = ll;
+			ll /= 10;
+			remainder = oldval - ll * 10;
+			out[0] = '0' + remainder;
+			len++;
+
+			/* write exponent */
+			out[len++] = 'e';
+			intermediate = INT2NUM(exp10i);
+
+			return neg + len + pg_text_enc_integer(conv, Qnil, out + len, &intermediate, enc_idx);
+		} else {
+			/* write the float in non exponent format (0.001234 or 123450.0) */
+
+			/* write digits from right to left */
+			int lz = exp10i < 0 ? 0 : exp10i;
+			for( i = MAX_DOUBLE_DIGITS - (exp10i < 0 ? exp10i : 0); i >= 0; i-- ){
+				oldval = ll;
+				ll /= 10;
+				remainder = oldval - ll * 10;
+				/* write decimal point */
+				if( i - 1 == lz ){
+					out[i--] = '.';
+					len++;
+				}
+				/* if possible then omit trailing zeros */
+				if(remainder != 0 || len || i - 2 == lz) {
+					out[i] = '0' + remainder;
+					len++;
+				}
+			}
+			return neg + len;
+		}
 	}else{
-		return 23;
+		return 1 /*sign*/ + MAX_DOUBLE_DIGITS + 1 /*dot*/ + 1 /*e*/ + 1 /*exp sign*/ + 3 /*exp digits*/;
+	}
+}
+
+
+/*
+ * Document-class: PG::TextEncoder::Numeric < PG::SimpleEncoder
+ *
+ * This is the encoder class for the PostgreSQL numeric types.
+ *
+ * It converts Integer, Float and BigDecimal objects.
+ * All other objects are expected to respond to +to_s+.
+ *
+ * As soon as this class is used, it requires the 'bigdecimal' gem.
+ */
+static int
+pg_text_enc_numeric(t_pg_coder *this, VALUE value, char *out, VALUE *intermediate, int enc_idx)
+{
+	switch(TYPE(value)){
+		case T_FIXNUM:
+		case T_BIGNUM:
+			return pg_text_enc_integer(this, value, out, intermediate, enc_idx);
+		case T_FLOAT:
+			return pg_text_enc_float(this, value, out, intermediate, enc_idx);
+		default:
+			if(out){ /* second pass */
+				rb_bug("unexpected value type: %d", TYPE(value));
+			} else { /* first pass */
+				if( rb_obj_is_kind_of(value, s_cBigDecimal) ){
+					/* value.to_s('F') */
+					*intermediate = rb_funcall(value, s_id_to_s, 1, s_str_F);
+					return -1; /* no second pass */
+				} else {
+					return pg_coder_enc_to_s(this, value, NULL, intermediate, enc_idx);
+					/* no second pass */
+				}
+			}
 	}
 }
 
+/* called per autoload when TextEncoder::Numeric is used */
+static VALUE
+init_pg_text_encoder_numeric(VALUE rb_mPG_TextDecoder)
+{
+	s_str_F = rb_str_freeze(rb_str_new_cstr("F"));
+	rb_global_variable(&s_str_F);
+	rb_funcall(rb_mPG, rb_intern("require_bigdecimal_without_warning"), 0);
+	s_cBigDecimal = rb_const_get(rb_cObject, rb_intern("BigDecimal"));
+
+	/* dummy = rb_define_class_under( rb_mPG_TextEncoder, "Numeric", rb_cPG_SimpleEncoder ); */
+	pg_define_coder( "Numeric", pg_text_enc_numeric, rb_cPG_SimpleEncoder, rb_mPG_TextEncoder );
+
+	return Qnil;
+}
+
+
 static const char hextab[] = {
 	'0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'a', 'b', 'c', 'd', 'e', 'f'
 };
@@ -261,13 +400,16 @@ static const char hextab[] = {
 /*
  * Document-class: PG::TextEncoder::Bytea < PG::SimpleEncoder
  *
- * This is an encoder class for the PostgreSQL bytea type for server version 9.0
- * or newer.
+ * This is an encoder class for the PostgreSQL +bytea+ type.
  *
  * The binary String is converted to hexadecimal representation for transmission
  * in text format. For query bind parameters it is recommended to use
- * PG::BinaryEncoder::Bytea instead, in order to decrease network traffic and
- * CPU usage.
+ * PG::BinaryEncoder::Bytea or the hash form <tt>{value: binary_string, format: 1}</tt> instead,
+ * in order to decrease network traffic and CPU usage.
+ * See PG::Connection#exec_params for using the hash form.
+ *
+ * This encoder is particular useful when PG::TextEncoder::CopyRow is used with the COPY command.
+ * In this case there's no way to change the format of a single column to binary, so that the data have to be converted to bytea hex representation.
  *
  */
 static int
@@ -286,11 +428,11 @@ pg_text_enc_bytea(t_pg_coder *conv, VALUE value, char *out, VALUE *intermediate,
 			*optr++ = hextab[c >> 4];
 			*optr++ = hextab[c & 0xf];
 		}
-		return optr - out;
+		return (int)(optr - out);
 	}else{
 		*intermediate = rb_obj_as_string(value);
 		/* The output starts with "\x" and each character is converted to hex. */
-		return 2 + RSTRING_LEN(*intermediate) * 2;
+		return 2 + RSTRING_LENINT(*intermediate) * 2;
 	}
 }
 
@@ -301,7 +443,7 @@ quote_array_buffer( void *_this, char *p_in, int strlen, char *p_out ){
 	t_pg_composite_coder *this = _this;
 	char *ptr1;
 	char *ptr2;
-	int backslashs = 0;
+	int backslashes = 0;
 	int needquote;
 
 	/* count data plus backslashes; detect chars needing quotes */
@@ -318,7 +460,7 @@ quote_array_buffer( void *_this, char *p_in, int strlen, char *p_out ){
 
 		if (ch == '"' || ch == '\\'){
 			needquote = 1;
-			backslashs++;
+			backslashes++;
 		} else if (ch == '{' || ch == '}' || ch == this->delimiter ||
 					ch == ' ' || ch == '\t' || ch == '\n' || ch == '\r' || ch == '\v' || ch == '\f'){
 			needquote = 1;
@@ -327,12 +469,12 @@ quote_array_buffer( void *_this, char *p_in, int strlen, char *p_out ){
 
 	if( needquote ){
 		ptr1 = p_in + strlen;
-		ptr2 = p_out + strlen + backslashs + 2;
+		ptr2 = p_out + strlen + backslashes + 2;
 		/* Write end quote */
 		*--ptr2 = '"';
 
 		/* Then store the escaped string on the final position, walking
-			* right to left, until all backslashs are placed. */
+			* right to left, until all backslashes are placed. */
 		while( ptr1 != p_in ) {
 			*--ptr2 = *--ptr1;
 			if(*ptr2 == '"' || *ptr2 == '\\'){
@@ -341,7 +483,7 @@ quote_array_buffer( void *_this, char *p_in, int strlen, char *p_out ){
 		}
 		/* Write start quote */
 		*p_out = '"';
-		return strlen + backslashs + 2;
+		return strlen + backslashes + 2;
 	} else {
 		if( p_in != p_out )
 			memcpy( p_out, p_in, strlen );
@@ -468,20 +610,19 @@ pg_text_enc_array(t_pg_coder *conv, VALUE value, char *out, VALUE *intermediate,
 static char *
 quote_identifier( VALUE value, VALUE out_string, char *current_out ){
 	char *p_in = RSTRING_PTR(value);
-	char *ptr1;
 	size_t strlen = RSTRING_LEN(value);
+	char *p_inend = p_in + strlen;
 	char *end_capa = current_out;
 
 	PG_RB_STR_ENSURE_CAPA( out_string, strlen + 2, current_out, end_capa );
 	*current_out++ = '"';
-	for(ptr1 = p_in; ptr1 != p_in + strlen; ptr1++) {
-		char c = *ptr1;
+	for(; p_in != p_inend; p_in++) {
+		char c = *p_in;
 		if (c == '"'){
-			strlen++;
-			PG_RB_STR_ENSURE_CAPA( out_string, p_in - ptr1 + strlen + 1, current_out, end_capa );
+			PG_RB_STR_ENSURE_CAPA( out_string, p_inend - p_in + 2, current_out, end_capa );
 			*current_out++ = '"';
 		} else if (c == 0){
-			break;
+			rb_raise(rb_eArgError, "string contains null byte");
 		}
 		*current_out++ = c;
 	}
@@ -494,8 +635,8 @@ quote_identifier( VALUE value, VALUE out_string, char *current_out ){
 static char *
 pg_text_enc_array_identifier(VALUE value, VALUE string, char *out, int enc_idx)
 {
-	int i;
-	int nr_elems;
+	long i;
+	long nr_elems;
 
 	Check_Type(value, T_ARRAY);
 	nr_elems = RARRAY_LEN(value);
@@ -521,7 +662,8 @@ pg_text_enc_array_identifier(VALUE value, VALUE string, char *out, int enc_idx)
  *
  * This is the encoder class for PostgreSQL identifiers.
  *
- * An Array value can be used for "schema.table.column" type identifiers:
+ * An Array value can be used for identifiers of the kind "schema.table.column".
+ * This ensures that each element is properly quoted:
  *   PG::TextEncoder::Identifier.new.encode(['schema', 'table', 'column'])
  *      => '"schema"."table"."column"'
  *
@@ -556,22 +698,22 @@ static int
 quote_literal_buffer( void *_this, char *p_in, int strlen, char *p_out ){
 	char *ptr1;
 	char *ptr2;
-	int backslashs = 0;
+	int backslashes = 0;
 
 	/* count required backlashs */
 	for(ptr1 = p_in; ptr1 != p_in + strlen; ptr1++) {
 		if (*ptr1 == '\''){
-			backslashs++;
+			backslashes++;
 		}
 	}
 
 	ptr1 = p_in + strlen;
-	ptr2 = p_out + strlen + backslashs + 2;
+	ptr2 = p_out + strlen + backslashes + 2;
 	/* Write end quote */
 	*--ptr2 = '\'';
 
 	/* Then store the escaped string on the final position, walking
-		* right to left, until all backslashs are placed. */
+		* right to left, until all backslashes are placed. */
 	while( ptr1 != p_in ) {
 		*--ptr2 = *--ptr1;
 		if(*ptr2 == '\''){
@@ -580,7 +722,7 @@ quote_literal_buffer( void *_this, char *p_in, int strlen, char *p_out ){
 	}
 	/* Write start quote */
 	*p_out = '\'';
-	return strlen + backslashs + 2;
+	return strlen + backslashes + 2;
 }
 
 
@@ -589,7 +731,13 @@ quote_literal_buffer( void *_this, char *p_in, int strlen, char *p_out ){
  *
  * This is the encoder class for PostgreSQL literals.
  *
- * A literal is quoted and escaped by the +'+ character.
+ * A literal is quoted and escaped by the <tt>'</tt> character, so that it can be inserted into SQL queries.
+ * It works equal to PG::Connection#escape_literal, but integrates into the type cast system of ruby-pg.
+ *
+ * Both expressions have the same result:
+ *   conn.escape_literal(PG::TextEncoder::Array.new.encode(["v1","v2"])) # => "'{v1,v2}'"
+ *   PG::TextEncoder::QuotedLiteral.new(elements_type: PG::TextEncoder::Array.new).encode(["v1","v2"]) # => "'{v1,v2}'"
+ * While escape_literal requires a intermediate ruby string allocation, QuotedLiteral encodes the values directly to the result string.
  *
  */
 static int
@@ -652,13 +800,15 @@ pg_text_enc_to_base64(t_pg_coder *conv, VALUE value, char *out, VALUE *intermedi
 
 
 void
-init_pg_text_encoder()
+init_pg_text_encoder(void)
 {
 	s_id_encode = rb_intern("encode");
 	s_id_to_i = rb_intern("to_i");
+	s_id_to_s = rb_intern("to_s");
 
 	/* This module encapsulates all encoder classes with text output format */
 	rb_mPG_TextEncoder = rb_define_module_under( rb_mPG, "TextEncoder" );
+	rb_define_private_method(rb_singleton_class(rb_mPG_TextEncoder), "init_numeric", init_pg_text_encoder_numeric, 0);
 
 	/* Make RDoc aware of the encoder classes... */
 	/* dummy = rb_define_class_under( rb_mPG_TextEncoder, "Boolean", rb_cPG_SimpleEncoder ); */