/*
* normalize_expr - Normalize statements or expressions.
*
* Mask constants, strip unnecessary whitespaces and upcase keywords. expr is
* modified in-place (destructively). If readability is more important than
* uniqueness, preserve_space puts one space for one existent whitespace for
* more readability.
*/
void
normalize_expr(char *expr, bool preserve_space)
{
core_yyscan_t yyscanner;
core_yy_extra_type yyextra;
core_YYSTYPE yylval;
YYLTYPE yylloc;
YYLTYPE lastloc;
YYLTYPE start;
char *wp;
int tok, lasttok;
wp = expr;
yyscanner = scanner_init(expr,
&yyextra,
ScanKeywords,
NumScanKeywords);
lasttok = 0;
lastloc = -1;
for (;;)
{
tok = norm_yylex(expr, &yylval, &yylloc, yyscanner);
start = yylloc;
if (lastloc >= 0)
{
int i, i2;
/* Skipping preceding whitespaces */
for(i = lastloc ; i < start && IS_WSCHAR(expr[i]) ; i++);
/* Searching for trailing whitespace */
for(i2 = i; i2 < start && !IS_WSCHAR(expr[i2]) ; i2++);
if (lasttok == IDENT)
{
/* Identifiers are copied in case-sensitive manner. */
memcpy(wp, expr + i, i2 - i);
wp += i2 - i;
}
else
{
/* Upcase keywords */
char *sp;
for (sp = expr + i ; sp < expr + i2 ; sp++, wp++)
*wp = (*sp >= 'a' && *sp <= 'z' ?
*sp - ('a' - 'A') : *sp);
}
/*
* Because of destructive writing, wp must not go advance the
* reading point.
* Although this function's output does not need any validity as a
* statement or an expression, spaces are added where it should be
* to keep some extent of sanity. If readability is more important
* than uniqueness, preserve_space adds one space for each
* existent whitespace.
*/
if (tok > 0 &&
i2 < start &&
(preserve_space ||
(tok >= IDENT && lasttok >= IDENT &&
!IS_CONST(tok) && !IS_CONST(lasttok))))
*wp++ = ' ';
start = i2;
}
/* Exit on parse error. */
if (tok < 0)
{
*wp = 0;
return;
}
/*
* Negative signs before numbers are tokenized separately. And
* explicit positive signs won't appear in deparsed expressions.
*/
if (tok == '-')
tok = norm_yylex(expr, &yylval, &yylloc, yyscanner);
/* Exit on parse error. */
if (tok < 0)
{
*wp = 0;
return;
}
if (IS_CONST(tok))
{
YYLTYPE end;
tok = norm_yylex(expr, &yylval, &end, yyscanner);
/* Exit on parse error. */
if (tok < 0)
{
*wp = 0;
return;
}
/*
* Negative values may be surrounded with parens by the
* deparser. Mask involving them.
*/
if (lasttok == '(' && tok == ')')
{
wp -= (start - lastloc);
start = lastloc;
end++;
}
while (expr[end - 1] == ' ') end--;
*wp++ = '?';
yylloc = end;
}
if (tok == 0)
break;
lasttok = tok;
lastloc = yylloc;
}
*wp = 0;
}
void
normalize_expr(char *expr, bool preserve_space)
{
core_yyscan_t yyscanner;
core_yy_extra_type yyextra;
core_YYSTYPE yylval;
YYLTYPE yylloc;
YYLTYPE lastloc;
YYLTYPE start;
char *wp;
int tok, lasttok;
wp = expr;
yyscanner = scanner_init(expr,
&yyextra,
&ScanKeywords,
ScanKeywordTokens);
/*
* The warnings about nonstandard escape strings is already emitted in the
* core. Just silence them here.
*/
#if PG_VERSION_NUM >= 90500
yyextra.escape_string_warning = false;
#endif
lasttok = 0;
lastloc = -1;
for (;;)
{
tok = norm_yylex(expr, &yylval, &yylloc, yyscanner);
start = yylloc;
if (lastloc >= 0)
{
int i, i2;
/* Skipping preceding whitespaces */
for(i = lastloc ; i < start && IS_WSCHAR(expr[i]) ; i++);
/* Searching for trailing whitespace */
for(i2 = i; i2 < start && !IS_WSCHAR(expr[i2]) ; i2++);
if (lasttok == IDENT)
{
/* Identifiers are copied in case-sensitive manner. */
memcpy(wp, expr + i, i2 - i);
wp += i2 - i;
}
#if PG_VERSION_NUM >= 100000
/*
* Since PG10 pg_stat_statements doesn't store trailing semicolon
* in the column "query". Normalization is basically useless in the
* version but still usefull to match utility commands so follow
* the behavior change.
*/
else if (lasttok == ';')
{
/* Just do nothing */
}
#endif
else
{
/* Upcase keywords */
char *sp;
for (sp = expr + i ; sp < expr + i2 ; sp++, wp++)
*wp = (*sp >= 'a' && *sp <= 'z' ?
*sp - ('a' - 'A') : *sp);
}
/*
* Because of destructive writing, wp must not go advance the
* reading point.
* Although this function's output does not need any validity as a
* statement or an expression, spaces are added where it should be
* to keep some extent of sanity. If readability is more important
* than uniqueness, preserve_space adds one space for each
* existent whitespace.
*/
if (tok > 0 &&
i2 < start &&
(preserve_space ||
(tok >= IDENT && lasttok >= IDENT &&
!IS_CONST(tok) && !IS_CONST(lasttok))))
*wp++ = ' ';
start = i2;
}
/* Exit on parse error. */
if (tok < 0)
{
*wp = 0;
return;
}
/*
* Negative signs before numbers are tokenized separately. And
* explicit positive signs won't appear in deparsed expressions.
*/
if (tok == '-')
tok = norm_yylex(expr, &yylval, &yylloc, yyscanner);
/* Exit on parse error. */
if (tok < 0)
{
*wp = 0;
return;
}
if (IS_CONST(tok))
{
YYLTYPE end;
tok = norm_yylex(expr, &yylval, &end, yyscanner);
/* Exit on parse error. */
if (tok < 0)
{
*wp = 0;
return;
}
/*
* Negative values may be surrounded with parens by the
* deparser. Mask involving them.
*/
if (lasttok == '(' && tok == ')')
{
wp -= (start - lastloc);
start = lastloc;
end++;
}
while (expr[end - 1] == ' ') end--;
*wp++ = '?';
yylloc = end;
}
if (tok == 0)
break;
lasttok = tok;
lastloc = yylloc;
}
*wp = 0;
}
