Improve parser's one-extra-token lookahead mechanism.

There are a couple of places in our grammar that fail to be strict LALR(1),
by requiring more than a single token of lookahead to decide what to do.
Up to now we've dealt with that by using a filter between the lexer and
parser that merges adjacent tokens into one in the places where two tokens
of lookahead are necessary.  But that creates a number of user-visible
anomalies, for instance that you can't name a CTE "ordinality" because
"WITH ordinality AS ..." triggers folding of WITH and ORDINALITY into one
token.  I realized that there's a better way.

In this patch, we still do the lookahead basically as before, but we never
merge the second token into the first; we replace just the first token by
a special lookahead symbol when one of the lookahead pairs is seen.

This requires a couple extra productions in the grammar, but it involves
fewer special tokens, so that the grammar tables come out a bit smaller
than before.  The filter logic is no slower than before, perhaps a bit
faster.

I also fixed the filter logic so that when backing up after a lookahead,
the current token's terminator is correctly restored; this eliminates some
weird behavior in error message issuance, as is shown by the one change in
existing regression test outputs.

I believe that this patch entirely eliminates odd behaviors caused by
lookahead for WITH.  It doesn't really improve the situation for NULLS
followed by FIRST/LAST unfortunately: those sequences still act like a
reserved word, even though there are cases where they should be seen as two
ordinary identifiers, eg "SELECT nulls first FROM ...".  I experimented
with additional grammar hacks but couldn't find any simple solution for
that.  Still, this is better than before, and it seems much more likely
that we *could* somehow solve the NULLS case on the basis of this filter
behavior than the previous one.

1 files changed, 65 insertions, 51 deletions

diff --git a/src/interfaces/ecpg/preproc/parser.c b/src/interfaces/ecpg/preproc/parser.c
index f1188269533..099a213d118 100644
--- a/src/interfaces/ecpg/preproc/parser.c
+++ b/src/interfaces/ecpg/preproc/parser.c
@@ -3,11 +3,8 @@
  * parser.c
  *		Main entry point/driver for PostgreSQL grammar
  *
- * Note that the grammar is not allowed to perform any table access
- * (since we need to be able to do basic parsing even while inside an
- * aborted transaction).  Therefore, the data structures returned by
- * the grammar are "raw" parsetrees that still need to be analyzed by
- * analyze.c and related files.
+ * This should match src/backend/parser/parser.c, except that we do not
+ * need to bother with re-entrant interfaces.
  *
  *
  * Portions Copyright (c) 1996-2015, PostgreSQL Global Development Group
@@ -29,18 +26,21 @@ static bool have_lookahead;		/* is lookahead info valid? */
 static int	lookahead_token;	/* one-token lookahead */
 static YYSTYPE lookahead_yylval;	/* yylval for lookahead token */
 static YYLTYPE lookahead_yylloc;	/* yylloc for lookahead token */
+static char *lookahead_yytext;	/* start current token */
+static char *lookahead_end;		/* end of current token */
+static char lookahead_hold_char;	/* to be put back at *lookahead_end */
 
 
 /*
  * Intermediate filter between parser and base lexer (base_yylex in scan.l).
  *
- * The filter is needed because in some cases the standard SQL grammar
+ * This filter is needed because in some cases the standard SQL grammar
  * requires more than one token lookahead.  We reduce these cases to one-token
- * lookahead by combining tokens here, in order to keep the grammar LALR(1).
+ * lookahead by replacing tokens here, in order to keep the grammar LALR(1).
  *
  * Using a filter is simpler than trying to recognize multiword tokens
  * directly in scan.l, because we'd have to allow for comments between the
- * words.  Furthermore it's not clear how to do it without re-introducing
+ * words.  Furthermore it's not clear how to do that without re-introducing
  * scanner backtrack, which would cost more performance than this filter
  * layer does.
  */
@@ -49,8 +49,10 @@ filtered_base_yylex(void)
 {
 	int			cur_token;
 	int			next_token;
+	int			cur_token_length;
 	YYSTYPE		cur_yylval;
 	YYLTYPE		cur_yylloc;
+	char	   *cur_yytext;
 
 	/* Get next token --- we might already have it */
 	if (have_lookahead)
@@ -58,74 +60,86 @@ filtered_base_yylex(void)
 		cur_token = lookahead_token;
 		base_yylval = lookahead_yylval;
 		base_yylloc = lookahead_yylloc;
+		yytext = lookahead_yytext;
+		*lookahead_end = lookahead_hold_char;
 		have_lookahead = false;
 	}
 	else
 		cur_token = base_yylex();
 
-	/* Do we need to look ahead for a possible multiword token? */
+	/*
+	 * If this token isn't one that requires lookahead, just return it.  If it
+	 * does, determine the token length.  (We could get that via strlen(), but
+	 * since we have such a small set of possibilities, hardwiring seems
+	 * feasible and more efficient.)
+	 */
 	switch (cur_token)
 	{
 		case NULLS_P:
+			cur_token_length = 5;
+			break;
+		case WITH:
+			cur_token_length = 4;
+			break;
+		default:
+			return cur_token;
+	}
+
+	/*
+	 * Identify end+1 of current token.  base_yylex() has temporarily stored a
+	 * '\0' here, and will undo that when we call it again.  We need to redo
+	 * it to fully revert the lookahead call for error reporting purposes.
+	 */
+	lookahead_end = yytext + cur_token_length;
+	Assert(*lookahead_end == '\0');
+
+	/* Save and restore lexer output variables around the call */
+	cur_yylval = base_yylval;
+	cur_yylloc = base_yylloc;
+	cur_yytext = yytext;
+
+	/* Get next token, saving outputs into lookahead variables */
+	next_token = base_yylex();
+
+	lookahead_token = next_token;
+	lookahead_yylval = base_yylval;
+	lookahead_yylloc = base_yylloc;
+	lookahead_yytext = yytext;
+
+	base_yylval = cur_yylval;
+	base_yylloc = cur_yylloc;
+	yytext = cur_yytext;
+
+	/* Now revert the un-truncation of the current token */
+	lookahead_hold_char = *lookahead_end;
+	*lookahead_end = '\0';
+
+	have_lookahead = true;
 
-			/*
-			 * NULLS FIRST and NULLS LAST must be reduced to one token
-			 */
-			cur_yylval = base_yylval;
-			cur_yylloc = base_yylloc;
-			next_token = base_yylex();
+	/* Replace cur_token if needed, based on lookahead */
+	switch (cur_token)
+	{
+		case NULLS_P:
+			/* Replace NULLS_P by NULLS_LA if it's followed by FIRST or LAST */
 			switch (next_token)
 			{
 				case FIRST_P:
-					cur_token = NULLS_FIRST;
-					break;
 				case LAST_P:
-					cur_token = NULLS_LAST;
-					break;
-				default:
-					/* save the lookahead token for next time */
-					lookahead_token = next_token;
-					lookahead_yylval = base_yylval;
-					lookahead_yylloc = base_yylloc;
-					have_lookahead = true;
-					/* and back up the output info to cur_token */
-					base_yylval = cur_yylval;
-					base_yylloc = cur_yylloc;
+					cur_token = NULLS_LA;
 					break;
 			}
 			break;
 
 		case WITH:
-
-			/*
-			 * WITH TIME must be reduced to one token
-			 */
-			cur_yylval = base_yylval;
-			cur_yylloc = base_yylloc;
-			next_token = base_yylex();
+			/* Replace WITH by WITH_LA if it's followed by TIME or ORDINALITY */
 			switch (next_token)
 			{
 				case TIME:
-					cur_token = WITH_TIME;
-					break;
 				case ORDINALITY:
-					cur_token = WITH_ORDINALITY;
-					break;
-				default:
-					/* save the lookahead token for next time */
-					lookahead_token = next_token;
-					lookahead_yylval = base_yylval;
-					lookahead_yylloc = base_yylloc;
-					have_lookahead = true;
-					/* and back up the output info to cur_token */
-					base_yylval = cur_yylval;
-					base_yylloc = cur_yylloc;
+					cur_token = WITH_LA;
 					break;
 			}
 			break;
-
-		default:
-			break;
 	}
 
 	return cur_token;