1 /*-------------------------------------------------------------------------
2  *
3  * read.c
4  *	  routines to convert a string (legal ascii representation of node) back
5  *	  to nodes
6  *
7  * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
8  * Portions Copyright (c) 1994, Regents of the University of California
9  *
10  *
11  * IDENTIFICATION
12  *	  src/backend/nodes/read.c
13  *
14  * HISTORY
15  *	  AUTHOR			DATE			MAJOR EVENT
16  *	  Andrew Yu			Nov 2, 1994		file creation
17  *
18  *-------------------------------------------------------------------------
19  */
20 #include "postgres.h"
21 
22 #include <ctype.h>
23 
24 #include "nodes/pg_list.h"
25 #include "nodes/readfuncs.h"
26 #include "nodes/value.h"
27 
28 
29 /* Static state for pg_strtok */
30 static char *pg_strtok_ptr = NULL;
31 
32 
33 /*
34  * stringToNode -
35  *	  returns a Node with a given legal ASCII representation
36  */
37 void *
stringToNode(char * str)38 stringToNode(char *str)
39 {
40 	char	   *save_strtok;
41 	void	   *retval;
42 
43 	/*
44 	 * We save and restore the pre-existing state of pg_strtok. This makes the
45 	 * world safe for re-entrant invocation of stringToNode, without incurring
46 	 * a lot of notational overhead by having to pass the next-character
47 	 * pointer around through all the readfuncs.c code.
48 	 */
49 	save_strtok = pg_strtok_ptr;
50 
51 	pg_strtok_ptr = str;		/* point pg_strtok at the string to read */
52 
53 	retval = nodeRead(NULL, 0); /* do the reading */
54 
55 	pg_strtok_ptr = save_strtok;
56 
57 	return retval;
58 }
59 
60 /*****************************************************************************
61  *
62  * the lisp token parser
63  *
64  *****************************************************************************/
65 
66 /*
67  * pg_strtok --- retrieve next "token" from a string.
68  *
69  * Works kinda like strtok, except it never modifies the source string.
70  * (Instead of storing nulls into the string, the length of the token
71  * is returned to the caller.)
72  * Also, the rules about what is a token are hard-wired rather than being
73  * configured by passing a set of terminating characters.
74  *
75  * The string is assumed to have been initialized already by stringToNode.
76  *
77  * The rules for tokens are:
78  *	* Whitespace (space, tab, newline) always separates tokens.
79  *	* The characters '(', ')', '{', '}' form individual tokens even
80  *	  without any whitespace around them.
81  *	* Otherwise, a token is all the characters up to the next whitespace
82  *	  or occurrence of one of the four special characters.
83  *	* A backslash '\' can be used to quote whitespace or one of the four
84  *	  special characters, so that it is treated as a plain token character.
85  *	  Backslashes themselves must also be backslashed for consistency.
86  *	  Any other character can be, but need not be, backslashed as well.
87  *	* If the resulting token is '<>' (with no backslash), it is returned
88  *	  as a non-NULL pointer to the token but with length == 0.  Note that
89  *	  there is no other way to get a zero-length token.
90  *
91  * Returns a pointer to the start of the next token, and the length of the
92  * token (including any embedded backslashes!) in *length.  If there are
93  * no more tokens, NULL and 0 are returned.
94  *
95  * NOTE: this routine doesn't remove backslashes; the caller must do so
96  * if necessary (see "debackslash").
97  *
98  * NOTE: prior to release 7.0, this routine also had a special case to treat
99  * a token starting with '"' as extending to the next '"'.  This code was
100  * broken, however, since it would fail to cope with a string containing an
101  * embedded '"'.  I have therefore removed this special case, and instead
102  * introduced rules for using backslashes to quote characters.  Higher-level
103  * code should add backslashes to a string constant to ensure it is treated
104  * as a single token.
105  */
106 char *
pg_strtok(int * length)107 pg_strtok(int *length)
108 {
109 	char	   *local_str;		/* working pointer to string */
110 	char	   *ret_str;		/* start of token to return */
111 
112 	local_str = pg_strtok_ptr;
113 
114 	while (*local_str == ' ' || *local_str == '\n' || *local_str == '\t')
115 		local_str++;
116 
117 	if (*local_str == '\0')
118 	{
119 		*length = 0;
120 		pg_strtok_ptr = local_str;
121 		return NULL;			/* no more tokens */
122 	}
123 
124 	/*
125 	 * Now pointing at start of next token.
126 	 */
127 	ret_str = local_str;
128 
129 	if (*local_str == '(' || *local_str == ')' ||
130 		*local_str == '{' || *local_str == '}')
131 	{
132 		/* special 1-character token */
133 		local_str++;
134 	}
135 	else
136 	{
137 		/* Normal token, possibly containing backslashes */
138 		while (*local_str != '\0' &&
139 			   *local_str != ' ' && *local_str != '\n' &&
140 			   *local_str != '\t' &&
141 			   *local_str != '(' && *local_str != ')' &&
142 			   *local_str != '{' && *local_str != '}')
143 		{
144 			if (*local_str == '\\' && local_str[1] != '\0')
145 				local_str += 2;
146 			else
147 				local_str++;
148 		}
149 	}
150 
151 	*length = local_str - ret_str;
152 
153 	/* Recognize special case for "empty" token */
154 	if (*length == 2 && ret_str[0] == '<' && ret_str[1] == '>')
155 		*length = 0;
156 
157 	pg_strtok_ptr = local_str;
158 
159 	return ret_str;
160 }
161 
162 /*
163  * debackslash -
164  *	  create a palloc'd string holding the given token.
165  *	  any protective backslashes in the token are removed.
166  */
167 char *
debackslash(char * token,int length)168 debackslash(char *token, int length)
169 {
170 	char	   *result = palloc(length + 1);
171 	char	   *ptr = result;
172 
173 	while (length > 0)
174 	{
175 		if (*token == '\\' && length > 1)
176 			token++, length--;
177 		*ptr++ = *token++;
178 		length--;
179 	}
180 	*ptr = '\0';
181 	return result;
182 }
183 
184 #define RIGHT_PAREN (1000000 + 1)
185 #define LEFT_PAREN	(1000000 + 2)
186 #define LEFT_BRACE	(1000000 + 3)
187 #define OTHER_TOKEN (1000000 + 4)
188 
189 /*
190  * nodeTokenType -
191  *	  returns the type of the node token contained in token.
192  *	  It returns one of the following valid NodeTags:
193  *		T_Integer, T_Float, T_String, T_BitString
194  *	  and some of its own:
195  *		RIGHT_PAREN, LEFT_PAREN, LEFT_BRACE, OTHER_TOKEN
196  *
197  *	  Assumption: the ascii representation is legal
198  */
199 static NodeTag
nodeTokenType(char * token,int length)200 nodeTokenType(char *token, int length)
201 {
202 	NodeTag		retval;
203 	char	   *numptr;
204 	int			numlen;
205 
206 	/*
207 	 * Check if the token is a number
208 	 */
209 	numptr = token;
210 	numlen = length;
211 	if (*numptr == '+' || *numptr == '-')
212 		numptr++, numlen--;
213 	if ((numlen > 0 && isdigit((unsigned char) *numptr)) ||
214 		(numlen > 1 && *numptr == '.' && isdigit((unsigned char) numptr[1])))
215 	{
216 		/*
217 		 * Yes.  Figure out whether it is integral or float; this requires
218 		 * both a syntax check and a range check. strtol() can do both for us.
219 		 * We know the token will end at a character that strtol will stop at,
220 		 * so we do not need to modify the string.
221 		 */
222 		long		val;
223 		char	   *endptr;
224 
225 		errno = 0;
226 		val = strtol(token, &endptr, 10);
227 		(void) val;				/* avoid compiler warning if unused */
228 		if (endptr != token + length || errno == ERANGE
229 #ifdef HAVE_LONG_INT_64
230 		/* if long > 32 bits, check for overflow of int4 */
231 			|| val != (long) ((int32) val)
232 #endif
233 			)
234 			return T_Float;
235 		return T_Integer;
236 	}
237 
238 	/*
239 	 * these three cases do not need length checks, since pg_strtok() will
240 	 * always treat them as single-byte tokens
241 	 */
242 	else if (*token == '(')
243 		retval = LEFT_PAREN;
244 	else if (*token == ')')
245 		retval = RIGHT_PAREN;
246 	else if (*token == '{')
247 		retval = LEFT_BRACE;
248 	else if (*token == '"' && length > 1 && token[length - 1] == '"')
249 		retval = T_String;
250 	else if (*token == 'b')
251 		retval = T_BitString;
252 	else
253 		retval = OTHER_TOKEN;
254 	return retval;
255 }
256 
257 /*
258  * nodeRead -
259  *	  Slightly higher-level reader.
260  *
261  * This routine applies some semantic knowledge on top of the purely
262  * lexical tokenizer pg_strtok().   It can read
263  *	* Value token nodes (integers, floats, or strings);
264  *	* General nodes (via parseNodeString() from readfuncs.c);
265  *	* Lists of the above;
266  *	* Lists of integers or OIDs.
267  * The return value is declared void *, not Node *, to avoid having to
268  * cast it explicitly in callers that assign to fields of different types.
269  *
270  * External callers should always pass NULL/0 for the arguments.  Internally
271  * a non-NULL token may be passed when the upper recursion level has already
272  * scanned the first token of a node's representation.
273  *
274  * We assume pg_strtok is already initialized with a string to read (hence
275  * this should only be invoked from within a stringToNode operation).
276  */
277 void *
nodeRead(char * token,int tok_len)278 nodeRead(char *token, int tok_len)
279 {
280 	Node	   *result;
281 	NodeTag		type;
282 
283 	if (token == NULL)			/* need to read a token? */
284 	{
285 		token = pg_strtok(&tok_len);
286 
287 		if (token == NULL)		/* end of input */
288 			return NULL;
289 	}
290 
291 	type = nodeTokenType(token, tok_len);
292 
293 	switch ((int) type)
294 	{
295 		case LEFT_BRACE:
296 			result = parseNodeString();
297 			token = pg_strtok(&tok_len);
298 			if (token == NULL || token[0] != '}')
299 				elog(ERROR, "did not find '}' at end of input node");
300 			break;
301 		case LEFT_PAREN:
302 			{
303 				List	   *l = NIL;
304 
305 				/*----------
306 				 * Could be an integer list:	(i int int ...)
307 				 * or an OID list:				(o int int ...)
308 				 * or a list of nodes/values:	(node node ...)
309 				 *----------
310 				 */
311 				token = pg_strtok(&tok_len);
312 				if (token == NULL)
313 					elog(ERROR, "unterminated List structure");
314 				if (tok_len == 1 && token[0] == 'i')
315 				{
316 					/* List of integers */
317 					for (;;)
318 					{
319 						int			val;
320 						char	   *endptr;
321 
322 						token = pg_strtok(&tok_len);
323 						if (token == NULL)
324 							elog(ERROR, "unterminated List structure");
325 						if (token[0] == ')')
326 							break;
327 						val = (int) strtol(token, &endptr, 10);
328 						if (endptr != token + tok_len)
329 							elog(ERROR, "unrecognized integer: \"%.*s\"",
330 								 tok_len, token);
331 						l = lappend_int(l, val);
332 					}
333 				}
334 				else if (tok_len == 1 && token[0] == 'o')
335 				{
336 					/* List of OIDs */
337 					for (;;)
338 					{
339 						Oid			val;
340 						char	   *endptr;
341 
342 						token = pg_strtok(&tok_len);
343 						if (token == NULL)
344 							elog(ERROR, "unterminated List structure");
345 						if (token[0] == ')')
346 							break;
347 						val = (Oid) strtoul(token, &endptr, 10);
348 						if (endptr != token + tok_len)
349 							elog(ERROR, "unrecognized OID: \"%.*s\"",
350 								 tok_len, token);
351 						l = lappend_oid(l, val);
352 					}
353 				}
354 				else
355 				{
356 					/* List of other node types */
357 					for (;;)
358 					{
359 						/* We have already scanned next token... */
360 						if (token[0] == ')')
361 							break;
362 						l = lappend(l, nodeRead(token, tok_len));
363 						token = pg_strtok(&tok_len);
364 						if (token == NULL)
365 							elog(ERROR, "unterminated List structure");
366 					}
367 				}
368 				result = (Node *) l;
369 				break;
370 			}
371 		case RIGHT_PAREN:
372 			elog(ERROR, "unexpected right parenthesis");
373 			result = NULL;		/* keep compiler happy */
374 			break;
375 		case OTHER_TOKEN:
376 			if (tok_len == 0)
377 			{
378 				/* must be "<>" --- represents a null pointer */
379 				result = NULL;
380 			}
381 			else
382 			{
383 				elog(ERROR, "unrecognized token: \"%.*s\"", tok_len, token);
384 				result = NULL;	/* keep compiler happy */
385 			}
386 			break;
387 		case T_Integer:
388 
389 			/*
390 			 * we know that the token terminates on a char atol will stop at
391 			 */
392 			result = (Node *) makeInteger(atol(token));
393 			break;
394 		case T_Float:
395 			{
396 				char	   *fval = (char *) palloc(tok_len + 1);
397 
398 				memcpy(fval, token, tok_len);
399 				fval[tok_len] = '\0';
400 				result = (Node *) makeFloat(fval);
401 			}
402 			break;
403 		case T_String:
404 			/* need to remove leading and trailing quotes, and backslashes */
405 			result = (Node *) makeString(debackslash(token + 1, tok_len - 2));
406 			break;
407 		case T_BitString:
408 			{
409 				char	   *val = palloc(tok_len);
410 
411 				/* skip leading 'b' */
412 				memcpy(val, token + 1, tok_len - 1);
413 				val[tok_len - 1] = '\0';
414 				result = (Node *) makeBitString(val);
415 				break;
416 			}
417 		default:
418 			elog(ERROR, "unrecognized node type: %d", (int) type);
419 			result = NULL;		/* keep compiler happy */
420 			break;
421 	}
422 
423 	return (void *) result;
424 }
425