common/jsonapi: support libpq as a client

Based on a patch by Michael Paquier. For libpq, use PQExpBuffer instead of StringInfo. This requires us to track allocation failures so that we can return JSON_OUT_OF_MEMORY as needed rather than exit()ing. Author: Jacob Champion <jacob.champion@enterprisedb.com> Co-authored-by: Michael Paquier <michael@paquier.xyz> Co-authored-by: Daniel Gustafsson <daniel@yesql.se> Reviewed-by: Peter Eisentraut <peter@eisentraut.org> Discussion: https://www.postgresql.org/message-id/flat/d1b467a78e0e36ed85a09adf979d04cf124a9d4b.camel@vmware.com
2025-11-18 02:02:55 +03:00 · 2024-09-11 08:28:35 +02:00
parent 3beb945da9
commit 0785d1b8b2
9 changed files with 557 additions and 242 deletions
--- a/src/common/jsonapi.c
+++ b/src/common/jsonapi.c
@@ -21,10 +21,70 @@
 #include "mb/pg_wchar.h"
 #include "port/pg_lfind.h"

-#ifndef FRONTEND
+#ifdef JSONAPI_USE_PQEXPBUFFER
+#include "pqexpbuffer.h"
+#else
+#include "lib/stringinfo.h"
 #include "miscadmin.h"
 #endif

+/*
+ * By default, we will use palloc/pfree along with StringInfo.  In libpq,
+ * use malloc and PQExpBuffer, and return JSON_OUT_OF_MEMORY on out-of-memory.
+ */
+#ifdef JSONAPI_USE_PQEXPBUFFER
+
+#define STRDUP(s) strdup(s)
+#define ALLOC(size) malloc(size)
+#define ALLOC0(size) calloc(1, size)
+#define REALLOC realloc
+#define FREE(s) free(s)
+
+#define jsonapi_appendStringInfo			appendPQExpBuffer
+#define jsonapi_appendBinaryStringInfo		appendBinaryPQExpBuffer
+#define jsonapi_appendStringInfoChar		appendPQExpBufferChar
+/* XXX should we add a macro version to PQExpBuffer? */
+#define jsonapi_appendStringInfoCharMacro	appendPQExpBufferChar
+#define jsonapi_makeStringInfo				createPQExpBuffer
+#define jsonapi_initStringInfo				initPQExpBuffer
+#define jsonapi_resetStringInfo				resetPQExpBuffer
+#define jsonapi_termStringInfo				termPQExpBuffer
+#define jsonapi_destroyStringInfo			destroyPQExpBuffer
+
+#else							/* !JSONAPI_USE_PQEXPBUFFER */
+
+#define STRDUP(s) pstrdup(s)
+#define ALLOC(size) palloc(size)
+#define ALLOC0(size) palloc0(size)
+#define REALLOC repalloc
+
+#ifdef FRONTEND
+#define FREE pfree
+#else
+/*
+ * Backend pfree() doesn't handle NULL pointers like the frontend's does; smooth
+ * that over to reduce mental gymnastics. Avoid multiple evaluation of the macro
+ * argument to avoid future hair-pulling.
+ */
+#define FREE(s) do {	\
+	void *__v = (s);	\
+	if (__v)			\
+		pfree(__v);		\
+} while (0)
+#endif
+
+#define jsonapi_appendStringInfo			appendStringInfo
+#define jsonapi_appendBinaryStringInfo		appendBinaryStringInfo
+#define jsonapi_appendStringInfoChar		appendStringInfoChar
+#define jsonapi_appendStringInfoCharMacro	appendStringInfoCharMacro
+#define jsonapi_makeStringInfo				makeStringInfo
+#define jsonapi_initStringInfo				initStringInfo
+#define jsonapi_resetStringInfo				resetStringInfo
+#define jsonapi_termStringInfo(s)			pfree((s)->data)
+#define jsonapi_destroyStringInfo			destroyStringInfo
+
+#endif							/* JSONAPI_USE_PQEXPBUFFER */
+
 /*
 * The context of the parser is maintained by the recursive descent
 * mechanism, but is passed explicitly to the error reporting routine
@@ -103,7 +163,7 @@ struct JsonIncrementalState
 {
 	bool		is_last_chunk;
 	bool		partial_completed;
-	StringInfoData partial_token;
+	jsonapi_StrValType partial_token;
 };

 /*
@@ -219,6 +279,7 @@ static JsonParseErrorType parse_object(JsonLexContext *lex, const JsonSemAction
 static JsonParseErrorType parse_array_element(JsonLexContext *lex, const JsonSemAction *sem);
 static JsonParseErrorType parse_array(JsonLexContext *lex, const JsonSemAction *sem);
 static JsonParseErrorType report_parse_error(JsonParseContext ctx, JsonLexContext *lex);
+static bool allocate_incremental_state(JsonLexContext *lex);

 /* the null action object used for pure validation */
 const JsonSemAction nullSemAction =
@@ -227,6 +288,10 @@ const JsonSemAction nullSemAction =
 	NULL, NULL, NULL, NULL, NULL
 };

+/* sentinels used for out-of-memory conditions */
+static JsonLexContext failed_oom;
+static JsonIncrementalState failed_inc_oom;
+
 /* Parser support routines */

 /*
@@ -273,15 +338,11 @@ IsValidJsonNumber(const char *str, size_t len)
 {
 	bool		numeric_error;
 	size_t		total_len;
-	JsonLexContext dummy_lex;
+	JsonLexContext dummy_lex = {0};

 	if (len <= 0)
 		return false;

-	dummy_lex.incremental = false;
-	dummy_lex.inc_state = NULL;
-	dummy_lex.pstack = NULL;
-
 	/*
 	 * json_lex_number expects a leading  '-' to have been eaten already.
 	 *
@@ -321,6 +382,9 @@ IsValidJsonNumber(const char *str, size_t len)
 * responsible for freeing the returned struct, either by calling
 * freeJsonLexContext() or (in backend environment) via memory context
 * cleanup.
+ *
+ * In shlib code, any out-of-memory failures will be deferred to time
+ * of use; this function is guaranteed to return a valid JsonLexContext.
 */
 JsonLexContext *
 makeJsonLexContextCstringLen(JsonLexContext *lex, const char *json,
@@ -328,7 +392,9 @@ makeJsonLexContextCstringLen(JsonLexContext *lex, const char *json,
 {
 	if (lex == NULL)
 	{
-		lex = palloc0(sizeof(JsonLexContext));
+		lex = ALLOC0(sizeof(JsonLexContext));
+		if (!lex)
+			return &failed_oom;
 		lex->flags |= JSONLEX_FREE_STRUCT;
 	}
 	else
@@ -339,15 +405,73 @@ makeJsonLexContextCstringLen(JsonLexContext *lex, const char *json,
 	lex->line_number = 1;
 	lex->input_length = len;
 	lex->input_encoding = encoding;
+	lex->need_escapes = need_escapes;
 	if (need_escapes)
 	{
-		lex->strval = makeStringInfo();
+		/*
+		 * This call can fail in shlib code. We defer error handling to time
+		 * of use (json_lex_string()) since we might not need to parse any
+		 * strings anyway.
+		 */
+		lex->strval = jsonapi_makeStringInfo();
 		lex->flags |= JSONLEX_FREE_STRVAL;
 	}

 	return lex;
 }

+/*
+ * Allocates the internal bookkeeping structures for incremental parsing. This
+ * can only fail in-band with shlib code.
+ */
+#define JS_STACK_CHUNK_SIZE 64
+#define JS_MAX_PROD_LEN 10		/* more than we need */
+#define JSON_TD_MAX_STACK 6400	/* hard coded for now - this is a REALLY high
+								 * number */
+static bool
+allocate_incremental_state(JsonLexContext *lex)
+{
+	void	   *pstack,
+			   *prediction,
+			   *fnames,
+			   *fnull;
+
+	lex->inc_state = ALLOC0(sizeof(JsonIncrementalState));
+	pstack = ALLOC(sizeof(JsonParserStack));
+	prediction = ALLOC(JS_STACK_CHUNK_SIZE * JS_MAX_PROD_LEN);
+	fnames = ALLOC(JS_STACK_CHUNK_SIZE * sizeof(char *));
+	fnull = ALLOC(JS_STACK_CHUNK_SIZE * sizeof(bool));
+
+#ifdef JSONAPI_USE_PQEXPBUFFER
+	if (!lex->inc_state
+		|| !pstack
+		|| !prediction
+		|| !fnames
+		|| !fnull)
+	{
+		FREE(lex->inc_state);
+		FREE(pstack);
+		FREE(prediction);
+		FREE(fnames);
+		FREE(fnull);
+
+		lex->inc_state = &failed_inc_oom;
+		return false;
+	}
+#endif
+
+	jsonapi_initStringInfo(&(lex->inc_state->partial_token));
+	lex->pstack = pstack;
+	lex->pstack->stack_size = JS_STACK_CHUNK_SIZE;
+	lex->pstack->prediction = prediction;
+	lex->pstack->pred_index = 0;
+	lex->pstack->fnames = fnames;
+	lex->pstack->fnull = fnull;
+
+	lex->incremental = true;
+	return true;
+}
+

 /*
 * makeJsonLexContextIncremental
@@ -357,19 +481,20 @@ makeJsonLexContextCstringLen(JsonLexContext *lex, const char *json,
 * we don't need the input, that will be handed in bit by bit to the
 * parse routine. We also need an accumulator for partial tokens in case
 * the boundary between chunks happens to fall in the middle of a token.
+ *
+ * In shlib code, any out-of-memory failures will be deferred to time of use;
+ * this function is guaranteed to return a valid JsonLexContext.
 */
-#define JS_STACK_CHUNK_SIZE 64
-#define JS_MAX_PROD_LEN 10		/* more than we need */
-#define JSON_TD_MAX_STACK 6400	/* hard coded for now - this is a REALLY high
-								 * number */
-
 JsonLexContext *
 makeJsonLexContextIncremental(JsonLexContext *lex, int encoding,
 							  bool need_escapes)
 {
 	if (lex == NULL)
 	{
-		lex = palloc0(sizeof(JsonLexContext));
+		lex = ALLOC0(sizeof(JsonLexContext));
+		if (!lex)
+			return &failed_oom;
+
 		lex->flags |= JSONLEX_FREE_STRUCT;
 	}
 	else
@@ -377,42 +502,65 @@ makeJsonLexContextIncremental(JsonLexContext *lex, int encoding,

 	lex->line_number = 1;
 	lex->input_encoding = encoding;
-	lex->incremental = true;
-	lex->inc_state = palloc0(sizeof(JsonIncrementalState));
-	initStringInfo(&(lex->inc_state->partial_token));
-	lex->pstack = palloc(sizeof(JsonParserStack));
-	lex->pstack->stack_size = JS_STACK_CHUNK_SIZE;
-	lex->pstack->prediction = palloc(JS_STACK_CHUNK_SIZE * JS_MAX_PROD_LEN);
-	lex->pstack->pred_index = 0;
-	lex->pstack->fnames = palloc(JS_STACK_CHUNK_SIZE * sizeof(char *));
-	lex->pstack->fnull = palloc(JS_STACK_CHUNK_SIZE * sizeof(bool));
+
+	if (!allocate_incremental_state(lex))
+	{
+		if (lex->flags & JSONLEX_FREE_STRUCT)
+		{
+			FREE(lex);
+			return &failed_oom;
+		}
+
+		/* lex->inc_state tracks the OOM failure; we can return here. */
+		return lex;
+	}
+
+	lex->need_escapes = need_escapes;
 	if (need_escapes)
 	{
-		lex->strval = makeStringInfo();
+		/*
+		 * This call can fail in shlib code. We defer error handling to time
+		 * of use (json_lex_string()) since we might not need to parse any
+		 * strings anyway.
+		 */
+		lex->strval = jsonapi_makeStringInfo();
 		lex->flags |= JSONLEX_FREE_STRVAL;
 	}
+
 	return lex;
 }

-static inline void
+static inline bool
 inc_lex_level(JsonLexContext *lex)
 {
-	lex->lex_level += 1;
-
-	if (lex->incremental && lex->lex_level >= lex->pstack->stack_size)
+	if (lex->incremental && (lex->lex_level + 1) >= lex->pstack->stack_size)
 	{
-		lex->pstack->stack_size += JS_STACK_CHUNK_SIZE;
-		lex->pstack->prediction =
-			repalloc(lex->pstack->prediction,
-					 lex->pstack->stack_size * JS_MAX_PROD_LEN);
-		if (lex->pstack->fnames)
-			lex->pstack->fnames =
-				repalloc(lex->pstack->fnames,
-						 lex->pstack->stack_size * sizeof(char *));
-		if (lex->pstack->fnull)
-			lex->pstack->fnull =
-				repalloc(lex->pstack->fnull, lex->pstack->stack_size * sizeof(bool));
+		size_t		new_stack_size;
+		char	   *new_prediction;
+		char	  **new_fnames;
+		bool	   *new_fnull;
+
+		new_stack_size = lex->pstack->stack_size + JS_STACK_CHUNK_SIZE;
+
+		new_prediction = REALLOC(lex->pstack->prediction,
+								 new_stack_size * JS_MAX_PROD_LEN);
+		new_fnames = REALLOC(lex->pstack->fnames,
+							 new_stack_size * sizeof(char *));
+		new_fnull = REALLOC(lex->pstack->fnull, new_stack_size * sizeof(bool));
+
+#ifdef JSONAPI_USE_PQEXPBUFFER
+		if (!new_prediction || !new_fnames || !new_fnull)
+			return false;
+#endif
+
+		lex->pstack->stack_size = new_stack_size;
+		lex->pstack->prediction = new_prediction;
+		lex->pstack->fnames = new_fnames;
+		lex->pstack->fnull = new_fnull;
 	}
+
+	lex->lex_level += 1;
+	return true;
 }

 static inline void
@@ -482,24 +630,31 @@ get_fnull(JsonLexContext *lex)
 void
 freeJsonLexContext(JsonLexContext *lex)
 {
+	static const JsonLexContext empty = {0};
+
+	if (!lex || lex == &failed_oom)
+		return;
+
 	if (lex->flags & JSONLEX_FREE_STRVAL)
-		destroyStringInfo(lex->strval);
+		jsonapi_destroyStringInfo(lex->strval);

 	if (lex->errormsg)
-		destroyStringInfo(lex->errormsg);
+		jsonapi_destroyStringInfo(lex->errormsg);

 	if (lex->incremental)
 	{
-		pfree(lex->inc_state->partial_token.data);
-		pfree(lex->inc_state);
-		pfree(lex->pstack->prediction);
-		pfree(lex->pstack->fnames);
-		pfree(lex->pstack->fnull);
-		pfree(lex->pstack);
+		jsonapi_termStringInfo(&lex->inc_state->partial_token);
+		FREE(lex->inc_state);
+		FREE(lex->pstack->prediction);
+		FREE(lex->pstack->fnames);
+		FREE(lex->pstack->fnull);
+		FREE(lex->pstack);
 	}

 	if (lex->flags & JSONLEX_FREE_STRUCT)
-		pfree(lex);
+		FREE(lex);
+	else
+		*lex = empty;
 }

 /*
@@ -522,22 +677,13 @@ JsonParseErrorType
 pg_parse_json(JsonLexContext *lex, const JsonSemAction *sem)
 {
 #ifdef FORCE_JSON_PSTACK
-
-	lex->incremental = true;
-	lex->inc_state = palloc0(sizeof(JsonIncrementalState));
-
 	/*
 	 * We don't need partial token processing, there is only one chunk. But we
 	 * still need to init the partial token string so that freeJsonLexContext
-	 * works.
+	 * works, so perform the full incremental initialization.
 	 */
-	initStringInfo(&(lex->inc_state->partial_token));
-	lex->pstack = palloc(sizeof(JsonParserStack));
-	lex->pstack->stack_size = JS_STACK_CHUNK_SIZE;
-	lex->pstack->prediction = palloc(JS_STACK_CHUNK_SIZE * JS_MAX_PROD_LEN);
-	lex->pstack->pred_index = 0;
-	lex->pstack->fnames = palloc(JS_STACK_CHUNK_SIZE * sizeof(char *));
-	lex->pstack->fnull = palloc(JS_STACK_CHUNK_SIZE * sizeof(bool));
+	if (!allocate_incremental_state(lex))
+		return JSON_OUT_OF_MEMORY;

 	return pg_parse_json_incremental(lex, sem, lex->input, lex->input_length, true);

@@ -546,6 +692,8 @@ pg_parse_json(JsonLexContext *lex, const JsonSemAction *sem)
 	JsonTokenType tok;
 	JsonParseErrorType result;

+	if (lex == &failed_oom)
+		return JSON_OUT_OF_MEMORY;
 	if (lex->incremental)
 		return JSON_INVALID_LEXER_TYPE;

@@ -591,13 +739,16 @@ json_count_array_elements(JsonLexContext *lex, int *elements)
 	int			count;
 	JsonParseErrorType result;

+	if (lex == &failed_oom)
+		return JSON_OUT_OF_MEMORY;
+
 	/*
 	 * It's safe to do this with a shallow copy because the lexical routines
 	 * don't scribble on the input. They do scribble on the other pointers
 	 * etc, so doing this with a copy makes that safe.
 	 */
 	memcpy(&copylex, lex, sizeof(JsonLexContext));
-	copylex.strval = NULL;		/* not interested in values here */
+	copylex.need_escapes = false;	/* not interested in values here */
 	copylex.lex_level++;

 	count = 0;
@@ -658,7 +809,8 @@ pg_parse_json_incremental(JsonLexContext *lex,
 	JsonParseContext ctx = JSON_PARSE_VALUE;
 	JsonParserStack *pstack = lex->pstack;

-
+	if (lex == &failed_oom || lex->inc_state == &failed_inc_oom)
+		return JSON_OUT_OF_MEMORY;
 	if (!lex->incremental)
 		return JSON_INVALID_LEXER_TYPE;

@@ -737,7 +889,9 @@ pg_parse_json_incremental(JsonLexContext *lex,
 							if (result != JSON_SUCCESS)
 								return result;
 						}
-						inc_lex_level(lex);
+
+						if (!inc_lex_level(lex))
+							return JSON_OUT_OF_MEMORY;
 					}
 					break;
 				case JSON_SEM_OEND:
@@ -766,7 +920,9 @@ pg_parse_json_incremental(JsonLexContext *lex,
 							if (result != JSON_SUCCESS)
 								return result;
 						}
-						inc_lex_level(lex);
+
+						if (!inc_lex_level(lex))
+							return JSON_OUT_OF_MEMORY;
 					}
 					break;
 				case JSON_SEM_AEND:
@@ -793,9 +949,11 @@ pg_parse_json_incremental(JsonLexContext *lex,
 						json_ofield_action ostart = sem->object_field_start;
 						json_ofield_action oend = sem->object_field_end;

-						if ((ostart != NULL || oend != NULL) && lex->strval != NULL)
+						if ((ostart != NULL || oend != NULL) && lex->need_escapes)
 						{
-							fname = pstrdup(lex->strval->data);
+							fname = STRDUP(lex->strval->data);
+							if (fname == NULL)
+								return JSON_OUT_OF_MEMORY;
 						}
 						set_fname(lex, fname);
 					}
@@ -883,14 +1041,21 @@ pg_parse_json_incremental(JsonLexContext *lex,
 							 */
 							if (tok == JSON_TOKEN_STRING)
 							{
-								if (lex->strval != NULL)
-									pstack->scalar_val = pstrdup(lex->strval->data);
+								if (lex->need_escapes)
+								{
+									pstack->scalar_val = STRDUP(lex->strval->data);
+									if (pstack->scalar_val == NULL)
+										return JSON_OUT_OF_MEMORY;
+								}
 							}
 							else
 							{
 								ptrdiff_t	tlen = (lex->token_terminator - lex->token_start);

-								pstack->scalar_val = palloc(tlen + 1);
+								pstack->scalar_val = ALLOC(tlen + 1);
+								if (pstack->scalar_val == NULL)
+									return JSON_OUT_OF_MEMORY;
+
 								memcpy(pstack->scalar_val, lex->token_start, tlen);
 								pstack->scalar_val[tlen] = '\0';
 							}
@@ -1025,14 +1190,21 @@ parse_scalar(JsonLexContext *lex, const JsonSemAction *sem)
 	/* extract the de-escaped string value, or the raw lexeme */
 	if (lex_peek(lex) == JSON_TOKEN_STRING)
 	{
-		if (lex->strval != NULL)
-			val = pstrdup(lex->strval->data);
+		if (lex->need_escapes)
+		{
+			val = STRDUP(lex->strval->data);
+			if (val == NULL)
+				return JSON_OUT_OF_MEMORY;
+		}
 	}
 	else
 	{
 		int			len = (lex->token_terminator - lex->token_start);

-		val = palloc(len + 1);
+		val = ALLOC(len + 1);
+		if (val == NULL)
+			return JSON_OUT_OF_MEMORY;
+
 		memcpy(val, lex->token_start, len);
 		val[len] = '\0';
 	}
@@ -1066,8 +1238,12 @@ parse_object_field(JsonLexContext *lex, const JsonSemAction *sem)

 	if (lex_peek(lex) != JSON_TOKEN_STRING)
 		return report_parse_error(JSON_PARSE_STRING, lex);
-	if ((ostart != NULL || oend != NULL) && lex->strval != NULL)
-		fname = pstrdup(lex->strval->data);
+	if ((ostart != NULL || oend != NULL) && lex->need_escapes)
+	{
+		fname = STRDUP(lex->strval->data);
+		if (fname == NULL)
+			return JSON_OUT_OF_MEMORY;
+	}
 	result = json_lex(lex);
 	if (result != JSON_SUCCESS)
 		return result;
@@ -1123,6 +1299,11 @@ parse_object(JsonLexContext *lex, const JsonSemAction *sem)
 	JsonParseErrorType result;

 #ifndef FRONTEND
+
+	/*
+	 * TODO: clients need some way to put a bound on stack growth. Parse level
+	 * limits maybe?
+	 */
 	check_stack_depth();
 #endif

@@ -1312,15 +1493,27 @@ json_lex(JsonLexContext *lex)
 	const char *const end = lex->input + lex->input_length;
 	JsonParseErrorType result;

-	if (lex->incremental && lex->inc_state->partial_completed)
+	if (lex == &failed_oom || lex->inc_state == &failed_inc_oom)
+		return JSON_OUT_OF_MEMORY;
+
+	if (lex->incremental)
 	{
-		/*
-		 * We just lexed a completed partial token on the last call, so reset
-		 * everything
-		 */
-		resetStringInfo(&(lex->inc_state->partial_token));
-		lex->token_terminator = lex->input;
-		lex->inc_state->partial_completed = false;
+		if (lex->inc_state->partial_completed)
+		{
+			/*
+			 * We just lexed a completed partial token on the last call, so
+			 * reset everything
+			 */
+			jsonapi_resetStringInfo(&(lex->inc_state->partial_token));
+			lex->token_terminator = lex->input;
+			lex->inc_state->partial_completed = false;
+		}
+
+#ifdef JSONAPI_USE_PQEXPBUFFER
+		/* Make sure our partial token buffer is valid before using it below. */
+		if (PQExpBufferDataBroken(lex->inc_state->partial_token))
+			return JSON_OUT_OF_MEMORY;
+#endif
 	}

 	s = lex->token_terminator;
@@ -1331,7 +1524,7 @@ json_lex(JsonLexContext *lex)
 		 * We have a partial token. Extend it and if completed lex it by a
 		 * recursive call
 		 */
-		StringInfo	ptok = &(lex->inc_state->partial_token);
+		jsonapi_StrValType *ptok = &(lex->inc_state->partial_token);
 		size_t		added = 0;
 		bool		tok_done = false;
 		JsonLexContext dummy_lex;
@@ -1358,7 +1551,7 @@ json_lex(JsonLexContext *lex)
 			{
 				char		c = lex->input[i];

-				appendStringInfoCharMacro(ptok, c);
+				jsonapi_appendStringInfoCharMacro(ptok, c);
 				added++;
 				if (c == '"' && escapes % 2 == 0)
 				{
@@ -1403,7 +1596,7 @@ json_lex(JsonLexContext *lex)
 						case '8':
 						case '9':
 							{
-								appendStringInfoCharMacro(ptok, cc);
+								jsonapi_appendStringInfoCharMacro(ptok, cc);
 								added++;
 							}
 							break;
@@ -1424,7 +1617,7 @@ json_lex(JsonLexContext *lex)

 				if (JSON_ALPHANUMERIC_CHAR(cc))
 				{
-					appendStringInfoCharMacro(ptok, cc);
+					jsonapi_appendStringInfoCharMacro(ptok, cc);
 					added++;
 				}
 				else
@@ -1467,6 +1660,7 @@ json_lex(JsonLexContext *lex)
 		dummy_lex.input_length = ptok->len;
 		dummy_lex.input_encoding = lex->input_encoding;
 		dummy_lex.incremental = false;
+		dummy_lex.need_escapes = lex->need_escapes;
 		dummy_lex.strval = lex->strval;

 		partial_result = json_lex(&dummy_lex);
@@ -1622,8 +1816,7 @@ json_lex(JsonLexContext *lex)
 					if (lex->incremental && !lex->inc_state->is_last_chunk &&
 						p == lex->input + lex->input_length)
 					{
-						appendBinaryStringInfo(
-											   &(lex->inc_state->partial_token), s, end - s);
+						jsonapi_appendBinaryStringInfo(&(lex->inc_state->partial_token), s, end - s);
 						return JSON_INCOMPLETE;
 					}

@@ -1680,8 +1873,9 @@ json_lex_string(JsonLexContext *lex)
 	do { \
 		if (lex->incremental && !lex->inc_state->is_last_chunk) \
 		{ \
-			appendBinaryStringInfo(&lex->inc_state->partial_token, \
-								   lex->token_start, end - lex->token_start); \
+			jsonapi_appendBinaryStringInfo(&lex->inc_state->partial_token, \
+										   lex->token_start, \
+										   end - lex->token_start); \
 			return JSON_INCOMPLETE; \
 		} \
 		lex->token_terminator = s; \
@@ -1694,8 +1888,15 @@ json_lex_string(JsonLexContext *lex)
 		return code; \
 	} while (0)

-	if (lex->strval != NULL)
-		resetStringInfo(lex->strval);
+	if (lex->need_escapes)
+	{
+#ifdef JSONAPI_USE_PQEXPBUFFER
+		/* make sure initialization succeeded */
+		if (lex->strval == NULL)
+			return JSON_OUT_OF_MEMORY;
+#endif
+		jsonapi_resetStringInfo(lex->strval);
+	}

 	Assert(lex->input_length > 0);
 	s = lex->token_start;
@@ -1732,7 +1933,7 @@ json_lex_string(JsonLexContext *lex)
 					else
 						FAIL_AT_CHAR_END(JSON_UNICODE_ESCAPE_FORMAT);
 				}
-				if (lex->strval != NULL)
+				if (lex->need_escapes)
 				{
 					/*
 					 * Combine surrogate pairs.
@@ -1789,19 +1990,19 @@ json_lex_string(JsonLexContext *lex)

 						unicode_to_utf8(ch, (unsigned char *) utf8str);
 						utf8len = pg_utf_mblen((unsigned char *) utf8str);
-						appendBinaryStringInfo(lex->strval, utf8str, utf8len);
+						jsonapi_appendBinaryStringInfo(lex->strval, utf8str, utf8len);
 					}
 					else if (ch <= 0x007f)
 					{
 						/* The ASCII range is the same in all encodings */
-						appendStringInfoChar(lex->strval, (char) ch);
+						jsonapi_appendStringInfoChar(lex->strval, (char) ch);
 					}
 					else
 						FAIL_AT_CHAR_END(JSON_UNICODE_HIGH_ESCAPE);
 #endif							/* FRONTEND */
 				}
 			}
-			else if (lex->strval != NULL)
+			else if (lex->need_escapes)
 			{
 				if (hi_surrogate != -1)
 					FAIL_AT_CHAR_END(JSON_UNICODE_LOW_SURROGATE);
@@ -1811,22 +2012,22 @@ json_lex_string(JsonLexContext *lex)
 					case '"':
 					case '\\':
 					case '/':
-						appendStringInfoChar(lex->strval, *s);
+						jsonapi_appendStringInfoChar(lex->strval, *s);
 						break;
 					case 'b':
-						appendStringInfoChar(lex->strval, '\b');
+						jsonapi_appendStringInfoChar(lex->strval, '\b');
 						break;
 					case 'f':
-						appendStringInfoChar(lex->strval, '\f');
+						jsonapi_appendStringInfoChar(lex->strval, '\f');
 						break;
 					case 'n':
-						appendStringInfoChar(lex->strval, '\n');
+						jsonapi_appendStringInfoChar(lex->strval, '\n');
 						break;
 					case 'r':
-						appendStringInfoChar(lex->strval, '\r');
+						jsonapi_appendStringInfoChar(lex->strval, '\r');
 						break;
 					case 't':
-						appendStringInfoChar(lex->strval, '\t');
+						jsonapi_appendStringInfoChar(lex->strval, '\t');
 						break;
 					default:

@@ -1861,7 +2062,7 @@ json_lex_string(JsonLexContext *lex)

 			/*
 			 * Skip to the first byte that requires special handling, so we
-			 * can batch calls to appendBinaryStringInfo.
+			 * can batch calls to jsonapi_appendBinaryStringInfo.
 			 */
 			while (p < end - sizeof(Vector8) &&
 				   !pg_lfind8('\\', (uint8 *) p, sizeof(Vector8)) &&
@@ -1885,8 +2086,8 @@ json_lex_string(JsonLexContext *lex)
 				}
 			}

-			if (lex->strval != NULL)
-				appendBinaryStringInfo(lex->strval, s, p - s);
+			if (lex->need_escapes)
+				jsonapi_appendBinaryStringInfo(lex->strval, s, p - s);

 			/*
 			 * s will be incremented at the top of the loop, so set it to just
@@ -1902,6 +2103,11 @@ json_lex_string(JsonLexContext *lex)
 		return JSON_UNICODE_LOW_SURROGATE;
 	}

+#ifdef JSONAPI_USE_PQEXPBUFFER
+	if (lex->need_escapes && PQExpBufferBroken(lex->strval))
+		return JSON_OUT_OF_MEMORY;
+#endif
+
 	/* Hooray, we found the end of the string! */
 	lex->prev_token_terminator = lex->token_terminator;
 	lex->token_terminator = s + 1;
@@ -2019,8 +2225,8 @@ json_lex_number(JsonLexContext *lex, const char *s,
 	if (lex->incremental && !lex->inc_state->is_last_chunk &&
 		len >= lex->input_length)
 	{
-		appendBinaryStringInfo(&lex->inc_state->partial_token,
-							   lex->token_start, s - lex->token_start);
+		jsonapi_appendBinaryStringInfo(&lex->inc_state->partial_token,
+									   lex->token_start, s - lex->token_start);
 		if (num_err != NULL)
 			*num_err = error;

@@ -2096,19 +2302,25 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
 char *
 json_errdetail(JsonParseErrorType error, JsonLexContext *lex)
 {
+	if (error == JSON_OUT_OF_MEMORY || lex == &failed_oom)
+	{
+		/* Short circuit. Allocating anything for this case is unhelpful. */
+		return _("out of memory");
+	}
+
 	if (lex->errormsg)
-		resetStringInfo(lex->errormsg);
+		jsonapi_resetStringInfo(lex->errormsg);
 	else
-		lex->errormsg = makeStringInfo();
+		lex->errormsg = jsonapi_makeStringInfo();

 	/*
 	 * A helper for error messages that should print the current token. The
 	 * format must contain exactly one %.*s specifier.
 	 */
 #define json_token_error(lex, format) \
-	appendStringInfo((lex)->errormsg, _(format), \
-					 (int) ((lex)->token_terminator - (lex)->token_start), \
-					 (lex)->token_start);
+	jsonapi_appendStringInfo((lex)->errormsg, _(format), \
+							 (int) ((lex)->token_terminator - (lex)->token_start), \
+							 (lex)->token_start);

 	switch (error)
 	{
@@ -2127,9 +2339,9 @@ json_errdetail(JsonParseErrorType error, JsonLexContext *lex)
 			json_token_error(lex, "Escape sequence \"\\%.*s\" is invalid.");
 			break;
 		case JSON_ESCAPING_REQUIRED:
-			appendStringInfo(lex->errormsg,
-							 _("Character with value 0x%02x must be escaped."),
-							 (unsigned char) *(lex->token_terminator));
+			jsonapi_appendStringInfo(lex->errormsg,
+									 _("Character with value 0x%02x must be escaped."),
+									 (unsigned char) *(lex->token_terminator));
 			break;
 		case JSON_EXPECTED_END:
 			json_token_error(lex, "Expected end of input, but found \"%.*s\".");
@@ -2160,6 +2372,9 @@ json_errdetail(JsonParseErrorType error, JsonLexContext *lex)
 		case JSON_INVALID_TOKEN:
 			json_token_error(lex, "Token \"%.*s\" is invalid.");
 			break;
+		case JSON_OUT_OF_MEMORY:
+			/* should have been handled above; use the error path */
+			break;
 		case JSON_UNICODE_CODE_POINT_ZERO:
 			return _("\\u0000 cannot be converted to text.");
 		case JSON_UNICODE_ESCAPE_FORMAT:
@@ -2191,15 +2406,23 @@ json_errdetail(JsonParseErrorType error, JsonLexContext *lex)
 	}
 #undef json_token_error

-	/*
-	 * We don't use a default: case, so that the compiler will warn about
-	 * unhandled enum values.  But this needs to be here anyway to cover the
-	 * possibility of an incorrect input.
-	 */
-	if (lex->errormsg->len == 0)
-		appendStringInfo(lex->errormsg,
-						 "unexpected json parse error type: %d",
-						 (int) error);
+	/* Note that lex->errormsg can be NULL in shlib code. */
+	if (lex->errormsg && lex->errormsg->len == 0)
+	{
+		/*
+		 * We don't use a default: case, so that the compiler will warn about
+		 * unhandled enum values.  But this needs to be here anyway to cover
+		 * the possibility of an incorrect input.
+		 */
+		jsonapi_appendStringInfo(lex->errormsg,
+								 "unexpected json parse error type: %d",
+								 (int) error);
+	}
+
+#ifdef JSONAPI_USE_PQEXPBUFFER
+	if (PQExpBufferBroken(lex->errormsg))
+		return _("out of memory while constructing error description");
+#endif

 	return lex->errormsg->data;
 }