Introduce a non-recursive JSON parser

This parser uses an explicit prediction stack, unlike the present
recursive descent parser where the parser state is represented on the
call stack. This difference makes the new parser suitable for use in
incremental parsing of huge JSON documents that cannot be conveniently
handled piece-wise by the recursive descent parser. One potential use
for this will be in parsing large backup manifests associated with
incremental backups.

Because this parser is somewhat slower than the recursive descent
parser, it  is not replacing that parser, but is an additional parser
available to callers.

For testing purposes, if the build is done with -DFORCE_JSON_PSTACK, all
JSON parsing is done with the non-recursive parser, in which case only
trivial regression differences in error messages should be observed.

Author: Andrew Dunstan
Reviewed-By: Jacob Champion

Discussion: https://postgr.es/m/7b0a51d6-0d9d-7366-3a1a-f74397a02f55@dunslane.net

Author: adunstan

src/common/jsonapi.c

@@ -36,6 +36,9 @@ typedef enum JsonTokenType
 typedef enum JsonParseErrorType
 {
 	JSON_SUCCESS,
+	JSON_INCOMPLETE,
+	JSON_INVALID_LEXER_TYPE,
+	JSON_NESTING_TOO_DEEP,
 	JSON_ESCAPING_INVALID,
 	JSON_ESCAPING_REQUIRED,
 	JSON_EXPECTED_ARRAY_FIRST,
@@ -57,6 +60,9 @@ typedef enum JsonParseErrorType
 	JSON_SEM_ACTION_FAILED,		/* error should already be reported */
 } JsonParseErrorType;
 
+/* Parser state private to jsonapi.c */
+typedef struct JsonParserStack JsonParserStack;
+typedef struct JsonIncrementalState JsonIncrementalState;
 
 /*
  * All the fields in this structure should be treated as read-only.
@@ -71,6 +77,11 @@ typedef enum JsonParseErrorType
  * AFTER the end of the token, i.e. where there would be a nul byte
  * if we were using nul-terminated strings.
  *
+ * The prev_token_terminator field should not be used when incremental is
+ * true, as the previous token might have started in a previous piece of input,
+ * and thus it can't be used in any pointer arithmetic or other operations in
+ * conjunction with token_start.
+ *
  * JSONLEX_FREE_STRUCT/STRVAL are used to drive freeJsonLexContext.
  */
 #define JSONLEX_FREE_STRUCT			(1 << 0)
@@ -83,11 +94,14 @@ typedef struct JsonLexContext
 	char	   *token_start;
 	char	   *token_terminator;
 	char	   *prev_token_terminator;
+	bool		incremental;
 	JsonTokenType token_type;
 	int			lex_level;
 	bits32		flags;
 	int			line_number;	/* line number, starting from 1 */
 	char	   *line_start;		/* where that line starts within input */
+	JsonParserStack *pstack;
+	JsonIncrementalState *inc_state;
 	StringInfo	strval;
 	StringInfo	errormsg;
 } JsonLexContext;
@@ -141,6 +155,12 @@ typedef struct JsonSemAction
 extern JsonParseErrorType pg_parse_json(JsonLexContext *lex,
 										JsonSemAction *sem);
 
+extern JsonParseErrorType pg_parse_json_incremental(JsonLexContext *lex,
+													JsonSemAction *sem,
+													char *json,
+													int len,
+													bool is_last);
+
 /* the null action object used for pure validation */
 extern PGDLLIMPORT JsonSemAction nullSemAction;
 
@@ -176,6 +196,16 @@ extern JsonLexContext *makeJsonLexContextCstringLen(JsonLexContext *lex,
 													int len,
 													int encoding,
 													bool need_escapes);
+
+/*
+ * make a JsonLexContext suitable for incremental parsing.
+ * the string chunks will be handed to pg_parse_json_incremental,
+ * so there's no need for them here.
+ */
+extern JsonLexContext *makeJsonLexContextIncremental(JsonLexContext *lex,
+													 int encoding,
+													 bool need_escapes);
+
 extern void freeJsonLexContext(JsonLexContext *lex);
 
 /* lex one token */

src/include/common/jsonapi.h

@@ -240,6 +240,13 @@
  *------------------------------------------------------------------------
  */
 
+/*
+ * Force use of the non-recursive JSON parser in all cases. This is useful
+ * to validate the working of the parser, and the regression tests should
+ * pass except for some different error messages about the stack limit.
+ */
+/* #define FORCE_JSON_PSTACK */
+
 /*
  * Include Valgrind "client requests", mostly in the memory allocator, so
  * Valgrind understands PostgreSQL memory contexts.  This permits detecting

src/include/pg_config_manual.h

@@ -22,6 +22,7 @@ SUBDIRS = \
 		  test_extensions \
 		  test_ginpostinglist \
 		  test_integerset \
+		  test_json_parser \
 		  test_lfind \
 		  test_misc \
 		  test_oat_hooks \

src/test/modules/Makefile

@@ -21,6 +21,7 @@ subdir('test_dsm_registry')
 subdir('test_extensions')
 subdir('test_ginpostinglist')
 subdir('test_integerset')
+subdir('test_json_parser')
 subdir('test_lfind')
 subdir('test_misc')
 subdir('test_oat_hooks')

src/test/modules/meson.build

@@ -0,0 +1,36 @@
+
+PGFILEDESC = "standalone json parser tester"
+PGAPPICON = win32
+
+TAP_TESTS = 1
+
+OBJS = test_json_parser_incremental.o test_json_parser_perf.o
+
+ifdef USE_PGXS
+PG_CONFIG = pg_config
+PGXS := $(shell $(PG_CONFIG) --pgxs)
+include $(PGXS)
+else
+subdir = src/test/modules/test_json_parser
+top_builddir = ../../../..
+include $(top_builddir)/src/Makefile.global
+include $(top_srcdir)/contrib/contrib-global.mk
+endif
+
+all: test_json_parser_incremental$(X) test_json_parser_perf$(X)
+
+%.o: $(top_srcdir)/$(subdir)/%.c
+
+PARSER_LIBS = $(top_builddir)/src/common/libpgcommon.a $(top_builddir)/src/port/libpgport.a
+
+test_json_parser_incremental$(X): test_json_parser_incremental.o $(PARSER_LIBS)
+	$(CC) $(CFLAGS) $^ -o $@
+
+test_json_parser_perf$(X): test_json_parser_perf.o $(PARSER_LIBS)
+	$(CC) $(CFLAGS) $^ -o $@
+
+speed-check: test_json_parser_perf$(X)
+	@echo Standard parser:
+	time ./test_json_parser_perf 10000 $(top_srcdir)/$(subdir)/tiny.json
+	@echo Incremental parser:
+	time ./test_json_parser_perf -i 10000 $(top_srcdir)/$(subdir)/tiny.json

src/test/modules/test_json_parser/Makefile

@@ -0,0 +1,25 @@
+Module `test_json_parser`
+=========================
+
+This module contains two programs for testing the json parsers.
+
+- `test_json_parser_incremental` is for testing the incremental parser, It
+  reads in a file and pases it in very small chunks (60 bytes at a time) to
+  the incremental parser. It's not meant to be a speed test but to test the
+  accuracy of the incremental parser. It takes one argument: the name of the
+  input file.
+- `test_json_parser_perf` is for speed testing both the standard
+  recursive descent parser and the non-recursive incremental
+  parser. If given the `-i` flag it uses the non-recursive parser,
+  otherwise the stardard parser. The remaining flags are the number of
+  parsing iterations and the file containing the input. Even when
+  using the non-recursive parser, the input is passed to the parser in a
+  single chunk. The results are thus comparable to those of the
+  standard parser.
+
+The easiest way to use these is to run `make check` and `make speed-check`
+
+The sample input file is a small extract from a list of `delicious`
+bookmarks taken some years ago, all wrapped in a single json
+array. 10,000 iterations of parsing this file gives a reasonable
+benchmark, and that is what the `speed-check` target does.

src/test/modules/test_json_parser/README

@@ -0,0 +1,52 @@
+# Copyright (c) 2024, PostgreSQL Global Development Group
+
+test_json_parser_incremental_sources = files(
+  'test_json_parser_incremental.c',
+)
+
+if host_system == 'windows'
+  test_json_parser_incremental_sources += rc_bin_gen.process(win32ver_rc, extra_args: [
+    '--NAME', 'test_json_parser_incremental',
+    '--FILEDESC', 'standalone json parser tester',
+  ])
+endif
+
+test_json_parser_incremental = executable('test_json_parser_incremental',
+  test_json_parser_incremental_sources,
+  dependencies: [frontend_code],
+  kwargs: default_bin_args + {
+    'install': false,
+  },
+)
+
+test_json_parser_perf_sources = files(
+  'test_json_parser_perf.c',
+)
+
+if host_system == 'windows'
+  test_json_parser_perf_sources += rc_bin_gen.process(win32ver_rc, extra_args: [
+    '--NAME', 'test_json_parser_perf',
+    '--FILEDESC', 'standalone json parser tester',
+  ])
+endif
+
+test_json_parser_perf = executable('test_json_parser_perf',
+  test_json_parser_perf_sources,
+  dependencies: [frontend_code],
+  kwargs: default_bin_args + {
+    'install': false,
+  },
+)
+
+tests += {
+  'name': 'test_json_parser',
+  'sd': meson.current_source_dir(),
+  'bd': meson.current_build_dir(),
+  'tap': {
+    'tests': [
+      't/001_test_json_parser_incremental.pl',
+      't/002_inline.pl',
+      't/003_test_semantic.pl'
+    ],
+  },
+}

src/test/modules/test_json_parser/meson.build

@@ -0,0 +1,23 @@
+
+use strict;
+use warnings;
+
+use PostgreSQL::Test::Utils;
+use Test::More;
+use FindBin;
+
+use File::Temp qw(tempfile);
+
+my $test_file = "$FindBin::RealBin/../tiny.json";
+
+my $exe = "test_json_parser_incremental";
+
+for (my $size = 64; $size > 0; $size--)
+{
+	my ($stdout, $stderr) = run_command( [$exe, "-c", $size, $test_file] );
+
+	like($stdout, qr/SUCCESS/, "chunk size $size: test succeeds");
+	is($stderr, "", "chunk size $size: no error output");
+}
+
+done_testing();

src/test/modules/test_json_parser/t/001_test_json_parser_incremental.pl

@@ -0,0 +1,83 @@
+use strict;
+use warnings;
+
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+use File::Temp qw(tempfile);
+
+sub test
+{
+	local $Test::Builder::Level = $Test::Builder::Level + 1;
+
+	my ($name, $json, %params) = @_;
+	my $exe = "test_json_parser_incremental";
+	my $chunk = length($json);
+
+	if ($chunk > 64)
+	{
+		$chunk = 64;
+	}
+
+	my ($fh, $fname) = tempfile(UNLINK => 1);
+	print $fh "$json";
+	close($fh);
+
+	foreach my $size (reverse(1..$chunk))
+	{
+		my ($stdout, $stderr) = run_command( [$exe, "-c", $size, $fname] );
+
+		if (defined($params{error}))
+		{
+			unlike($stdout, qr/SUCCESS/, "$name, chunk size $size: test fails");
+			like($stderr, $params{error}, "$name, chunk size $size: correct error output");
+		}
+		else
+		{
+			like($stdout, qr/SUCCESS/, "$name, chunk size $size: test succeeds");
+			is($stderr, "", "$name, chunk size $size: no error output");
+		}
+	}
+}
+
+test("number", "12345");
+test("string", '"hello"');
+test("false", "false");
+test("true", "true");
+test("null", "null");
+test("empty object", "{}");
+test("empty array", "[]");
+test("array with number", "[12345]");
+test("array with numbers", "[12345,67890]");
+test("array with null", "[null]");
+test("array with string", '["hello"]');
+test("array with boolean", '[false]');
+test("single pair", '{"key": "value"}');
+test("heavily nested array", "[" x 3200 . "]" x 3200);
+test("serial escapes", '"\\\\\\\\\\\\\\\\"');
+test("interrupted escapes", '"\\\\\\"\\\\\\\\\\"\\\\"');
+test("whitespace", '     ""     ');
+
+test("unclosed empty object", "{", error => qr/input string ended unexpectedly/);
+test("bad key", "{{", error => qr/Expected string or "}", but found "\{"/);
+test("bad key", "{{}", error => qr/Expected string or "}", but found "\{"/);
+test("numeric key", "{1234: 2}", error => qr/Expected string or "}", but found "1234"/);
+test("second numeric key", '{"a": "a", 1234: 2}', error => qr/Expected string, but found "1234"/);
+test("unclosed object with pair", '{"key": "value"', error => qr/input string ended unexpectedly/);
+test("missing key value", '{"key": }', error => qr/Expected JSON value, but found "}"/);
+test("missing colon", '{"key" 12345}', error => qr/Expected ":", but found "12345"/);
+test("missing comma", '{"key": 12345 12345}', error => qr/Expected "," or "}", but found "12345"/);
+test("overnested array", "[" x 6401, error => qr/maximum permitted depth is 6400/);
+test("overclosed array", "[]]", error => qr/Expected end of input, but found "]"/);
+test("unexpected token in array", "[ }}} ]", error => qr/Expected array element or "]", but found "}"/);
+test("junk punctuation", "[ ||| ]", error => qr/Token "|" is invalid/);
+test("missing comma in array", "[123 123]", error => qr/Expected "," or "]", but found "123"/);
+test("misspelled boolean", "tru", error => qr/Token "tru" is invalid/);
+test("misspelled boolean in array", "[tru]", error => qr/Token "tru" is invalid/);
+test("smashed top-level scalar", "12zz", error => qr/Token "12zz" is invalid/);
+test("smashed scalar in array", "[12zz]", error => qr/Token "12zz" is invalid/);
+test("unknown escape sequence", '"hello\vworld"', error => qr/Escape sequence "\\v" is invalid/);
+test("unescaped control", "\"hello\tworld\"", error => qr/Character with value 0x09 must be escaped/);
+test("incorrect escape count", '"\\\\\\\\\\\\\\"', error => qr/Token ""\\\\\\\\\\\\\\"" is invalid/);
+
+done_testing();

src/test/modules/test_json_parser/t/002_inline.pl

@@ -0,0 +1,36 @@
+use strict;
+use warnings;
+
+use PostgreSQL::Test::Utils;
+use Test::More;
+use FindBin;
+
+use File::Temp qw(tempfile);
+
+my $test_file = "$FindBin::RealBin/../tiny.json";
+my $test_out = "$FindBin::RealBin/../tiny.out";
+
+my $exe = "test_json_parser_incremental";
+
+my ($stdout, $stderr) = run_command( [$exe, "-s", $test_file] );
+
+is($stderr, "", "no error output");
+
+my ($fh, $fname) = tempfile();
+
+print $fh $stdout,"\n";
+
+close($fh);
+
+($stdout, $stderr) = run_command(["diff", "-u", $fname, $test_out]);
+
+is($stdout, "", "no output diff");
+is($stderr, "", "no diff error");
+
+done_testing();
+
+
+
+
+
+