Add reactors to the JSON parser

Jay Berkenbilt
1 parent f5dd6381
Showing 17 changed files with 401 additions and 14 deletions
ChangeLog
include/qpdf/JSON.hh
libqpdf/JSON.cc
libtests/json_parse.cc
libtests/qtest/json_parse.test
libtests/qtest/json_parse/good-01-react.out
libtests/qtest/json_parse/good-02-react.out
libtests/qtest/json_parse/good-03-react.out
libtests/qtest/json_parse/good-04-react.out
libtests/qtest/json_parse/good-05-react.out
libtests/qtest/json_parse/good-06-react.out
libtests/qtest/json_parse/good-07-react.out
libtests/qtest/json_parse/good-08-react.out
libtests/qtest/json_parse/good-09-react.out
libtests/qtest/json_parse/good-10-react.out
libtests/qtest/json_parse/good-10.json
libtests/qtest/json_parse/save-10.json
+2022-05-01  Jay Berkenbilt  <ejb@ql.org>
+
+        * JSON: add reactors to the JSON parser, making it possible to
+        react to JSON parsing events as they occur and to block the
+        results from being stored. This makes it possible to incrementally
+        parse arbitrarily large JSON inputs.
+
 2022-04-30  Jay Berkenbilt  <ejb@ql.org>
         * QPDFWriter: change encryption API calls
@@ -141,9 +141,86 @@ class JSON
     QPDF_DLL
     bool checkSchema(JSON schema, std::list<std::string>& errors);
-    // Create a JSON object from a string.
+    // An pointer to a Reactor class can be passed to parse, which
+    // will enable the caller to react to incremental events in the
+    // construction of the JSON object. This makes it possible to
+    // implement SAX-like handling of very large JSON objects.
+    class QPDF_DLL_CLASS Reactor
+    {
+      public:
+        QPDF_DLL
+        virtual ~Reactor() = default;
+
+        // The start/end methods are called when parsing of a
+        // dictionary or array is started or ended. The item methods
+        // are called when an item is added to a dictionary or array.
+        // See important notes in "Item methods" below.
+
+        // During parsing of a JSON string, the parser is operating on
+        // a single object at a time. When a dictionary or array is
+        // started, a new context begins, and when that dictionary or
+        // array is ended, the previous context is resumed. So, for
+        // example, if you have `{"a": [1]}`, you will receive the
+        // following method calls
+        //
+        // dictionaryStart -- current object is the top-level dictionary
+        // arrayStart      -- current object is the array
+        // arrayItem       -- called with the "1" object
+        // containerEnd    -- now current object is the dictionary again
+        // dictionaryItem  -- called with "a" and the just-completed array
+        // containerEnd    -- current object is undefined
+        //
+        // If the top-level item in a JSON string is a scalar, the
+        // topLevelScalar() method will be called. No argument is
+        // passed since the object is the same as what is returned by
+        // parse().
+
+        QPDF_DLL
+        virtual void dictionaryStart() = 0;
+        QPDF_DLL
+        virtual void arrayStart() = 0;
+        QPDF_DLL
+        virtual void containerEnd(JSON const& value) = 0;
+        QPDF_DLL
+        virtual void topLevelScalar() = 0;
+
+        // Item methods:
+        //
+        // The return value of the item methods indicate whether the
+        // item has been "consumed". If the item method returns true,
+        // then the item will not be added to the containing JSON
+        // object. This is what allows arbitrarily large JSON objects
+        // to be parsed and not have to be kept in memory.
+        //
+        // NOTE: When a dictionary or an array is added to a
+        // container, the dictionaryItem or arrayItem method is called
+        // when the child item's start delimiter is encountered, so
+        // the JSON object passed in at that time will always be
+        // in its initial, empty state.
+
+        QPDF_DLL
+        virtual bool
+        dictionaryItem(std::string const& key, JSON const& value) = 0;
+        QPDF_DLL
+        virtual bool arrayItem(JSON const& value) = 0;
+    };
+
+    // Create a JSON object from a string. See above for information
+    // about how to use the Reactor.
+    QPDF_DLL
+    static JSON parse(std::string const&, Reactor* reactor = nullptr);
+
+    // parse calls setOffsets to set the inclusive start and
+    // non-inclusive end offsets of an object relative to its input
+    // string. Otherwise, both values are 0.
+    QPDF_DLL
+    void setStart(size_t);
+    QPDF_DLL
+    void setEnd(size_t);
+    QPDF_DLL
+    size_t getStart() const;
     QPDF_DLL
-    static JSON parse(std::string const&);
+    size_t getEnd() const;
   private:
     static std::string encode_string(std::string const& utf8);
@@ -217,6 +294,9 @@ class JSON
         Members(Members const&) = delete;
         std::shared_ptr<JSON_value> value;
+        // start and end are only populated for objects created by parse
+        size_t start;
+        size_t end;
     };
     std::shared_ptr<Members> m;
 #include <qpdf/JSON.hh>
+#include <qpdf/QIntC.hh>
 #include <qpdf/QTC.hh>
 #include <qpdf/QUtil.hh>
 #include <cstring>
 #include <stdexcept>
 JSON::Members::Members(std::shared_ptr<JSON_value> value) :
-    value(value)
+    value(value),
+    start(0),
+    end(0)
 {
 }
@@ -455,7 +458,8 @@ namespace
     class JSONParser
     {
       public:
-        JSONParser() :
+        JSONParser(JSON::Reactor* reactor) :
+            reactor(reactor),
             lex_state(ls_top),
             number_before_point(0),
             number_after_point(0),
@@ -499,6 +503,7 @@ namespace
             ls_backslash,
         };
+        JSON::Reactor* reactor;
         lex_state_e lex_state;
         size_t number_before_point;
         size_t number_after_point;
@@ -828,10 +833,18 @@ JSONParser::handleToken()
         switch (*tok_start) {
         case '{':
             item = std::make_shared<JSON>(JSON::makeDictionary());
+            item->setStart(QIntC::to_size(tok_start - cstr));
+            if (reactor) {
+                reactor->dictionaryStart();
+            }
             break;
         case '[':
             item = std::make_shared<JSON>(JSON::makeArray());
+            item->setStart(QIntC::to_size(tok_start - cstr));
+            if (reactor) {
+                reactor->arrayStart();
+            }
             break;
         default:
@@ -997,6 +1010,11 @@ JSONParser::handleToken()
     } else if ((delimiter == '}') || (delimiter == ']')) {
         next_state = ps_stack.back();
         ps_stack.pop_back();
+        auto tos = stack.back();
+        tos->setEnd(QIntC::to_size(tok_end - cstr));
+        if (reactor) {
+            reactor->containerEnd(*tos);
+        }
         if (next_state != ps_done) {
             stack.pop_back();
         }
@@ -1004,6 +1022,11 @@ JSONParser::handleToken()
         throw std::logic_error(
             "JSONParser::handleToken: unexpected delimiter in transition");
     } else if (item.get()) {
+        if (!(item->isArray() || item->isDictionary())) {
+            item->setStart(QIntC::to_size(tok_start - cstr));
+            item->setEnd(QIntC::to_size(tok_end - cstr));
+        }
+
         std::shared_ptr<JSON> tos;
         if (!stack.empty()) {
             tos = stack.back();
@@ -1017,14 +1040,18 @@ JSONParser::handleToken()
             break;
         case ps_dict_after_colon:
-            tos->addDictionaryMember(dict_key, *item);
+            if (!reactor || !reactor->dictionaryItem(dict_key, *item)) {
+                tos->addDictionaryMember(dict_key, *item);
+            }
             next_state = ps_dict_after_item;
             break;
         case ps_array_begin:
         case ps_array_after_comma:
+            if (!reactor || !reactor->arrayItem(*item)) {
+                tos->addArrayElement(*item);
+            }
             next_state = ps_array_after_item;
-            tos->addArrayElement(*item);
             break;
         case ps_top:
@@ -1083,12 +1110,40 @@ JSONParser::parse(std::string const&amp; s)
         QTC::TC("libtests", "JSON parse premature EOF");
         throw std::runtime_error("JSON: premature end of input");
     }
-    return stack.back();
+    auto const& tos = stack.back();
+    if (reactor && tos.get() && !(tos->isArray() || tos->isDictionary())) {
+        reactor->topLevelScalar();
+    }
+    return tos;
 }
 JSON
-JSON::parse(std::string const& s)
+JSON::parse(std::string const& s, Reactor* reactor)
 {
-    JSONParser jp;
+    JSONParser jp(reactor);
     return *jp.parse(s);
 }
+
+void
+JSON::setStart(size_t start)
+{
+    this->m->start = start;
+}
+
+void
+JSON::setEnd(size_t end)
+{
+    this->m->end = end;
+}
+
+size_t
+JSON::getStart() const
+{
+    return this->m->start;
+}
+
+size_t
+JSON::getEnd() const
+{
+    return this->m->end;
+}
 #include <qpdf/JSON.hh>
 #include <qpdf/QUtil.hh>
+#include <cstdlib>
+#include <cstring>
 #include <iostream>
+#include <memory>
+
+namespace
+{
+    class Reactor: public JSON::Reactor
+    {
+      public:
+        virtual ~Reactor() = default;
+        virtual void dictionaryStart() override;
+        virtual void arrayStart() override;
+        virtual void containerEnd(JSON const& value) override;
+        virtual void topLevelScalar() override;
+        virtual bool
+        dictionaryItem(std::string const& key, JSON const& value) override;
+        virtual bool arrayItem(JSON const& value) override;
+
+      private:
+        void printItem(JSON const&);
+    };
+} // namespace
+
+void
+Reactor::dictionaryStart()
+{
+    std::cout << "dictionary start" << std::endl;
+}
+
+void
+Reactor::arrayStart()
+{
+    std::cout << "array start" << std::endl;
+}
+
+void
+Reactor::containerEnd(JSON const& value)
+{
+    std::cout << "container end: ";
+    printItem(value);
+}
+
+void
+Reactor::topLevelScalar()
+{
+    std::cout << "top-level scalar" << std::endl;
+}
+
+bool
+Reactor::dictionaryItem(std::string const& key, JSON const& value)
+{
+    std::cout << "dictionary item: " << key << " -> ";
+    printItem(value);
+    if (key == "keep") {
+        return false;
+    }
+    return true;
+}
+
+bool
+Reactor::arrayItem(JSON const& value)
+{
+    std::cout << "array item: ";
+    printItem(value);
+    std::string n;
+    if (value.getString(n) && n == "keep") {
+        return false;
+    }
+    return true;
+}
+
+void
+Reactor::printItem(JSON const& j)
+{
+    std::cout << "[" << j.getStart() << ", " << j.getEnd()
+              << "): " << j.unparse() << std::endl;
+}
+
+static void
+usage()
+{
+    std::cerr << "Usage: json_parse file [--react]" << std::endl;
+    exit(2);
+}
 int
 main(int argc, char* argv[])
 {
-    if (argc != 2) {
-        std::cerr << "Usage: json_parse file" << std::endl;
+    if ((argc < 2) || (argc > 3)) {
+        usage();
         return 2;
     }
     char const* filename = argv[1];
+    std::shared_ptr<Reactor> reactor;
+    if (argc == 3) {
+        if (strcmp(argv[2], "--react") == 0) {
+            reactor = std::make_shared<Reactor>();
+        } else {
+            usage();
+        }
+    }
     try {
         std::shared_ptr<char> buf;
         size_t size;
         QUtil::read_file_into_memory(filename, buf, size);
         std::string s(buf.get(), size);
-        std::cout << JSON::parse(s).unparse() << std::endl;
+        std::cout << JSON::parse(s, reactor.get()).unparse() << std::endl;
     } catch (std::exception& e) {
         std::cerr << "exception: " << filename << ": " << e.what() << std::endl;
         return 2;
@@ -32,7 +32,7 @@ if ($^O ne &#39;msys&#39;)
 cleanup();
-my $good = 9;
+my $good = 10;
 for (my $i = 1; $i <= $good; ++$i)
 {
@@ -73,6 +73,11 @@ for (my $i = 1; $i &lt;= $good; ++$i)
                      {$td->FILE => "out.json"},
                      {$td->STRING => ""});
     }
+
+    $td->runtest("good $n reactor",
+                 {$td->COMMAND => "json_parse good-$n.json --react"},
+                 {$td->FILE => "good-$n-react.out", $td->EXIT_STATUS => 0},
+                 $td->NORMALIZE_NEWLINES);
 }
 my @bad = (
@@ -127,7 +132,7 @@ foreach my $d (@bad)
 cleanup();
-$td->report((2 * $good) + scalar(@bad));
+$td->report((3 * $good) + scalar(@bad));
 sub cleanup
 {
+dictionary start
+dictionary item: a -> [6, 11): "bcd"
+array start
+dictionary item: e -> [18, 0): []
+array item: [19, 20): 1
+array item: [41, 42): 2
+array item: [44, 45): 3
+array item: [46, 47): 4
+array item: [48, 54): "five"
+dictionary start
+array item: [56, 0): {}
+dictionary item: six -> [64, 65): 7
+dictionary item: 8 -> [72, 73): 9
+container end: [56, 74): {}
+array item: [76, 80): null
+array item: [82, 86): true
+array item: [107, 112): false
+array item: [114, 134): "a\b\f\n\r\t\\\"/z"
+container end: [18, 135): []
+container end: [0, 136): {}
+{}
+dictionary start
+container end: [0, 2): {}
+{}
+array start
+container end: [0, 2): []
+[]
+array start
+array start
+array item: [1, 0): []
+array start
+array item: [2, 0): []
+dictionary start
+array item: [3, 0): {}
+container end: [3, 5): {}
+container end: [2, 6): []
+dictionary start
+array item: [8, 0): {}
+dictionary start
+dictionary item:  -> [13, 0): {}
+container end: [13, 15): {}
+container end: [8, 16): {}
+container end: [1, 17): []
+container end: [0, 18): []
+[]
+top-level scalar
+"x"
+top-level scalar
+123
+top-level scalar
+-123
+array start
+array item: [1, 2): 1
+array item: [4, 6): -2
+array item: [8, 11): 3.4
+array item: [13, 17): -5.6
+array item: [19, 23): -9e1
+array item: [25, 29): 10e2
+array item: [31, 37): 12.3e5
+array item: [39, 46): 12.6e-7
+container end: [0, 47): []
+[]
+array start
+array item: [1, 7): "aπb"
+array item: [9, 23): "a\b\f\n\r\tc"
+array item: [25, 42): "aπbπc"
+array item: [44, 52): "π"
+array item: [54, 71): "a\u0018bʬc"
+container end: [0, 72): []
+[]
+dictionary start
+array start
+dictionary item: a -> [9, 0): []
+array item: [10, 11): 1
+array item: [13, 14): 2
+dictionary start
+array item: [16, 0): {}
+dictionary item: x -> [22, 25): "y"
+container end: [16, 26): {}
+array item: [28, 29): 3
+dictionary start
+array item: [31, 0): {}
+dictionary item: keep -> [40, 61): "not in final output"
+container end: [31, 62): {
+  "keep": "not in final output"
+}
+container end: [9, 63): []
+array start
+dictionary item: keep -> [75, 0): []
+array item: [76, 77): 1
+array item: [79, 83): null
+array item: [85, 86): 2
+array item: [88, 93): false
+array item: [95, 101): "keep"
+array item: [103, 104): 3
+array start
+array item: [106, 0): []
+array item: [107, 113): "this"
+array item: [115, 121): "keep"
+array item: [123, 128): "not"
+array item: [130, 137): "final"
+container end: [106, 138): [
+  "keep"
+]
+container end: [75, 139): [
+  "keep"
+]
+container end: [0, 141): {
+  "keep": [
+    "keep"
+  ]
+}
+{
+  "keep": [
+    "keep"
+  ]
+}
+{
+  "a": [1, 2, {"x": "y"}, 3, {"keep": "not in final output"}],
+  "keep": [1, null, 2, false, "keep", 3, ["this", "keep", "not", "final"]]
+}
+{
+  "a": [
+    1,
+    2,
+    {
+      "x": "y"
+    },
+    3,
+    {
+      "keep": "not in final output"
+    }
+  ],
+  "keep": [
+    1,
+    null,
+    2,
+    false,
+    "keep",
+    3,
+    [
+      "this",
+      "keep",
+      "not",
+      "final"
+    ]
+  ]
+}