Merge pull request 'parser3: keep the stackSize needed by a document independent of chunking (reserve transient token slots)' (#72) from weaselbot/weaseljson:weaselbot/issue-71 into main

Reviewed-on: #72

.gitea/workflows/ci.yml

@@ -12,7 +12,15 @@ jobs:
         run: pipx install pre-commit
 
       - name: Run pre-commit
-        run: ~/.local/bin/pre-commit run --all-files --show-diff-on-failure
+        # `pipx install` writes the app to PIPX_BIN_DIR, which is not
+        # necessarily ~/.local/bin and not necessarily on PATH, so don't
+        # hardcode a path that only some runner images happen to use.
+        run: |
+          if command -v pre-commit >/dev/null 2>&1; then
+            pre-commit run --all-files --show-diff-on-failure
+          else
+            pipx run pre-commit run --all-files --show-diff-on-failure
+          fi
 
   build:
     strategy:

include/weaseljson.h

@@ -36,7 +36,7 @@ enum WeaselJsonStatus {
   WeaselJson_AGAIN,
   /** Invalid json */
   WeaselJson_REJECT,
-  /** json is too deeply nested */
+  /** json is too deeply nested for the parser's stack size */
   WeaselJson_OVERFLOW,
   /** Tried to call parse on a null parser */
   WeaselJson_NULL,
@@ -50,10 +50,12 @@ enum WeaselJsonFlags {
 };
 
 /** Create a parser. Increasing stack size increases memory usage but also
- * increases the depth of nested json accepted. `callbacks` and `userdata` must
- * outlive the returned parser. Returns null if there's insufficient available
- * memory, or if `stackSize` is negative or too small to hold a minimal
- * document. */
+ * increases the depth of nested json accepted. Whether a document is accepted,
+ * rejected or reported as too deeply nested depends only on its bytes and on
+ * `stackSize`, not on how the bytes are split across WeaselJsonParser_parse
+ * calls. `callbacks` and `userdata` must outlive the returned parser. Returns
+ * null if there's insufficient available memory, or if `stackSize` is negative
+ * or too small to hold a minimal document. */
 WeaselJsonParser *WeaselJsonParser_create(int stackSize,
                                           const WeaselJsonCallbacks *callbacks,
                                           void *userdata, int flags);

src/fuzz.cpp

@@ -5,12 +5,12 @@
 
 #include <simdjson.h>
 
-std::pair<std::string, WeaselJsonStatus> runStreaming(std::string copy,
-                                                      int stride) {
+std::pair<std::string, WeaselJsonStatus>
+runStreaming(std::string copy, int stackSize, int stride) {
   SerializeState state;
   auto c = serializeCallbacks();
   std::unique_ptr<WeaselJsonParser, decltype(&WeaselJsonParser_destroy)> parser{
-      WeaselJsonParser_create(1024, &c, &state,
+      WeaselJsonParser_create(stackSize, &c, &state,
                               copy.size() % 2 == 0 ? WeaselJsonRaw : 0),
       WeaselJsonParser_destroy};
   if (stride == 0) {
@@ -31,11 +31,12 @@ std::pair<std::string, WeaselJsonStatus> runStreaming(std::string copy,
   return {state.result, s};
 }
 
-std::pair<std::string, WeaselJsonStatus> runBatch(std::string copy) {
+std::pair<std::string, WeaselJsonStatus> runBatch(std::string copy,
+                                                  int stackSize) {
   SerializeState state;
   auto c = serializeCallbacks();
   std::unique_ptr<WeaselJsonParser, decltype(&WeaselJsonParser_destroy)> parser{
-      WeaselJsonParser_create(1 << 20, &c, &state,
+      WeaselJsonParser_create(stackSize, &c, &state,
                               copy.size() % 2 == 0 ? WeaselJsonRaw : 0),
       WeaselJsonParser_destroy};
   auto s = WeaselJsonParser_parse(parser.get(), copy.data(), copy.size());
@@ -47,11 +48,11 @@ std::pair<std::string, WeaselJsonStatus> runBatch(std::string copy) {
 }
 
 std::pair<std::string, WeaselJsonStatus> runPrefix(std::string copy,
-                                                   int prefix) {
+                                                   int stackSize, int prefix) {
   SerializeState state;
   auto c = serializeCallbacks();
   std::unique_ptr<WeaselJsonParser, decltype(&WeaselJsonParser_destroy)> parser{
-      WeaselJsonParser_create(1024, &c, &state,
+      WeaselJsonParser_create(stackSize, &c, &state,
                               copy.size() % 2 == 0 ? WeaselJsonRaw : 0),
       WeaselJsonParser_destroy};
   auto s = WeaselJsonParser_parse(parser.get(), copy.data(), prefix);
@@ -67,52 +68,51 @@ std::pair<std::string, WeaselJsonStatus> runPrefix(std::string copy,
   return {state.result, s};
 }
 
-void testStreaming(std::string const &json) {
-  auto batch = runBatch(json);
+void checkStreamingResult(
+    int stackSize, char const *kind,
+    std::pair<std::string, WeaselJsonStatus> const &got,
+    std::pair<std::string, WeaselJsonStatus> const &batch) {
+  if (got.second == WeaselJson_AGAIN) {
+    abort();
+  }
+  // The status must be identical however the document is chunked; the events
+  // reported for a document that failed to parse need not match.
+  bool ok = got.second == batch.second &&
+            (got.second != WeaselJson_OK || got.first == batch.first);
+  if (!ok) {
+    printf("%s (stackSize %d): %d, %s\n", kind, stackSize, (int)got.second,
+           got.first.c_str());
+    printf("batch (stackSize %d): %d, %s\n", stackSize, (int)batch.second,
+           batch.first.c_str());
+    abort();
+  }
+}
+
+void testStreaming(std::string const &json, int stackSize) {
+  auto batch = runBatch(json, stackSize);
   if (batch.second == WeaselJson_AGAIN) {
     abort();
   }
   for (int stride = 1; stride < 16; ++stride) {
-    auto streaming = runStreaming(json, stride);
-    if (streaming != batch) {
-      if (streaming.second == WeaselJson_AGAIN) {
-        abort();
-      }
-      bool streamingOk = streaming.second == WeaselJson_OK;
-      bool batchOk = batch.second == WeaselJson_OK;
-      if (streamingOk == batchOk && !batchOk) {
-        // It's ok if the processed data doesn't match if parsing failed
-      } else {
-        printf("streaming: %s, %s\n",
-               streaming.second == WeaselJson_OK ? "accept" : "reject",
-               streaming.first.c_str());
-        printf("batch: %s, %s\n",
-               batch.second == WeaselJson_OK ? "accept" : "reject",
-               batch.first.c_str());
-        abort();
-      }
-    }
+    checkStreamingResult(stackSize, "streaming",
+                         runStreaming(json, stackSize, stride), batch);
     if (int(json.size()) > stride) {
-      auto prefix = runPrefix(json, stride);
-      if (prefix != batch) {
-        if (prefix.second == WeaselJson_AGAIN) {
-          abort();
-        }
-        bool prefixOk = prefix.second == WeaselJson_OK;
-        bool batchOk = batch.second == WeaselJson_OK;
-        if (prefixOk == batchOk && !batchOk) {
-          // It's ok if the processed data doesn't match if parsing failed
-        } else {
-          printf("prefix: %s, %s\n",
-                 prefix.second == WeaselJson_OK ? "accept" : "reject",
-                 prefix.first.c_str());
-          printf("batch: %s, %s\n",
-                 batch.second == WeaselJson_OK ? "accept" : "reject",
-                 batch.first.c_str());
-          abort();
+      checkStreamingResult(stackSize, "prefix",
+                           runPrefix(json, stackSize, stride), batch);
     }
   }
 }
+
+void testStreaming(std::string const &json) {
+  // Whether a document parses must not depend on the streaming pattern: the
+  // parser hands the tokenizer progress for a literal or a \uXXXX escape that
+  // straddles a chunk boundary to the pushdown stack, so a stackSize picked for
+  // the document's nesting depth used to overflow for some chunkings only.
+  // Sweep stack sizes around the minimum (3 is the smallest accepted by
+  // WeaselJsonParser_create) so that the property is exercised, then a large
+  // one for documents that need the room.
+  for (int stackSize : {3, 4, 5, 6, 7, 8, 12, 1024}) {
+    testStreaming(json, stackSize);
   }
 }
 

src/lib.cpp

@@ -13,8 +13,7 @@ WeaselJsonParser_create(int stackSize, const WeaselJsonCallbacks *callbacks,
   if (stackSize < 0) { // avoid (size_t)stackSize wrapping to a huge allocation
     return nullptr;
   }
-  auto *buf =
-      malloc(sizeof(Parser3) + (size_t)stackSize * sizeof(*Parser3::stackPtr));
+  auto *buf = malloc(Parser3::allocationSize(stackSize));
   if (buf == nullptr) {
     return nullptr;
   }

src/parser3.h

@@ -72,6 +72,30 @@ enum Symbol : uint8_t {
   T_BACKSLASH,
   N_SYMBOL_COUNT, // Must be last
 };
+
+// Some symbols are on the stack only because a literal (`true`/`false`/`null`)
+// or a `\uXXXX` escape straddles the end of a chunk. The whole-buffer path
+// consumes those characters inline and pushes nothing, so these symbols must
+// not spend the caller's nesting budget: otherwise the stackSize a document
+// needs would depend on where the parse() boundaries happen to fall (issue
+// #71). WeaselJsonParser_create() allocates this many slots beyond `stackSize`
+// to hold them, and Parser3::pushTransient() pushes them there.
+//
+// All the pushes of such symbols, with their sizes:
+//   n_value()                    3, 4, 3  (true, false, null)
+//   n_string2()                  1        (N_STRING_FOLLOWING_ESCAPE)
+//   n_string_following_escape()  4        (the four hex digits)
+//   t_hex2()                     6        (utf-16 surrogate pair)
+//
+// Upper bound on how many of them are on the stack at once: 6. These are
+// pushdown symbols, so the ones that are on the stack are consumed before
+// anything else is pushed; and none of those continuations pushes while
+// another's symbols are still pending, because each pops the symbol it took
+// before pushing (the literal ones pop N_VALUE before n_value() pushes, and the
+// rest pop themselves). The number of pending symbols is therefore bounded by
+// the largest single push, the six symbols that start a surrogate pair.
+inline constexpr int kTransientSymbols = 6;
+
 struct Parser3 {
   Parser3(const WeaselJsonCallbacks *callbacks, void *userdata, int stackSize,
           int flags)
@@ -80,6 +104,14 @@ struct Parser3 {
     reset();
   }
 
+  // Size of the allocation that holds a parser with a nesting budget of
+  // `stackSize` symbols: the parser itself, that budget, and the reserve for
+  // transient tokenizer progress symbols.
+  static constexpr size_t allocationSize(int stackSize) {
+    return sizeof(Parser3) +
+           (size_t(stackSize) + kTransientSymbols) * sizeof(Symbol);
+  }
+
   [[nodiscard]] WeaselJsonStatus parse(char *buf, int len);
 
   void flushNumber(bool done, char *buf) {
@@ -114,9 +146,12 @@ struct Parser3 {
   }
 
   // Pushing symbols onto the stack roughly corresponds with "parse these
-  // nonterminals/terminals in order"
-  [[nodiscard]] WeaselJsonStatus push(std::initializer_list<Symbol> symbols) {
-    if (stackEnd - stackPtr < ptrdiff_t(symbols.size())) [[unlikely]] {
+  // nonterminals/terminals in order". `end` is the first slot the symbols may
+  // not occupy.
+  [[nodiscard]] static WeaselJsonStatus
+  pushOnto(Symbol *&stackPtr, Symbol *end,
+           std::initializer_list<Symbol> symbols) {
+    if (end - stackPtr < ptrdiff_t(symbols.size())) [[unlikely]] {
       return WeaselJson_OVERFLOW;
     }
     for (int i = symbols.size() - 1; i >= 0; --i) {
@@ -125,6 +160,23 @@ struct Parser3 {
     return WeaselJson_OK;
   }
 
+  // Push symbols that model the nesting of the document. This is what
+  // WeaselJson_OVERFLOW reports, so it's bounded by the caller's stackSize.
+  [[nodiscard]] WeaselJsonStatus push(std::initializer_list<Symbol> symbols) {
+    return pushOnto(stackPtr, stackEnd, symbols);
+  }
+
+  // Push tokenizer progress symbols for a literal or escape that straddles a
+  // chunk boundary. Unlike push(), these may grow past `stackEnd` into the
+  // reserve, so they don't spend the caller's nesting budget and can never make
+  // the parser overflow on input it would have accepted in one piece. See
+  // kTransientSymbols for why the reserve is big enough to always succeed.
+  [[nodiscard]] WeaselJsonStatus
+  pushTransient(std::initializer_list<Symbol> symbols) {
+    assert(ptrdiff_t(symbols.size()) <= kTransientSymbols);
+    return pushOnto(stackPtr, stackEnd + kTransientSymbols, symbols);
+  }
+
   [[nodiscard]] Symbol top() const {
     assert(!empty());
     return *(stackPtr - 1);
@@ -376,7 +428,7 @@ inline PRESERVE_NONE ContinuationStatus n_value(Parser3 *self, char *buf,
         return WeaselJson_REJECT;
       }
     } else {
-      if (auto s = self->push({T_R, T_U, N_TRUE})) {
+      if (auto s = self->pushTransient({T_R, T_U, N_TRUE})) {
         return s;
       }
     }
@@ -392,7 +444,7 @@ inline PRESERVE_NONE ContinuationStatus n_value(Parser3 *self, char *buf,
         return WeaselJson_REJECT;
       }
     } else {
-      if (auto s = self->push({T_A, T_L, T_S, N_FALSE})) {
+      if (auto s = self->pushTransient({T_A, T_L, T_S, N_FALSE})) {
         return s;
       }
     }
@@ -408,7 +460,7 @@ inline PRESERVE_NONE ContinuationStatus n_value(Parser3 *self, char *buf,
         return WeaselJson_REJECT;
       }
     } else {
-      if (auto s = self->push({T_U, T_L, N_NULL})) {
+      if (auto s = self->pushTransient({T_U, T_L, N_NULL})) {
         return s;
       }
     }
@@ -609,7 +661,7 @@ inline PRESERVE_NONE ContinuationStatus n_string2(Parser3 *self, char *buf,
   case '\\':
     ++buf;
     if (bufEnd - buf < /*strlen("u0000\\u0000")*/ 11) {
-      if (auto s = self->push({N_STRING_FOLLOWING_ESCAPE})) {
+      if (auto s = self->pushTransient({N_STRING_FOLLOWING_ESCAPE})) {
         return s;
       }
       if (buf == bufEnd) {
@@ -718,7 +770,7 @@ n_string_following_escape(Parser3 *self, char *buf, char *bufEnd) {
     ++buf;
     self->utf8Codepoint = 0;
     self->pop();
-    if (auto s = self->push({T_HEX, T_HEX, T_HEX, T_HEX2})) {
+    if (auto s = self->pushTransient({T_HEX, T_HEX, T_HEX, T_HEX2})) {
       return s;
     }
     break;
@@ -800,8 +852,8 @@ inline PRESERVE_NONE ContinuationStatus t_hex2(Parser3 *self, char *buf,
       self->utf16Surrogate = self->utf8Codepoint;
       self->utf8Codepoint = 0;
       self->pop();
-      if (auto s =
-              self->push({T_BACKSLASH, T_U2, T_HEX, T_HEX, T_HEX, T_HEX3})) {
+      if (auto s = self->pushTransient(
+              {T_BACKSLASH, T_U2, T_HEX, T_HEX, T_HEX, T_HEX3})) {
         return s;
       }
       if (buf == bufEnd) {

src/test.cpp

@@ -4,8 +4,10 @@
 #include <cstring>
 #include <fstream>
 
+#include <algorithm>
 #include <limits>
 #include <string>
+#include <vector>
 
 #include <doctest.h>
 #include <nanobench.h>
@@ -152,6 +154,47 @@ void testStreaming(std::string const &json) {
   CHECK(streaming.result == batch.result);
 }
 
+// Feed `doc` to a fresh parser with the given stackSize, cutting it into chunks
+// at the offsets in `cuts` (ascending, each > 0 and < doc.size(); an empty
+// `cuts` feeds the whole document at once), and return the final status.
+WeaselJsonStatus parseChunked(std::string const &doc, int stackSize,
+                              std::vector<size_t> const &cuts) {
+  auto c = noopCallbacks();
+  std::string copy = doc;
+  auto *parser = WeaselJsonParser_create(stackSize, &c, nullptr, 0);
+  REQUIRE(parser != nullptr);
+  WeaselJsonStatus status = WeaselJson_AGAIN;
+  size_t pos = 0;
+  for (size_t cut : cuts) {
+    if (status != WeaselJson_AGAIN) {
+      break;
+    }
+    if (cut > pos && cut < copy.size()) {
+      status =
+          WeaselJsonParser_parse(parser, copy.data() + pos, int(cut - pos));
+      pos = cut;
+    }
+  }
+  if (status == WeaselJson_AGAIN && pos < copy.size()) {
+    status = WeaselJsonParser_parse(parser, copy.data() + pos,
+                                    int(copy.size() - pos));
+  }
+  if (status == WeaselJson_AGAIN) {
+    status = WeaselJsonParser_parse(parser, nullptr, 0);
+  }
+  WeaselJsonParser_destroy(parser);
+  return status;
+}
+
+// The offsets to cut `doc` into chunks of `stride` bytes.
+std::vector<size_t> stridedCuts(std::string const &doc, size_t stride) {
+  std::vector<size_t> cuts;
+  for (size_t cut = stride; cut < doc.size(); cut += stride) {
+    cuts.push_back(cut);
+  }
+  return cuts;
+}
+
 } // namespace
 
 TEST_CASE("parser3") {
@@ -258,6 +301,55 @@ TEST_CASE("reset clears overflow state") {
   WeaselJsonParser_destroy(parser);
 }
 
+TEST_CASE("acceptance is independent of the chunking") {
+  // Literals and \uXXXX escapes that straddle the end of a chunk are tokenized
+  // by pushing their not-yet-matched symbols onto the pushdown stack, so they
+  // used to spend the caller's nesting budget: a document could be accepted
+  // when fed whole and reported as WeaselJson_OVERFLOW when streamed in smaller
+  // pieces. The decision must depend only on the document and on `stackSize`.
+  std::vector<std::string> docs = {
+      "1",
+      "true",
+      "false",
+      "null",
+      "[true]",
+      "[true,false,null]",
+      "\"\\u00e9abcdefgh\"",
+      // Needs the full transient reserve: a surrogate pair split across chunks
+      // pushes six symbols at once.
+      "\"\\ud801\\udc37\"",
+      "[\"\\u00e9abcdefgh\"]",
+      "{\"\\ud801\\udc37\":[true,{\"x\":null}]}",
+      "[[[[[[[[[[[[true]]]]]]]]]]]]",
+      // Whether an invalid document is rejected must not depend on the chunking
+      // either.
+      "tru",
+      "truX",
+      "[[[[x",
+      "\"\\ud801x\"",
+  };
+  for (auto const &doc : docs) {
+    CAPTURE(doc);
+    // 3 is the smallest stackSize WeaselJsonParser_create accepts, so this
+    // sweeps from "everything overflows" to "everything fits".
+    for (int stackSize = 3; stackSize <= 16; ++stackSize) {
+      CAPTURE(stackSize);
+      auto whole = parseChunked(doc, stackSize, {});
+      // Every way to split the document in two, so that every literal and
+      // escape in `docs` straddles a chunk boundary at some point.
+      for (size_t cut = 1; cut < doc.size(); ++cut) {
+        CAPTURE(cut);
+        REQUIRE(parseChunked(doc, stackSize, {cut}) == whole);
+      }
+      for (size_t stride = 1; stride <= 12; ++stride) {
+        CAPTURE(stride);
+        REQUIRE(parseChunked(doc, stackSize, stridedCuts(doc, stride)) ==
+                whole);
+      }
+    }
+  }
+}
+
 TEST_CASE("rejected state is sticky") {
   auto c = noopCallbacks();
   auto *parser = WeaselJsonParser_create(1024, &c, nullptr, 0);