More decoder fixes, and slightly changed parse call semantics.

Prior to this change, if an error was returned, it would be
guaranteed to always return a short byte count.  Now the two
concepts are a bit more orthogonal.  There are cases where
the entire input is consumed even though an error was
encountered.

tests/pb/test_decoder.cc

@@ -202,6 +202,16 @@ string submsg(uint32_t fn, const string& buf) {
   return cat( tag(fn, UPB_WIRE_TYPE_DELIMITED), delim(buf) );
 }
 
+// Like delim()/submsg(), but intentionally encodes an incorrect length.
+// These help test when a delimited boundary doesn't land in the right place.
+string badlen_delim(int err, const string& buf) {
+  return cat(varint(buf.size() + err), buf);
+}
+
+string badlen_submsg(int err, uint32_t fn, const string& buf) {
+  return cat( tag(fn, UPB_WIRE_TYPE_DELIMITED), badlen_delim(err, buf) );
+}
+
 
 /* A set of handlers that covers all .proto types *****************************/
 
@@ -436,6 +446,21 @@ upb::reffed_ptr<const upb::MessageDef> NewMessageDef() {
   ASSERT(f->set_message_subdef(md.get(), NULL));
   ASSERT(md->AddField(f.get(), NULL));
 
+  f = upb::FieldDef::New();
+  ASSERT(f->set_name("f_group", NULL));
+  ASSERT(f->set_number(UPB_DESCRIPTOR_TYPE_GROUP, NULL));
+  f->set_descriptor_type(UPB_DESCRIPTOR_TYPE_GROUP);
+  ASSERT(f->set_message_subdef(md.get(), NULL));
+  ASSERT(md->AddField(f.get(), NULL));
+
+  f = upb::FieldDef::New();
+  ASSERT(f->set_name("r_group", NULL));
+  ASSERT(f->set_number(rep_fn(UPB_DESCRIPTOR_TYPE_GROUP), NULL));
+  f->set_label(UPB_LABEL_REPEATED);
+  f->set_descriptor_type(UPB_DESCRIPTOR_TYPE_GROUP);
+  ASSERT(f->set_message_subdef(md.get(), NULL));
+  ASSERT(md->AddField(f.get(), NULL));
+
   upb::reffed_ptr<upb::EnumDef> e = upb::EnumDef::New();
   ASSERT(e->AddValue("FOO", 1, NULL));
   ASSERT(e->Freeze(NULL));
@@ -492,6 +517,8 @@ upb::reffed_ptr<const upb::Handlers> NewHandlers(TestMode mode) {
     // to this type, eg: message M { optional M m = 1; }
     reg_subm(h.get(), UPB_DESCRIPTOR_TYPE_MESSAGE);
     reg_subm(h.get(), rep_fn(UPB_DESCRIPTOR_TYPE_MESSAGE));
+    reg_subm(h.get(), UPB_DESCRIPTOR_TYPE_GROUP);
+    reg_subm(h.get(), rep_fn(UPB_DESCRIPTOR_TYPE_GROUP));
 
     // For NOP_FIELD we register no handlers, so we can pad a proto freely without
     // changing the output.
@@ -645,6 +672,12 @@ void assert_successful_parse(const string& proto,
 
 void assert_does_not_parse_at_eof(const string& proto) {
   run_decoder(proto, NULL);
+
+  // Also test that we fail to parse at end-of-submessage, not just
+  // end-of-message.
+  run_decoder(submsg(UPB_DESCRIPTOR_TYPE_MESSAGE, proto), NULL);
+  run_decoder(cat(submsg(UPB_DESCRIPTOR_TYPE_MESSAGE, proto), thirty_byte_nop),
+              NULL);
 }
 
 void assert_does_not_parse(const string& proto) {
@@ -859,6 +892,18 @@ void test_invalid() {
                    tag(UPB_DESCRIPTOR_TYPE_GROUP, UPB_WIRE_TYPE_START_GROUP)),
             tag(UPB_DESCRIPTOR_TYPE_GROUP, UPB_WIRE_TYPE_END_GROUP)));
 
+  // Unknown string extends past enclosing submessage.
+  assert_does_not_parse(
+      cat (badlen_submsg(-1, UPB_DESCRIPTOR_TYPE_MESSAGE,
+                         submsg(12345, string("   "))),
+           submsg(UPB_DESCRIPTOR_TYPE_MESSAGE, string("     "))));
+
+  // Unknown fixed-length field extends past enclosing submessage.
+  assert_does_not_parse(
+      cat (badlen_submsg(-1, UPB_DESCRIPTOR_TYPE_MESSAGE,
+                         cat( tag(12345, UPB_WIRE_TYPE_64BIT), uint64(0))),
+           submsg(UPB_DESCRIPTOR_TYPE_MESSAGE, string("     "))));
+
   // Test exceeding the resource limit of stack depth.
   string buf;
   for (int i = 0; i <= MAX_NESTING; i++) {
@@ -942,8 +987,7 @@ void test_valid() {
 
   // Unknown field inside a known submessage.
   assert_successful_parse(
-      cat (submsg(UPB_DESCRIPTOR_TYPE_MESSAGE, 
-             submsg(12345, string("   "))),
+      cat (submsg(UPB_DESCRIPTOR_TYPE_MESSAGE, submsg(12345, string("   "))),
            tag(UPB_DESCRIPTOR_TYPE_INT32, UPB_WIRE_TYPE_VARINT),
            varint(5)),
       LINE("<")

tests/test_util.h

@@ -117,21 +117,6 @@ class VerboseParserEnvironment {
       }
     }
 
-    if (status_.ok() != (parsed >= bytes)) {
-      if (status_.ok()) {
-        fprintf(stderr,
-                "Error: decode function returned short byte count but set no "
-                "error status\n");
-      } else {
-        fprintf(stderr,
-                "Error: decode function returned complete byte count but set "
-                "error status\n");
-      }
-      fprintf(stderr, "Status: %s, parsed=%u, bytes=%u\n",
-              status_.error_message(), (unsigned)parsed, (unsigned)bytes);
-      ASSERT(false);
-    }
-
     if (!status_.ok())
       return false;
 

upb/pb/compile_decoder_x64.dasc

@@ -300,7 +300,7 @@ static void emit_static_asm(jitcompiler *jc) {
   |  mov   rcx, DELIMEND
   |  sub   rcx, PTR
   |  sub   rcx, rdx
-  |  jb    ->err  // Len is greater than enclosing message.
+  |  jb    >4   // Len is greater than enclosing message.
   |  mov   FRAME->end_ofs, rcx
   |  cmp   FRAME, DECODER->limit
   |  je    >3   // Stack overflow
@@ -319,7 +319,7 @@ static void emit_static_asm(jitcompiler *jc) {
   |2:
   |  ret
   |3:
-  |  // Error -- call seterr.
+  |  // Stack overflow error.
   |  mov   PTR, DECODER->checkpoint  // Rollback to before the delim len.
   |  // Prepare seterr args.
   |  mov   ARG1_64, DECODER
@@ -327,6 +327,15 @@ static void emit_static_asm(jitcompiler *jc) {
   |  callp upb_pbdecoder_seterr
   |  call  ->suspend
   |  jmp   <1
+  |4:
+  |  // Overextended len.
+  |  mov   PTR, DECODER->checkpoint  // Rollback to before the delim len.
+  |  // Prepare seterr args.
+  |  mov   ARG1_64, DECODER
+  |  ld64  kPbDecoderSubmessageTooLong
+  |  callp upb_pbdecoder_seterr
+  |  call  ->suspend
+  |  jmp   <1
   |
   | // For getting a value that spans a buffer seam.  Falls back to C.
   |.macro getvalue_slow, func, bytes

upb/pb/decoder.c

@@ -26,6 +26,8 @@
 
 /* Error messages that are shared between the bytecode and JIT decoders. */
 const char *kPbDecoderStackOverflow = "Nesting too deep.";
+const char *kPbDecoderSubmessageTooLong =
+    "Submessage end extends past enclosing submessage.";
 
 /* Error messages shared within this file. */
 static const char *kUnterminatedVarint = "Unterminated varint.";
@@ -185,8 +187,11 @@ static void checkpoint(upb_pbdecoder *d) {
  */
 static int32_t skip(upb_pbdecoder *d, size_t bytes) {
   assert(!in_residual_buf(d, d->ptr) || d->size_param == 0);
-  assert(bytes <= delim_remaining(d));
-  if (bufleft(d) > bytes) {
+  assert(d->skip == 0);
+  if (bytes > delim_remaining(d)) {
+    seterr(d, "Skipped value extended beyond enclosing submessage.");
+    return upb_pbdecoder_suspend(d);
+  } else if (bufleft(d) > bytes) {
     /* Skipped data is all in current buffer, and more is still available. */
     advance(d, bytes);
     d->skip = 0;
@@ -222,7 +227,9 @@ int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf,
   d->checkpoint = d->ptr;
 
   if (d->skip) {
-    CHECK_RETURN(skip(d, d->skip));
+    size_t skip_bytes = d->skip;
+    d->skip = 0;
+    CHECK_RETURN(skip(d, skip_bytes));
     d->checkpoint = d->ptr;
   }
 
@@ -461,7 +468,7 @@ static bool decoder_push(upb_pbdecoder *d, uint64_t end) {
   upb_pbdecoder_frame *fr = d->top;
 
   if (end > fr->end_ofs) {
-    seterr(d, "Submessage end extends past enclosing submessage.");
+    seterr(d, kPbDecoderSubmessageTooLong);
     return false;
   } else if (fr == d->limit) {
     seterr(d, kPbDecoderStackOverflow);
@@ -565,34 +572,7 @@ have_tag:
       return DECODE_OK;
     }
 
-    if (d->ptr == d->delim_end) {
-      seterr(d, "Enclosing submessage ended in the middle of value or group");
-      /* Unlike most errors we notice during parsing, right now we have consumed
-       * all of the user's input.
-       *
-       * There are three different options for how to handle this case:
-       *
-       *   1. decode() = short count, error = set
-       *   2. decode() = full count, error = set
-       *   3. decode() = full count, error NOT set, short count and error will
-       *      be reported on next call to decode() (or end())
-       *
-       * (1) and (3) have the advantage that they preserve the invariant that an
-       * error occurs iff decode() returns a short count.
-       *
-       * (2) and (3) have the advantage of reflecting the fact that all of the
-       * bytes were in fact parsed (and possibly delivered to the unknown field
-       * handler, in the future when that is supported).
-       *
-       * (3) requires extra state in the decode (a place to store the "permanent
-       * error" that we should return for all subsequent attempts to decode).
-       * But we likely want this anyway.
-       *
-       * Right now we do (1), thanks to the fact that we checkpoint *after* this
-       * check.  (3) may be a better choice long term; unclear at the moment. */
-      return upb_pbdecoder_suspend(d);
-    }
-
+    /* Unknown group -- continue looping over unknown fields. */
     checkpoint(d);
   }
 }

upb/pb/decoder.int.h

@@ -273,6 +273,7 @@ void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg);
 
 /* Error messages that are shared between the bytecode and JIT decoders. */
 extern const char *kPbDecoderStackOverflow;
+extern const char *kPbDecoderSubmessageTooLong;
 
 /* Access to decoderplan members needed by the decoder. */
 const char *upb_pbdecoder_getopname(unsigned int op);