Changed parser semantics around skipping.

Prior to this change:

  parse(buf, len) -> len + N

...would indicate that the next N bytes of the input are not
needed, *and* would advance the decoding position by this
much.

After this change:

  parse(buf, len) -> len + N
  parse(NULL, N) -> N

...can be used to achieve the same thing.  But skipping the
N bytes is not explicitly performed by the user.  A user that
doesn't want/need to skip can just say:

  parsed = parse(buf, len);

  if (parsed < len) {
    // Handle suspend, advance stream by "parsed".
  } else {
    // Stream was advanced by "len" (even if parsed > len).
  }

Updated unit tests to test this new behavior, and refactored
test utility code a bit to support it.

3 files changed, 196 insertions, 144 deletions

diff --git a/tests/json/test_json.cc b/tests/json/test_json.cc
index 8dd4062..c483cf0 100644
--- a/tests/json/test_json.cc
+++ b/tests/json/test_json.cc
@@ -288,33 +288,22 @@ void test_json_roundtrip_message(const char* json_src,
                                  const char* json_expected,
                                  const upb::Handlers* serialize_handlers,
                                  int seam) {
-  upb::Status st;
-  upb::Environment env;
-  env.ReportErrorsTo(&st);
+  VerboseParserEnvironment env(verbose);
   StringSink data_sink;
-  upb::json::Printer* printer =
-      upb::json::Printer::Create(&env, serialize_handlers, data_sink.Sink());
-  upb::json::Parser* parser = upb::json::Parser::Create(&env, printer->input());
-
-  upb::BytesSink* input = parser->input();
-  void *sub;
-  size_t len = strlen(json_src);
-  size_t ofs = 0;
-
-  bool ok = input->Start(0, &sub) &&
-            parse_buffer(input, sub, json_src, 0, seam, &ofs, &st, verbose) &&
-            parse_buffer(input, sub, json_src, seam, len, &ofs, &st, verbose) &&
-            ofs == len;
-
-  if (ok) {
-    if (verbose) {
-      fprintf(stderr, "calling end()\n");
-    }
-    ok = input->End();
-  }
+  upb::json::Printer* printer = upb::json::Printer::Create(
+      env.env(), serialize_handlers, data_sink.Sink());
+  upb::json::Parser* parser =
+      upb::json::Parser::Create(env.env(), printer->input());
+  env.ResetBytesSink(parser->input());
+  env.Reset(json_src, strlen(json_src), false);
+
+  bool ok = env.Start() &&
+            env.ParseBuffer(seam) &&
+            env.ParseBuffer(-1) &&
+            env.End();
 
   if (!ok) {
-    fprintf(stderr, "upb parse error: %s\n", st.error_message());
+    fprintf(stderr, "upb parse error: %s\n", env.status().error_message());
   }
   ASSERT(ok);
 
diff --git a/tests/pb/test_decoder.cc b/tests/pb/test_decoder.cc
index 6bb53fc..310f8c1 100644
--- a/tests/pb/test_decoder.cc
+++ b/tests/pb/test_decoder.cc
@@ -63,6 +63,8 @@
 
 #define MAX_NESTING 64
 
+#define LINE(x) x "\n"
+
 uint32_t filter_hash = 0;
 double completed;
 double total;
@@ -517,12 +519,13 @@ upb::pb::Decoder* CreateDecoder(upb::Environment* env,
 }
 
 uint32_t Hash(const string& proto, const string* expected_output, size_t seam1,
-              size_t seam2) {
+              size_t seam2, bool may_skip) {
   uint32_t hash = MurmurHash2(proto.c_str(), proto.size(), 0);
   if (expected_output)
     hash = MurmurHash2(expected_output->c_str(), expected_output->size(), hash);
   hash = MurmurHash2(&seam1, sizeof(seam1), hash);
   hash = MurmurHash2(&seam2, sizeof(seam2), hash);
+  hash = MurmurHash2(&may_skip, sizeof(may_skip), hash);
   return hash;
 }
 
@@ -538,91 +541,88 @@ void CheckBytesParsed(const upb::pb::Decoder& decoder, size_t ofs) {
   ASSERT(ofs <= (decoder.BytesParsed() + MAX_BUFFERED));
 }
 
-static bool parse(upb::pb::Decoder* decoder, void* subc, const char* buf,
-                  size_t start, size_t end, size_t* ofs, upb::Status* status) {
-  CheckBytesParsed(*decoder, *ofs);
-  bool ret = parse_buffer(decoder->input(), subc, buf, start, end, ofs, status,
-                          filter_hash != 0);
+static bool parse(VerboseParserEnvironment* env,
+                  const upb::pb::Decoder& decoder, int bytes) {
+  CheckBytesParsed(decoder, env->ofs());
+  bool ret = env->ParseBuffer(bytes);
   if (ret) {
-    CheckBytesParsed(*decoder, *ofs);
+    CheckBytesParsed(decoder, env->ofs());
   }
 
   return ret;
 }
 
-#define LINE(x) x "\n"
-void run_decoder(const string& proto, const string* expected_output) {
-  upb::Status status;
-  upb::Sink sink(global_handlers, &closures[0]);
-  for (size_t i = 0; i < proto.size(); i++) {
-    for (size_t j = i; j < UPB_MIN(proto.size(), i + 5); j++) {
-      // TODO(haberman): hoist this again once the environment supports reset.
-      upb::Environment env;
-      env.ReportErrorsTo(&status);
-      upb::pb::Decoder *decoder = CreateDecoder(&env, global_method, &sink);
-
-      testhash = Hash(proto, expected_output, i, j);
-      if (filter_hash && testhash != filter_hash) continue;
-      if (test_mode != COUNT_ONLY) {
-        output.clear();
-        status.Clear();
-        size_t ofs = 0;
-        upb::BytesSink* input = decoder->input();
-        void *sub;
-
-        if (filter_hash) {
-          fprintf(stderr, "RUNNING TEST CASE, hash=%x\n", testhash);
-          fprintf(stderr, "JIT on: %s\n",
-                  global_method->is_native() ? "true" : "false");
-          fprintf(stderr, "Input (len=%u): ", (unsigned)proto.size());
-          PrintBinary(proto);
-          fprintf(stderr, "\n");
-          if (expected_output) {
-            fprintf(stderr, "Expected output: %s\n", expected_output->c_str());
-          } else {
-            fprintf(stderr, "Expected to FAIL\n");
-          }
-          fprintf(stderr, "Calling start()\n");
-        }
+void do_run_decoder(VerboseParserEnvironment* env, upb::pb::Decoder* decoder,
+                    const string& proto, const string* expected_output,
+                    size_t i, size_t j, bool may_skip) {
+  env->Reset(proto.c_str(), proto.size(), may_skip);
+  decoder->Reset();
 
-        bool ok = input->Start(proto.size(), &sub) &&
-                  parse(decoder, sub, proto.c_str(), 0, i, &ofs, &status) &&
-                  parse(decoder, sub, proto.c_str(), i, j, &ofs, &status) &&
-                  parse(decoder, sub, proto.c_str(), j, proto.size(), &ofs,
-                        &status) &&
-                  ofs == proto.size();
+  testhash = Hash(proto, expected_output, i, j, may_skip);
+  if (filter_hash && testhash != filter_hash) return;
+  if (test_mode != COUNT_ONLY) {
+    output.clear();
+
+    if (filter_hash) {
+      fprintf(stderr, "RUNNING TEST CASE, hash=%x\n", testhash);
+      fprintf(stderr, "JIT on: %s\n",
+              global_method->is_native() ? "true" : "false");
+      fprintf(stderr, "Input (len=%u): ", (unsigned)proto.size());
+      PrintBinary(proto);
+      fprintf(stderr, "\n");
+      if (expected_output) {
+        fprintf(stderr, "Expected output: %s\n", expected_output->c_str());
+      } else {
+        fprintf(stderr, "Expected to FAIL\n");
+      }
+      fprintf(stderr, "Calling start()\n");
+    }
 
+    bool ok = env->Start() &&
+              parse(env, *decoder, i) &&
+              parse(env, *decoder, j - i) &&
+              parse(env, *decoder, -1) &&
+              env->End();
+
+    ASSERT(ok == env->status().ok());
+
+    if (test_mode == ALL_HANDLERS) {
+      if (expected_output) {
+        if (output != *expected_output) {
+          fprintf(stderr, "Text mismatch: '%s' vs '%s'\n",
+                  output.c_str(), expected_output->c_str());
+        }
+        if (!ok) {
+          fprintf(stderr, "Failed: %s\n", env->status().error_message());
+        }
+        ASSERT(ok);
+        ASSERT(output == *expected_output);
+      } else {
         if (ok) {
-          if (filter_hash) {
-            fprintf(stderr, "calling end()\n");
-          }
-          ok = input->End();
+          fprintf(stderr, "Didn't expect ok result, but got output: '%s'\n",
+                  output.c_str());
+        } else if (filter_hash) {
+          fprintf(stderr, "Failed as we expected, with message: %s\n",
+                  env->status().error_message());
         }
+        ASSERT(!ok);
+      }
+    }
+  }
+  (*count)++;
+}
 
-        if (test_mode == ALL_HANDLERS) {
-          if (expected_output) {
-            if (output != *expected_output) {
-              fprintf(stderr, "Text mismatch: '%s' vs '%s'\n",
-                      output.c_str(), expected_output->c_str());
-            }
-            if (!ok) {
-              fprintf(stderr, "Failed: %s\n", status.error_message());
-            }
-            ASSERT(ok);
-            ASSERT(output == *expected_output);
-          } else {
-            if (ok) {
-              fprintf(stderr, "Didn't expect ok result, but got output: '%s'\n",
-                      output.c_str());
-            } else if (filter_hash) {
-              fprintf(stderr, "Failed as we expected, with message: %s\n",
-                      status.error_message());
-            }
-            ASSERT(!ok);
-          }
-        }
+void run_decoder(const string& proto, const string* expected_output) {
+  VerboseParserEnvironment env(filter_hash != 0);
+  upb::Sink sink(global_handlers, &closures[0]);
+  upb::pb::Decoder *decoder = CreateDecoder(env.env(), global_method, &sink);
+  env.ResetBytesSink(decoder->input());
+  for (size_t i = 0; i < proto.size(); i++) {
+    for (size_t j = i; j < UPB_MIN(proto.size(), i + 5); j++) {
+      do_run_decoder(&env, decoder, proto, expected_output, i, j, true);
+      if (env.SkippedWithNull()) {
+        do_run_decoder(&env, decoder, proto, expected_output, i, j, false);
       }
-      (*count)++;
     }
   }
   testhash = 0;
@@ -1146,20 +1146,14 @@ upb::reffed_ptr<const upb::pb::DecoderMethod> method =
     { NULL, 0 },
   };
   for (int i = 0; testdata[i].data; i++) {
-    upb::Environment env;
-    upb::Status status;
-    env.ReportErrorsTo(&status);
+    VerboseParserEnvironment env(filter_hash != 0);
     upb::Sink sink(method->dest_handlers(), &closures[0]);
-    upb::pb::Decoder* decoder = CreateDecoder(&env, method.get(), &sink);
-    upb::BytesSink* input = decoder->input();
-    void* subc;
-    ASSERT(input->Start(0, &subc));
-    size_t ofs = 0;
-    ASSERT(parse_buffer(input, subc,
-                        testdata[i].data, 0, testdata[i].length,
-                        &ofs, &status, false));
-    ASSERT(ofs == testdata[i].length);
-    ASSERT(input->End());
+    upb::pb::Decoder* decoder = CreateDecoder(env.env(), method.get(), &sink);
+    env.ResetBytesSink(decoder->input());
+    env.Reset(testdata[i].data, testdata[i].length, true);
+    ASSERT(env.Start());
+    ASSERT(env.ParseBuffer(-1));
+    ASSERT(env.End());
   }
 }
 
diff --git a/tests/test_util.h b/tests/test_util.h
index d9e8d25..6408557 100644
--- a/tests/test_util.h
+++ b/tests/test_util.h
@@ -8,59 +8,90 @@
 #include <stdio.h>
 #include <math.h>
 #include "tests/upb_test.h"
+#include "upb/env.h"
 #include "upb/sink.h"
 
 upb::BufferHandle global_handle;
 
-// Puts a region of the given buffer [start, end) into the given sink (which
-// probably represents a parser.  Can gracefully handle the case where the
-// parser returns a "parsed" length that is less or greater than the input
-// buffer length, and tracks the overall parse offset in *ofs.
-//
-// Pass verbose=true to print detailed diagnostics to stderr.
-bool parse_buffer(upb::BytesSink* sink, void* subc, const char* buf,
-                  size_t start, size_t end, size_t* ofs,
-                  upb::Status* status, bool verbose) {
-  start = UPB_MAX(start, *ofs);
+class VerboseParserEnvironment {
+ public:
+  VerboseParserEnvironment(bool verbose) : verbose_(verbose) {
+    env_.ReportErrorsTo(&status_);
+  }
+
+  void Reset(const char *buf, size_t len, bool may_skip) {
+    buf_ = buf;
+    len_ = len;
+    ofs_ = 0;
+    skip_until_ = may_skip ? 0 : -1;
+    status_.Clear();
+  }
+
+  bool Start() {
+    return sink_->Start(len_, &subc_);
+  }
+
+  bool End() {
+    return sink_->End();
+  }
+
+  // Puts a region of the given buffer [start, end) into the given sink (which
+  // probably represents a parser.  Can gracefully handle the case where the
+  // parser returns a "parsed" length that is less or greater than the input
+  // buffer length, and tracks the overall parse offset in *ofs.
+  //
+  // Pass verbose=true to print detailed diagnostics to stderr.
+  bool ParseBuffer(int bytes) {
+    if (bytes < 0) {
+      bytes = len_ - ofs_;
+    }
 
-  if (start <= end) {
-    size_t len = end - start;
+    ASSERT((size_t)bytes <= (len_ - ofs_));
 
     // Copy buffer into a separate, temporary buffer.
     // This is necessary to verify that the parser is not erroneously
     // reading outside the specified bounds.
-    char *buf2 = (char*)malloc(len);
-    assert(buf2);
-    memcpy(buf2, buf + start, len);
+    char *buf2 = NULL;
 
-    if (verbose) {
+    if ((int)(ofs_ + bytes) > skip_until_) {
+      buf2 = (char*)malloc(bytes);
+      assert(buf2);
+      memcpy(buf2, buf_ + ofs_, bytes);
+    }
+
+    if (buf2 == NULL && bytes == 0) {
+      // Decoders dont' support buf=NULL, bytes=0.
+      return true;
+    }
+
+    if (verbose_) {
       fprintf(stderr, "Calling parse(%u) for bytes %u-%u of the input\n",
-              (unsigned)len, (unsigned)start, (unsigned)end);
+              (unsigned)bytes, (unsigned)ofs_, (unsigned)(ofs_ + bytes));
     }
 
-    size_t parsed = sink->PutBuffer(subc, buf2, len, &global_handle);
+    int parsed = sink_->PutBuffer(subc_, buf2, bytes, &global_handle);
     free(buf2);
 
-    if (verbose) {
-      if (parsed == len) {
+    if (verbose_) {
+      if (parsed == bytes) {
         fprintf(stderr,
                 "parse(%u) = %u, complete byte count indicates success\n",
-                (unsigned)len, (unsigned)len);
-      } else if (parsed > len) {
+                (unsigned)bytes, (unsigned)bytes);
+      } else if (parsed > bytes) {
         fprintf(stderr,
-                "parse(%u) = %u, long byte count indicates success and skip"
+                "parse(%u) = %u, long byte count indicates success and skip "
                 "of the next %u bytes\n",
-                (unsigned)len, (unsigned)parsed, (unsigned)(parsed - len));
+                (unsigned)bytes, (unsigned)parsed, (unsigned)(parsed - bytes));
       } else {
         fprintf(stderr,
                 "parse(%u) = %u, short byte count indicates failure; "
                 "last %u bytes were not consumed\n",
-                (unsigned)len, (unsigned)parsed, (unsigned)(len - parsed));
+                (unsigned)bytes, (unsigned)parsed, (unsigned)(bytes - parsed));
       }
     }
 
-    if (status->ok() != (parsed >= len)) {
-      if (status->ok()) {
+    if (status_.ok() != (parsed >= bytes)) {
+      if (status_.ok()) {
         fprintf(stderr,
                 "Error: decode function returned short byte count but set no "
                 "error status\n");
@@ -69,17 +100,55 @@ bool parse_buffer(upb::BytesSink* sink, void* subc, const char* buf,
                 "Error: decode function returned complete byte count but set "
                 "error status\n");
       }
-      fprintf(stderr, "Status: %s, parsed=%u, len=%u\n",
-              status->error_message(), (unsigned)parsed, (unsigned)len);
+      fprintf(stderr, "Status: %s, parsed=%u, bytes=%u\n",
+              status_.error_message(), (unsigned)parsed, (unsigned)bytes);
       ASSERT(false);
     }
 
-    if (!status->ok())
+    if (!status_.ok())
       return false;
 
-    *ofs += parsed;
+    if (parsed > bytes && skip_until_ >= 0) {
+      skip_until_ = ofs_ + parsed;
+    }
+
+    ofs_ += UPB_MIN(parsed, bytes);
+
+    return true;
+  }
+
+  void ResetBytesSink(upb::BytesSink* sink) {
+    sink_ = sink;
   }
-  return true;
-}
+
+  const upb::Status& status() { return status_; }
+
+  size_t ofs() { return ofs_; }
+  upb::Environment* env() { return &env_; }
+
+  bool SkippedWithNull() { return skip_until_ > 0; }
+
+ private:
+  upb::Environment env_;
+  upb::Status status_;
+  upb::BytesSink* sink_;
+  const char* buf_;
+  size_t len_;
+  bool verbose_;
+  size_t ofs_;
+  void *subc_;
+
+  // When our parse call returns a value greater than the number of bytes
+  // we passed in, the decoder is indicating to us that the next N bytes
+  // in the stream are not needed and can be skipped.  The user is allowed
+  // to pass a NULL buffer for those N bytes.
+  //
+  // skip_until_ is initially set to 0 if we should do this NULL-buffer
+  // skipping or -1 if we should not.  If we are open to doing NULL-buffer
+  // skipping and we get an opportunity to do it, we set skip_until to the
+  // stream offset where we can skip until.  The user can then test whether
+  // this happened by testing SkippedWithNull().
+  int skip_until_;
+};
 
 #endif