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.

1 files changed, 6 insertions, 0 deletions

diff --git a/upb/pb/decoder.int.h b/upb/pb/decoder.int.h
index 2d4485a..be5d044 100644
--- a/upb/pb/decoder.int.h
+++ b/upb/pb/decoder.int.h
@@ -225,6 +225,12 @@ struct upb_pbdecoder {
   char residual[12];
   char *residual_end;
 
+  /* Bytes of data that should be discarded from the input beore we start
+   * parsing again.  We set this when we internally determine that we can
+   * safely skip the next N bytes, but this region extends past the current
+   * user buffer. */
+  size_t skip;
+
   /* Stores the user buffer passed to our decode function. */
   const char *buf_param;
   size_t size_param;