IndexedDB: Fixed support for empty blobs.

net/base/file_stream_context_posix.cc

Index: net/base/file_stream_context_posix.cc
diff --git a/net/base/file_stream_context_posix.cc b/net/base/file_stream_context_posix.cc
index 44f59b1ef3b82c9b3032edcbef331901dbc4e518..268fe92e942074acf653cf25efbe74f53fd84b2d 100644
--- a/net/base/file_stream_context_posix.cc
+++ b/net/base/file_stream_context_posix.cc
@@ -103,6 +103,8 @@ FileStream::Context::IOResult FileStream::Context::ReadFileImpl(
 FileStream::Context::IOResult FileStream::Context::WriteFileImpl(
     scoped_refptr<IOBuffer> buf,
     int buf_len) {
+  if (buf_len == 0)

[jsbell, 2015/02/24 18:37:36]

Can you explain why we need this special case? Does
WriteAtCurrentPosNoBestEffort error out if the length is zero? If so, do we want
to change the semantics of this API, or short-circuit closer to the Blob code?

(This seems entirely reasonable, I'm just nervous...)

[cmumford, 2015/02/24 22:40:52]

buf refers to a NULL IOBuffer in this case. So operator-> is what's failing
without this check. I'd rather stay out of net/base, and will see if I can do
that. However, going two levels up the callstack FileStream::Write allows for
writing zero bytes (see DCHECK_GE), so maybe the solution should be in net/base.
What do you think?

[jsbell, 2015/02/25 00:10:30]

Although you took a different approach, it is weird if FileStream::Write is
effectively calling WriteFileImpl(nullptr, 0) !

+    return IOResult(0, 0);
   int res = file_.WriteAtCurrentPosNoBestEffort(buf->data(), buf_len);
   if (res == -1)
     return IOResult::FromOSError(errno);