pkg/net/rpc$ gotest under ThreadSanitizer
(http://code.google.com/p/data-race-test/wiki/ThreadSanitizerGo) outputs the following
error:

WARNING: Possible data race during write of size 1 at 0xf8400010d0: {{{
   T2 (L{}):
    #0  net/rpc..input ~/go/src/pkg/net/rpc/client.go:132
    #1  runtime.goexit ~/go/src/pkg/runtime/proc.c:266
  Concurrent read(s) happened at (OR AFTER) these points:
   T1 (L{}):
    #0  net/rpc..Call ~/go/src/pkg/net/rpc/client.go:286
    #1  net/rpc.TestClientWriteError ~/go/src/pkg/net/rpc/server_test.go:487
    #2  testing.tRunner ~/go/src/pkg/testing/testing.go:254
    #3  runtime.goexit ~/go/src/pkg/runtime/proc.c:266
  Location 0xf8400010d0 is 48 bytes inside a block starting at 0xf8400010a0 of size 80 allocated by T1 from heap:
    #0  net/rpc..Go ~/go/src/pkg/net/rpc/client.go:255
    #1  net/rpc..Call ~/go/src/pkg/net/rpc/client.go:285
    #2  net/rpc.TestClientWriteError ~/go/src/pkg/net/rpc/server_test.go:487
    #3  testing.tRunner ~/go/src/pkg/testing/testing.go:254
    #4  runtime.goexit ~/go/src/pkg/runtime/proc.c:266
}}}

The root issue is that both send() and input() may set Call.Error and send to Call.Done.
This leads to 2 serious problems. First, there is a read-write data race (that the tool
reports) on Call.Error, since it's an interface the race can lead to arbitrary memory
corruptions and crashes. Second, if a user reuses Done channel, it will be totally
confused by two notifications about completion of a single Call.