Data racing issue with heavy load

[QianYizhou]

Hi,
Your SPMC_Queue is really inspiring! But it seems there's still a data racing issue.

My test steps:

1. in multhread.cc, extend array size defined in struct Msg to make it heavier, e.g.,
   struct Msg
   {
   uint64_t tsc;
   uint64_t i[1];
   };
   to
   struct Msg
   {
   uint64_t tsc;
   uint64_t i[100];
   double f[100];
   };
2. compare Msg.f as well
3. Loop more times (1e6 -> 1e7)

Then these's a possibility of assert fault.
I think it may related to the write() operation: sees that operation does't care if the data is occupied by a reader -- the data may be changed by write() while reader is reading.

[leeshineSZ]

I also have this question. how to deal with the competition?while reader is reading, the writer is going to write the same area.