Random exception by reading wrong position about crystal-mysql HOT 3 CLOSED

It seems that the last few bytes of null bitmap is "reused".

According to tcpdump, I received the following bytes for the failed row.

3500 0093 00a0 5300 b707

But my print debug in IO#read_fully?(slice : Bytes) tells me that

• It reads 4 bytes 35 00 00 93 as the beginning of the new row
• Then read 3 bytes a0 53 00 for null bitmap.
• Then read 4 bytes 53 00 b7 07 for id field, which is expected to be b7 07 00 00 = 1975
  So the 2 bytes "53 00" repeated.

Seems a crystal's bug.

from crystal-mysql.

I found the root cause. It only happens when a read request to mysql TCP socket is issued across buffer boundary.
For example, when there are only 2 bytes remaining in the buffer, if 4 bytes are requested, then IO::Buffered read only the first 2 bytes and do nothing for the last 2 bytes, returning 2 as number of bytes read. crystal-mysql still can read 4 bytes from the slice but the content of the last 2 bytes are unpredictable.

One solution is to check the number of bytes read and call @io.read() again if necessary.

But I feel the behavior of IO::Buffered is not intuitive. I think it should work transparently.

from crystal-mysql.

crystal-lang/crystal#4796
So it's not crystal's bug.

from crystal-mysql.