Fix misaligned buffer allocation method (fixes #1367)

[dmandalidis]

No description provided.

[mp911de]

Care to elaborate? Multiplying a float with an int value uses float multiplication for both values. After the multiplication, the value gets downcasted to int again.

It seems we need to differentiate whether we use ByteBufUtil.writeUtf8(…) for writing or our own method. ByteBufUtil requires a bit more memory for writing whereas the current implementation tries to optimize for memory.

[dmandalidis]

@mp911de I think that the problem is that both Lettuce and Netty are allocating buffers in a different way. This fix makes StringCodec aligned with the Netty way, but it seems to be a workaround in general. Is there anything I can do to help on this?

[mp911de]

Let's use the same estimation semantics for the UTF-8 case like netty does. That is calling ByteBufUtil.utf8MaxBytes(…) in sizeOf(…) when the codec uses utf8 mode.

[dmandalidis]

@mp911de thanks, would it be possible to include it in an upcoming 5.3.x release? (no rush, we have a workaround)

[mp911de]

Sure, the fix is backported to 5.3 and we will release a new version in about a week.