Alignment of var data and repeating groups

[mjpt777]

It seems the spec does not provide an option for alignment for the starting of repeating groups or var data. For example, the var data header could be a length field of type uint32 that is followed by a string of 13 bytes, then the following repeating group or var data field would have it header unaligned for the blockLength/length member.

It possible solution would be to have an alignment attribute for the schema that would be used to align fields such as repeating group headers and var data header on alignment boundaries.

[donmendelson]

That makes sense. We could also let an alignment instruction apply to a field as well as repeating group or var data. Then a message designer could just specify desired alignment instead of having to compute offset from the beginning of the block. Of course, an implementation still has to compute offsets in a buffer, but let a computer do the work.

[mjpt777]

As a minimum with alignment, if enabled, then align to the size of the type like C does.

[donmendelson]

Offset attribute is from the start of a block (message root or repeating group entry). The proposed alternative to offset would align a field or repeating group entry on a desired boundary. @mjpt777 , is it safe to assume for common languages and platforms that the start of a message is always properly aligned, or should alignment apply to messages as well?

[mjpt777]

We cannot assume the start of a message will be aligned with any language when it is encoded to a byte array. Most languages align on the size of the type. Some from of framing alignment will be required. For example, in Aeron we align messages on a 32 byte boundary for this and because it gets good processor support for copy operations.