Re: [protobuf] Zig Zag Encoding (vs two's complement encoding)
The reason ZigZag encoding evolved as it did is because it can be applied as a separate layer on top of varint encoding. Originally, ZigZag was not officially part of the protobuf spec, but just something some people did manually to encode negative numbers efficiently. Eventually we made it official. Your encoding would work equally well, but cannot really be implemented independently of varint itself, therefore could not have evolved into existence in this way. On Fri, Jan 14, 2011 at 12:07 AM, David Srbecky dsrbe...@gmail.com wrote: 126 is 0111 1110 in binary. We truncate it form the left to groups of seven: 000 110 and then encode it as usual (least significant group first): 1 110 0 000. It takes two bytes to encode, but so does the zig-zag encoding. In fact, the zig-zag encoding and the two's complement encoding always take the same number of bytes to encode a given number. -- You received this message because you are subscribed to the Google Groups Protocol Buffers group. To post to this group, send email to protobuf@googlegroups.com. To unsubscribe from this group, send email to protobuf+unsubscr...@googlegroups.comprotobuf%2bunsubscr...@googlegroups.com . For more options, visit this group at http://groups.google.com/group/protobuf?hl=en. -- You received this message because you are subscribed to the Google Groups Protocol Buffers group. To post to this group, send email to protobuf@googlegroups.com. To unsubscribe from this group, send email to protobuf+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/protobuf?hl=en.
Re: [protobuf] Zig Zag Encoding (vs two's complement encoding)
126 is 0111 1110 in binary. We truncate it form the left to groups of seven: 000 110 and then encode it as usual (least significant group first): 1 110 0 000. It takes two bytes to encode, but so does the zig-zag encoding. In fact, the zig-zag encoding and the two's complement encoding always take the same number of bytes to encode a given number. -- You received this message because you are subscribed to the Google Groups Protocol Buffers group. To post to this group, send email to protobuf@googlegroups.com. To unsubscribe from this group, send email to protobuf+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/protobuf?hl=en.
[protobuf] Zig Zag Encoding
Hello, I am wondering why the integers are encoded as zig zags. It would seem much more intuitive to me to just truncate the number from the left. That is, store -1 as 0111, -2 as 0110, 1 as 0001, 2 as 0010 and so on. I am not proposing to change it. I am just curious about the design. Regards, David Srbecky -- You received this message because you are subscribed to the Google Groups Protocol Buffers group. To post to this group, send email to protobuf@googlegroups.com. To unsubscribe from this group, send email to protobuf+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/protobuf?hl=en.
Re: [protobuf] Zig Zag Encoding
maybe I'm being slow (it is late here), but what would 126 look like?
You could say ah, pad to the expected length, but then -1 is encoded as
1110 - and as I see it, *that* is
what zip-zag attempts to avoid.
It also solves the problem of switching from 32 to 64 bit, as a zig-zag
number shouldn't change (IIRC).
Marc
On 13 January 2011 19:58, David Srbecky dsrbe...@gmail.com wrote:
Hello,
I am wondering why the integers are encoded as zig zags. It would
seem much more intuitive to me to just truncate the number from the
left. That is, store -1 as 0111, -2 as 0110, 1 as 0001, 2
as 0010 and so on.
I am not proposing to change it. I am just curious about the design.
Regards,
David Srbecky
--
You received this message because you are subscribed to the Google Groups
Protocol Buffers group.
To post to this group, send email to protobuf@googlegroups.com.
To unsubscribe from this group, send email to
protobuf+unsubscr...@googlegroups.comprotobuf%2bunsubscr...@googlegroups.com
.
For more options, visit this group at
http://groups.google.com/group/protobuf?hl=en.
--
Regards,
Marc
--
You received this message because you are subscribed to the Google Groups
Protocol Buffers group.
To post to this group, send email to protobuf@googlegroups.com.
To unsubscribe from this group, send email to
protobuf+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/protobuf?hl=en.
