On Fri, Oct 7, 2016 at 2:16 PM, Arpit Baldeva <abald...@gmail.com> wrote:
> Thanks for the info. > > I feel like without pack/unpack/Is method, the utility of Any will > diminish. For example, the rpc status proto (https://github.com/ > googleapis/googleapis/blob/master/google/rpc/status.proto) uses repeated > Any field. It'd not be possible to write code like one described here - > https://developers.google.com/protocol-buffers/docs/proto3#any because > you won't know if it is safe to convert value to a give message. I also > came across this post after my post which marks the request as a bug > currently - https://github.com/google/protobuf/issues/1974 > What you're saying makes sense, we might want to consider just updating Any to have first-class support for MessageLite. In C++ this would be straightforward but in Java, for example, we would need to think carefully about how to do it because in Java lite we don't currently have the message names available at runtime. Regarding the future of GetTypeName, though it has overhead, feel like it > could have many utilities outside of the Any support as well. I don't have > concrete use case in mind though as I am just starting on protobuf. This > brings another important question that I was wondering if somebody already > has data around. There are two options for reducing code bloat. One is Lite > and another is code_size. I understand that lite reduces code bloat by > removing descriptors/reflections related code (thereby reducing the library > size) and code_size reduces the code bloat by generating less code per > message but puts descriptors/reflectors back in(shared code). And the > recommendation is to choose code_size option if number of message are much > higher compared to the overhead caused by the size of lib. Are there any > benchmarks around what the size of the library is (and lite version) and > what is the per message overhead saved by the code_size option? And the > performance drop with code_size option? > Here's one way to break it down. SPEED: - Fixed overhead of full runtime (e.g. the Message class) - Per-message overhead of generated parsing/serialization code - Per-message overhead of generated descriptors LITE_RUNTIME: - Fixed overhead of lite runtime (e.g. includes MessageLite but not Message) - Per-message overhead of generated parsing/serialization code CODE_SIZE: - Fixed overhead of full runtime (e.g. the Message class) - Per-message overhead of generated descriptors SPEED and LITE_RUNTIME should be about the same speed because they both benefit from the fast generated code for parsing and serialization, while CODE_SIZE is much slower because it relies on reflection instead of generated code. My impression is that CODE_SIZE is not really a good choice unless you have an unusual situation where you have a large number of protos and are really tight on code size. A basic rule of thumb would be to use the default (SPEED) on servers and LITE_RUNTIME on mobile. I'm not sure offhand of the actual numbers for how binary size and speed differ between the three choices--Gerben (CC'd), do you happen to know some numbers for this question? -- You received this message because you are subscribed to the Google Groups "Protocol Buffers" group. To unsubscribe from this group and stop receiving emails from it, send an email to protobuf+unsubscr...@googlegroups.com. To post to this group, send email to protobuf@googlegroups.com. Visit this group at https://groups.google.com/group/protobuf. For more options, visit https://groups.google.com/d/optout.
