* size computations
* add `Protobuf.computeSize` to compute the encoded size of an object
* readd `VarIntLengthPrefix` support
* don't expose loosely typed `writeValue` for objects that might
conflict with other serializers
* ensure all `writeValue` and `writeField` are tagged with a `codec`
type
* avoid writing all-defaults proto3 object fields
* Fix sizer for reference
* Create TableObject in a template
---------
Co-authored-by: Ludovic Chenut <ludovic@status.im>
This is a cleanup / rewrite of the implementation increasing
interoperability with other implementations and fixing several bugs /
security issues
* remove special handling of nim:isms (`ptr`, `ref`, `Option`, `HashSet`
etc) (#4)
* these deviate from "standard" protobuf behavior making the library
less interoperable with other langagues
* supporting "custom" types should be done as part of an extension
framework instead (that can support any type / collection instead of a
few hand-picked special cases)
* don't allow encoding scalars in the core encoder (#31)
* `codec` can be used to encode simple scalars
* switch to `libp2p/minprotobuf`-like encoding base, fixing several
bugs, crashes and inaccuracies (#30, #32, #33)
* move parsing support to separate import
* the parser is a heavy dependency
* allow unknown fields
* unknown fields should be given an extension point allowing the user
to detect / handle them - standard behavior for protobuf is to ignore
them
* work around several faststreams bugs (#22)
* remove machine-word-dependent length prefix options (#35)
* actually, remove varint length prefix too for now (due to
faststreams bugs)
* update version
* verify that strings are valid utf-8 on parse
* fix warnings
* truncate values like C++ version
* allow unpacked fields in proto3
* protobuf2/3 -> proto2/3
* update docs
There's lots left to do here in terms of tests and features:
* Almost all tests are roundtrip tests - meaning they check that writing
and reading have the same bugs (vs outputting conforming protobuf)
* There are very few invalid-input tests
* There's a beginning of an extension mechanism, but it needs more work
* It's generally inefficient to copy data to a protobuf object and then
write it to a stream - the stream writers should probably be made more
general to handle this case better (either via callbacks or some other
"builder-like" mechanism - projects currently using `minprotobuf` will
likely see a performance regression using this library
* `required` semantics are still off - a set/not-set flag is needed for
every field in proto2
* possibly, when annotated with proto2, we should simply rewrite all
members to become `PBOption` (as well as rename the field)
Jacek Sieka
Zahary Karadjov
Luke Parker
Joey