* Pydantic V2 support
* Support Python 3.8
* Mark as classmethod
* Remove max int validation
* Run poe format
* Merge develop
* Revert dataclasses import
* Fix revert
PLACEHOLDER is a specific instance of an object, the test here should be "is not" instead of "!="
I am experimenting with adding ndarray support, and the equality test here causes problems.
Continuing work from #484 by @kevinaud to fix#419
* Implementing pickle methods
* Implement __copy__, __reduce__, and fix __setstate__
* Moved pickling tests into their own file
* Add test for caching
* Add support for streaming delimited messages
This allows developers to easily dump and load multiple messages
from a stream in a way that is compatible with official
protobuf implementations (such as Java's
`MessageLite#writeDelimitedTo(...)`).
* Add Java compatibility tests for streaming
These tests stream data such as messages to output files, have a
Java binary read them and then write them back using the
`protobuf-java` functions, and then read them back in on the Python
side to check that the returned data is as expected. This checks
that the official Java implementation (and so any other matching
implementations) can properly parse outputs from Betterproto, and
vice-versa, ensuring compatibility in these functions between the
two.
* Replace `xxxxableBuffer` with `SupportsXxxx`
This change ensures that deprecation warnings are only raised when
either a deprecated field is explicitly set or a deprecated message is
initialised.
Resolves: #347
* Update protobuf pregenerated files
* Update grpcio-tools to latest version
* Implement proto3 field presence
* Fix to_dict with None optional fields.
* Add test with optional enum
* Properly support optional enums
* Add tests for 64-bit ints and floats
* Support field presence for int64 types
* Fix oneof serialization with proto3 field presence (#292)
= Description
The serialization of a oneof message that contains a message with fields
with explicit presence was buggy.
For example:
```
message A {
oneof kind {
B b = 1;
C c = 2;
}
}
message B {}
message C {
optional bool z = 1;
}
```
Serializing `A(b=B())` would lead to this payload:
```
0A # tag1, length delimited
00 # length: 0
12 # tag2, length delimited
00 # length: 0
```
Which when deserialized, leads to the message `A(c=C())`.
= Explanation
The issue lies in the post_init method. All fields are introspected, and
if different from PLACEHOLDER, the message is marked as having been
"serialized_on_wire".
Then, when serializing `A(b=B())`, we go through each field of the
oneof:
- field 'b': this is the selected field from the group, so it is
serialized
- field 'c': marked as 'serialized_on_wire', so it is added as well.
= Fix
The issue is that support for explicit presence changed the default
value from PLACEHOLDER to None. This breaks the post_init method in that
case, which is relatively easy to fix: if a field is optional, and set
to None, this is considered as the default value (which it is).
This fix however has a side-effect: the group_current for this field (the
oneof trick for explicit presence) is no longer set. This changes the
behavior when serializing the message in JSON: as the value is the
default one (None), and the group is not set (which would force the
serialization of the field), so None fields are no longer serialized in
JSON. This break one test, and will be fixed in the next commit.
* fix: do not serialize None fields in JSON format
This is linked to the fix from the previous commit: after it, scalar
None fields were not included in the JSON format, but some were still
included.
This is all cleaned up: None fields are not added in JSON by default,
as they indicate the default value of fields with explicit presence.
However, if `include_default_values is set, they are included.
* Fix: use builtin annotation prefix
* Remove comment
Co-authored-by: roblabla <unfiltered@roblab.la>
Co-authored-by: Vincent Thiberville <vthib@pm.me>
Byteslike objects (like memoryview) do not have a decode function defined.
Instead, a string may be created from them by passing them to the str
constructor along with an encoding.
- Enable oneof_enum test case that passes now (removed the xfail)
- Switch from toml to tomlkit as a dev dep for better toml support
- upgrade poethepoet to latest stable release
- use full table format for poe tasks to avoid long lines in pyproject.toml
- remove redundant _WrappedMessage class
- fix various Mypy warnings
- reformat some comments for consistent line length
- Added support for the custom double values from
the protobuf json spec: "Infinity", "-Infinity", and "NaN"
- Added `infinite_floats` test data
- Updated Message.__eq__ to consider nan values
equal
- Updated `test_message_json` and
`test_binary_compatibility` to replace NaN float
values in dictionaries before comparison
(because two NaN values are not equal)
This means the betterproto plugin no longer needs to depend durectly on
protobuf.
This requires a small runtime hack to monkey patch some google types to
get around the fact that the compiler uses proto2, but betterproto
expects proto3.
Also:
- regenerate google.protobuf package
- fix a regex bug in the logic for determining whether to use a google
wrapper type.
- fix a bug causing comments to get mixed up when multiple proto files
generate code into a single python module
* Implement Message.__bool__ with similar semantics to a collection, such that any value being set on the message (i.e. having a non-default value) make the Message value truthy .
Co-authored-by: nat <n@natn.me>
