proc-macro2 0.3 tries to be compatible with the non-nightly
0.2 versions, even if the nightly f...

Remove the indirect and unneeded dependency on failure_derive. In our
prost-based application, ...

This is a followup / fix for e0317f83958892d716e99423f07525db5c7469e6.

While attempting to pu...

Signed-off-by: Brian Smith <[email protected]>

* Embed protobuf tools & includes instead of downloading during the build.

Implement the prop...

This drops the `ordermap` dependency, for example.

Contributed on behalf of Buoyant, Inc.

...

It isn't useful to convert the environment variable to Unicode since
`Command` deals with `OsSt...

Use env_logger 0.5 everywhere, instead of a mix of 0.5 and 0.4, to
avoid needing to download an...

Changing ServiceGenerator::{generate, finalize} to take `&mut self` is
motivated by tower-rs/tow...

Also fixes a bug in bytes default values with octal escapes.

* Add ServiceGenerator::finalize

It can be used to output things that need to exist only once...

* Allow generation of attributes on messages and fields

This allows messages to be for exampl...

The expected usage wasn't clear to me, it is likely it wouldn't be clear
for others.

If messages or oneof types contain an enum (encoded as i32), try to
convert it to the real enum ...

That matches the version in prost-derive to not depend on two version of the itertools crate.

Currently, `oneof`s that reference a parent type result in compile
errors.

By replicating the `...

These nesting tests are duplicates of the upstream
test_messages_proto[2,2] tests, however they ...

The constants were causing issues with docs.rs doc generation.
Should fix #28.

This directs Cargo to discard cache builds if the PROTOC or
PROTOC_INCLUDE environment variables...

Fixes a bunch of clippy warnings.

Removes a circular prost-build <-> prost-types dependency

Adds `Comments::append_with_indent` which makes it easy to handle
comments correctly in `Service...

This changes the generated accessor methods on optional enum fields to
return the default enum v...

This is a breaking change, because heck is aggressive about pruning
prefix and suffix underscore...

This fixes up some issues introduced by 05293211f2cc2df6ef3:

* Use prost-types by default in pr...

Allows two packages with the same terminal name to be mixed in a
project, like 'foo.bar' and 'qu...

fixes danburkert/prost#5

* Folds the prost-codegen crate into prost-build.
* Removes the protoc-gen-prost tool. It wasn't...

The protobuf crate downloads, compiles, and installs the
[C++ Protobuf project][1], including th...

This tweaks prost-build to download protoc during its own build, if
necessary, instead of downlo...

This also sneaks in a couple bug fixes which make this possible:

1) `group` fields are now filt...

This results in a pretty good improvement in message encoding performance:

name ...

Copies the varint decoding optimizations from the cpp protobuf
implementation. Performance is si...

This is a relatively modest change. Turns out removing the assert from the
inner loop still make...

Based on the cpp Protobuf implementation. This is a big win: encoded_len_varint
is now constant ...

The Cursor type is no longer used in tests.

This follows Cargo conventions. Cargo recently began warning when the
convention is not followed...

These were added arbitrarily; they can always be added back later if
they are shown to make a di...

The 'dummy const' pattern was copied from SerDe[1], but it causes
rustdoc not to generate docume...

Fixes #19

This is a breaking change.

This makes the API much nicer to work with. See the included fixups
for examples.

This is a bre...

The presence of Take in the public API was an unfortunate leak of an
internal implementation det...

Removes io::Error in favor of prost-specific error types. io::Error is
a bit heavyweight for pr...

Allows prost-build to be used with a system protoc.

This also expands the prost-build documentation.

This allows the rountrip checker to be simplified, and fixes some
remaining bugs around NaN equa...

As suggested by @stepancheg in [1].

This allowed a bunch of the `unwrap()` calls in codegen to ...