Change-Id: Ie70fb0a3632cdaa77dfd7427c5ea2684b8217512
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/708557
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Chressie Himpel <chressie@google.com>

I encountered this usage pattern in a production application
and noticed that when breaking the size calculation,
only the conformance_test fails (indirectly),
but there was no explicit test before this commit.

Change-Id: I15a125b7cff3364f24fe372e4a3a046bb7552413
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/710855
Reviewed-by: Chressie Himpel <chressie@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

Change-Id: I9cdc26a8bda576a76e93dc352a2a8bea9f504817
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/711200
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Chressie Himpel <chressie@google.com>

Without this, we allow non-option symbols from import option dependencies to be referenced inside of protos.  This can both break valid protos and allow invalid protos in various edge cases, and is inconsistent with protoc.

Change-Id: I12e657435173186b8fe4e84315fc2e1b9fd2dc6a
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/711015
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Michael Stapelberg <stapelberg@google.com>
Commit-Queue: Michael Stapelberg <stapelberg@google.com>

Change-Id: I4a786adbc7fc1a829e444dcaa0e43d35f42a3be4
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/713180
Reviewed-by: Chressie Himpel <chressie@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

… plugin.

Because protoc sends option dependencies to its plugins, this results in the go plugin enforcing the import path flag requirements on them.  This is unnecessary though, since a true option dependency will have no generated code importing it anyway.  This is especially problematic in (but not unique to) Bazel, since the aspect doesn't trigger on the option_deps attributes.

Change-Id: I3b8cec00e462f84ef02ba1ab6bdb4dbe843703d5
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/713342
Reviewed-by: Michael Stapelberg <stapelberg@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

Before this change, this test had a flakiness rate of 3 out of 1000 runs
(on Google’s internal testing infrastructure).

After this change, the test is not flaky, even when running with 50_000 runs.

testing.AllocsPerRun is meant to be robust against individual allocations
from background goroutines, but not in the way we called the function.

Change-Id: If99573734fb2de5d7128b5e61c2019867ccbb9b2
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/716520
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Christian Höppner <hoeppi@google.com>

proto3 legacy struct field tags are improperly parsed as proto2,
resulting in fd.L1.EditionFeatures.IsFieldPresence set to true when it
should have been false. This causes coderFieldInfo.isPointer to be true
for said fields. When the encode process (particularly
MessageInfo.sizePointerSlow) sees isPointer=true for the field, it
attempts to interpret the field as a pointer, which results in a "fatal
error: checkptr: misaligned pointer conversion" runtime error if the
field address happens to not be pointer-aligned. This only happens if
building with -race/-asan/-msan as they enable checkptr.

To avoid this issue, parse from proto3 edition defaults if the proto3
key is found in the struct field tag.

Change-Id: Id81ab38fe7efd437d0748a485653d27b1acc3afb
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/716360
TryBot-Bypass: Michael Stapelberg <stapelberg@google.com>
Auto-Submit: Nicolas Hillegeer <aktau@google.com>
Reviewed-by: Nicolas Hillegeer <aktau@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Run-TryBot: Damien Neil <dneil@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-by: Chressie Himpel <chressie@google.com>
Reviewed-by: Michael Stapelberg <stapelberg@google.com>

A user reported a data race when calling the IsMapEntry and Options accessors
concurrently: golang/protobuf#1701

Originally, I wanted to suggest moving the md.L1.IsMapEntry / md.L1.IsMessageSet
assignments into Message.unmarshalSeedOptions when I realized that these fields
are already set in unmarshalSeedOptions!

Therefore, we can entirely remove Message.unmarshalOptions in favor of
Message.unmarshalSeedOptions. All tests continue to pass, both within Go
Protobuf itself, and also all Google tests (Global Presubmit, cl/832261606).

Resolves golang/protobuf#1701

Change-Id: Ieb4dbddd11f4a123ba01d3c10fed4d16bee0791a
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/720980
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Lasse Folger <lassefolger@google.com>

… symbols.

This change adds g.AnnotateSymbol calls to generate Kythe metadata for extension variables, oneof case constants, and oneof Which methods. See b/461219786 for more details.

Change-Id: Id40e5492e7c32ec904f90c0ec1db7d8e781eaaed
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/723440
Reviewed-by: Lasse Folger <lassefolger@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

This will be useful in a follow-up change to reproduce a bug.

Change-Id: Ide5d174af24c054be75709b4ed77eb9956b411ec
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/727140
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Lasse Folger <lassefolger@google.com>

Change-Id: I6666a66f2ead3d53220e3ce941daa93b8408273f
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/727940
Reviewed-by: Lasse Folger <lassefolger@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

The UNSTABLE edition serves as a sandbox for all features in-development.
The --experimental_edition protoc flag is required to accept it.

Google-Internally, see go/edition-unstable for more details.

Originally authored by Rachel Goldfinger, who asked me to send the change.

Change-Id: If3acc284f70b7690e837d4746202823880a13355
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/727960
Reviewed-by: Lasse Folger <lassefolger@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

Authored by Rachel Goldfinger.

Change-Id: I57d211412850cc8e58f7d905770e82d2570e50f3
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/728681
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Lasse Folger <lassefolger@google.com>

Change the text-format parser to allow for more characters and formats
in the type URL prefixes of expanded Any protos. This follows a recent
change to the text-format spec [1,2] and the Go implementation is now
almost exactly implementing this spec (with some exceptions to ensure
backwards compatibility).

Note that we are also making the whitespace handling between [ and ]
more lenient compared to the current implementation. This is also
in line with the new spec.

Refs:
- [1] https://protobuf.dev/reference/protobuf/textformat-spec/#characters
- [2] https://protobuf.dev/reference/protobuf/textformat-spec/#field-names
- go/protobuf-improved-type-url-support

Change-Id: I55055becec0248c1d161e776f1937eaaa4af2066
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/726780
Reviewed-by: Michael Stapelberg <stapelberg@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Damien Neil <dneil@google.com>

The recursion depth limit was introduced in https://go.dev/cl/385854,
but without any tests.

When I added a basic test, I noticed that lazily decoded messages
and extensions circumvent the recursion limit:

At validation time, the limit was not checked. Later, lazy unmarshaling
(MessageInfo.unmarshalField) starts with the default recursion limit
(10'000) by copying impl.lazyUnmarshalOptions.

Before this commit, a 9999-deeply nested message could have an extension
which contains another 9999 nested messages, and so on.

I also noticed that maps did not count as a recursion (now also fixed).

(Google-internally tested globally as cl/842145661)

Change-Id: Ica036d8625225b21473de58ecd1afc8414eb6784
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/728680
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Lasse Folger <lassefolger@google.com>

Change-Id: I451fcbd61dc2be12f9df9cb8efd44d2649307b17
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/729640
Reviewed-by: Lasse Folger <lassefolger@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>