[veljjanoski]

Part of #633 (all of it except the proto2/proto3 features rule; see below).

DescriptorPool linked editions files that protoc 33.5 rejects in ValidateFieldFeatures and ValidateFileFeatures (descriptor.cc). Each case is now a new PoolError variant that names the field or file:

Input in an editions file	protoc message	New variant
a TYPE_GROUP field or extension	Group types are not allowed under editions.	GroupFieldInEditions
the packed option, either value	Field option packed is not allowed under editions.	PackedOptionInEditions
features.field_presence on a oneof member, a repeated or map field, or an extension; IMPLICIT on a message field	Oneof fields / Repeated fields / Extensions can't specify field presence; Message fields can't specify implicit presence.	InvalidFieldPresence
features.utf8_validation on a field that is not a string field or a map with a string key or value	Only string fields can specify utf8 validation.	InvalidUtf8Validation
features.message_encoding on a map field or a non-message field	Only message fields can specify message encoding.	InvalidMessageEncoding
file-level features.field_presence = LEGACY_REQUIRED	Required presence can't be specified by default.	LegacyRequiredFileDefault

The four protoc presence messages share one variant, as the two repeated_field_encoding messages already share InvalidRepeatedFieldEncoding. An extension set to LEGACY_REQUIRED still reports RequiredExtension.

Details that follow protoc:

• The field rules count only a feature the field sets itself, like the existing repeated_field_encoding check, so a file default passes over the fields it does not fit. They skip the key and value fields of a map entry: protoc copies the map field's features onto them, so map<string, int32> m = 1 [features.utf8_validation = NONE] gives the int32 value field that feature too.
• The rules apply when the file's syntax is editions and its edition is 2023 or later. protoc does not apply them to an earlier edition: a hand-built file with syntax editions and EDITION_PROTO2 is accepted with a packed option and a group. A new private is_editions_file sets LinkScope::editions; an edition this crate does not define counts as a later one. The pool's older rules (the required label, the proto3 rules) still go by the syntax string; I left them as they are.
• PackedOptionInEditions is checked before InvalidPackedOption, so packed = true on an editions string field reports the editions rule.

I checked every case against protoc 33.5 with .proto files, and the edition gate with a hand-built descriptor set passed through protoc --descriptor_set_in.

Left out: protoc also rejects any features in a proto2 or proto3 file ("Features are only valid under editions."). The pool can't take that rule as it stands: buffa-codegen's open_enums_in and override_feature_in write features.enum_type = OPEN into proto2 enum and field descriptors, and with generate_reflection the generated descriptor_pool() loads them (for example buffa-test's open_enums.proto configs), so it would start to panic. That needs a decision on how the overrides should reach the embedded pool, so I left it for you. protoc's ValidateFileFeatures also rejects the java_string_check_utf8 file option in editions files; #633 does not list it, so it is not in this PR either.

Tests: new buffa-descriptor/tests/pool_editions_features.rs, with rejected and accepted cases for each rule. The accepted ones are the neighbours protoc accepts: a DELIMITED message field, presence on a singular field, utf8_validation on string fields and string maps, message_encoding on singular and repeated message fields, EXPLICIT/IMPLICIT file defaults, file defaults for all three features over every kind of field, and files with syntax editions and EDITION_PROTO2 or EDITION_PROTO3. Group fields cover extensions, edition 2024 and an undefined edition number, and one test checks that a rejected file leaves the pool unchanged. Each new condition was mutation-checked: disabling it fails a test.

Run locally on Windows (x86_64-pc-windows-gnu, toolchain 1.95.0, protoc 33.5):

• cargo fmt --all --check
• cargo clippy --workspace --all-targets -- -D warnings, and cargo clippy -p buffa-descriptor --all-targets --all-features -- -D warnings
• cargo test --workspace --all-features: all pass except two buffa-types tests that fail on Windows only, because SystemTime there has 100 ns resolution
• task doc (rustdoc -D warnings)
• the host buffa-descriptor commands of task check-nostd
• task conformance: all eight runs pass with 0 unexpected failures, the reflect run included (it loads the protoc-built conformance protos, editions ones too, into a DescriptorPool)

AI-usage disclosure: Claude (Anthropic) was used as assistant.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}