Descriptor pools

A pool holds the message types zuproto can decode and encode. It is built from serialised FileDescriptorSets, which protoc and buf write:

Usage

proto_pool(x = NULL)

proto_pool_add(pool, x)

Arguments

x

A serialised FileDescriptorSet: a raw vector, a path to a file, or a connection. NULL for the well-known types only.

pool

A pool from proto_pool().

Value

proto_pool() returns a pool. proto_pool_add() returns pool, invisibly, after adding x to it.

Details

protoc --include_imports --descriptor_set_out=schema.binpb schema.proto
buf build -o schema.binpb

Every pool starts with the well-known types (google/protobuf/*.proto, including descriptor.proto), so a set need not include them.

Descriptor sets are trusted input: upb checks them, but less thoroughly than protoc does. Load sets you or your tools produced, not ones from an untrusted source. Message bytes, by contrast, are never trusted.

Files are added in dependency order, whatever order the set lists them in. A file already in the pool is skipped if its contents are identical and is a zuproto_pool_conflict if they are not; the same holds for a file listed twice. A file importing one that is neither in the set nor in the pool is a zuproto_pool_error naming both, and nothing from the set is added. Extensions are loaded with the files that declare them.

See also

proto_types() and proto_fields() to see what a pool holds.

Examples

pool <- proto_pool()
pool
#> <proto_pool: 11 files, 61 message types>
"google.protobuf.Timestamp" %in% proto_types(pool)$name
#> [1] TRUE