Python renamed fields (e.g. "type_") break text_format.Parse

[soumyasr-google]

Environment details

• OS type and version: Linux 6.16.12-1rodete2-amd64
• Python version: python --version: Python 3.13.7
• pip version: pip --version: pip 25.3
• gapic-generator version: pip show gapic-generator: Version: 1.30.0

Steps to reproduce

1. Import a cloud sdk that has fields in the list of reserved names
2. Try to load textproto using text_format.Parse

Since the same textproto output loads fine in other languages (verified with go, for example), this seems like a bug in the python version of the generator, specifically that the proto descriptor returned by .pb() does not have the same field names as the actual proto.

There should be some way for users to parse textproto files that correspond to the released protos using the python sdk.

Code example

from google.protobuf import text_format  
from google.cloud import dlp_v2

# inlined for easy repro, actual contents would come from the Google-provided GCS object
text_proto_content = """  
findings {  
    quote: "Test"  
    location {  
        container {  
            type: "Google Cloud Storage"  
        }  
    }  
}
"""

output_message = dlp_v2.SaveToGcsFindingsOutput.pb()()
text_format.Parse(text_proto_content, output_message)

Stack trace

google.protobuf.text_format.ParseError: 6:7 : Message type "google.privacy.dlp.v2.Container" has no field named "type".

[soumyasr-google]

(Related to internal bug b/461709077)

[daniel-sanche]

Marking this as a feature request, since the current behaviour is a known limitation, which was presumably a trade-off of the current design.

But it would be great to support test to proto for this kind of use-case, so it would be worth taking a look at this at some point