Clean up treating MessageDelta in ChatModels.ChatGoogleAI

[nallwhy]

No description provided.

[brainlid]

Hi @nallwhy!

These look like some significant changes. Is this a case of the Google model API changing? Or did it not work correctly before?

And can you confirm that these changes are working correctly for you against a live server?

[nallwhy]

This wasn’t a Google API change. It looks like some incorrect code had lingered, and this PR cleans that up.

What changed and why:

• role mapping — Updated per Google’s Content docs (https://ai.google.dev/api/caching#Content) so we’re setting roles correctly.
• Removed do_process_response/3 — That clause was never reachable; all callers match the functions above it, so the extra function could never be invoked.
• Finish reason handling — We were always marking MessageDelta as incomplete due to using a wrong key (complete). Now we correctly map Google’s finishReason and set completion state as intended.

I’ve already applied these changes in production code using both streaming and non-streaming flows, and they work correctly in both cases.

In the previous code, even when finishReason was present, the status of MessageDelta was set to incomplete. I’m curious what the intention behind that was. In ChatOpenAI, it seems that the status is set to complete based on the finishReason.

[nallwhy]

Sample stream

DECODED STREAM DATA: {
  [
   % {
      "candidates" => [
       % {
          "content" => % {
            "parts" => [
             % {
                "functionCall" => % { "args" => % {}, "name" => "current_datetime" },
                "thoughtSignature" => "CiQBVKhc7n3VmT2ZrGFTpttOeSFQJRKz82lAZ/hrQPU0Jze8ifoKZgFUqFzuVVfE3aYh3N/CrOTZ2wITKfi/CF0oAyo+ESh1hecgQNsyikC7fZwMs4bfw4c9J0nHWEDmgCg0t78ZmLxKepvaMRyXDjhn1eQgZ+x3nkGMkM4DOhccoboZ4kpjauJ+0UUS2wqhAQFUqFzuunrV3v4M420/JEhr2nNxS0/TgwkGUz4uI5ED4mdoXLl0kNW1iexW5UN+bzVA3034G5n6ZQwwkyf/ul5w4UMszw2+oZ+xOiZJHVUWuNcRCPsPaTcrk3KFsIchtASvxnSJtF8EZkZByiY6xgTRsxQRvIx+zbEVNp0RR06TKlRTUV6Gu4U4dVpOAdIAzRw7Gy87+6m5dkT7G1clkI7x"
              }
            ],
            "role" => "model"
          },
          "finishReason" => "STOP",
          "index" => 0
        }
      ],
      "modelVersion" => "gemini-2.5-flash",
      "responseId" => "ykyRaNTaE7WcqtsP2vHd6QU",
      "usageMetadata" => % {
        "candidatesTokenCount" => 10,
        "promptTokenCount" => 41,
        "promptTokensDetails" => [% { "modality" => "TEXT", "tokenCount" => 41 }],
        "thoughtsTokenCount" => 46,
        "totalTokenCount" => 97
      }
    }
  ], ""
}

DECODED STREAM DATA: {
  [
   % {
      "candidates" => [
       % {
          "content" => % {
            "parts" => [% { "text" => "The current time is" }],
            "role" => "model"
          },
          "index" => 0
        }
      ],
      "modelVersion" => "gemini-2.5-flash",
      "responseId" => "y0yRaIytCaf6qtsP8e7n-Q8",
      "usageMetadata" => % {
        "candidatesTokenCount" => 4,
        "promptTokenCount" => 103,
        "promptTokensDetails" => [% { "modality" => "TEXT", "tokenCount" => 103 }],
        "totalTokenCount" => 107
      }
    }
  ], ""
}
DECODED STREAM DATA: {
  [
   % {
      "candidates" => [
       % {
          "content" => % {
            "parts" => [% { "text" => " 00:14:02 UTC on August 5, 20" }],
            "role" => "model"
          },
          "index" => 0
        }
      ],
      "modelVersion" => "gemini-2.5-flash",
      "responseId" => "y0yRaIytCaf6qtsP8e7n-Q8",
      "usageMetadata" => % {
        "candidatesTokenCount" => 21,
        "promptTokenCount" => 103,
        "promptTokensDetails" => [% { "modality" => "TEXT", "tokenCount" => 103 }],
        "totalTokenCount" => 124
      }
    }
  ], ""
}
DECODED STREAM DATA: {
  [
   % {
      "candidates" => [
       % {
          "content" => % { "parts" => [% { "text" => "25." }], "role" => "model" },
          "finishReason" => "STOP",
          "index" => 0
        }
      ],
      "modelVersion" => "gemini-2.5-flash",
      "responseId" => "y0yRaIytCaf6qtsP8e7n-Q8",
      "usageMetadata" => % {
        "candidatesTokenCount" => 24,
        "promptTokenCount" => 103,
        "promptTokensDetails" => [% { "modality" => "TEXT", "tokenCount" => 103 }],
        "totalTokenCount" => 127
      }
    }
  ], ""
}

[brainlid]

Thanks for the clarification @nallwhy. This PR has a slight conflict with another PR of yours I just merged. After you resolve it (because you can test it and verify it works as expected), I'll merge this.

[nallwhy]

@brainlid I've rebased the commits.

[brainlid]

Thanks @nallwhy!
❤️💛💙💜