fix Gemini tool calling

[zikajk]

• Normalizes ECA’s interleaved tool history into OpenAI/Gemini-compatible turns: all parallel tool_calls first, then all tool outputs, without merging sequential tool-loop rounds (uses :tool-turn-id boundaries).

• Preserves Gemini extra_content.google.thought_signature across turns (stored as :external-id and emitted back on tool_calls), preventing 400 INVALID_ARGUMENT on Gemini 3 Flash tool calls.

• Makes streamed tool execution deterministic even when providers omit tool_calls[].index (fallback :stream-order) and avoids tool-loop stalls on invalid JSON tool arguments.

• Adds unit tests covering ordering, tool-turn boundaries, thought_signature preservation, and deterministic execution.

• I added a entry in changelog under unreleased section.

[ericdallo]

Looks good, thank you!

[stsquad]

Hmm /init still fails:

[server] Starting server...
[DB] No existing DB cache found for /home/alex/.cache/eca/db.transit.json
:db/read-cache 0ms
:db/read-cache 14ms
:eca/initialize 15ms
[LLM-API] Default LLM model 'anthropic/claude-sonnet-4.5' decision ':default'
[LLM-API] Default LLM model 'anthropic/claude-sonnet-4.5' decision ':default'
:eca/initialized 280ms
:eca/chat-query-commands 0ms
:eca/chat-prompt 15ms
:db/upsert-cache 1ms
[TOOLS] Calling tool 'eca__directory_tree' with args '{"path" "/home/alex/mysrc/hubi.git", "max_depth" 2}'
[OPENAI-CHAT] 3949 Unexpected response status: %s body: %s 400 [{
  "error": {
    "code": 400,
    "message": "Please ensure that function call turn comes immediately after a user turn or after a function response turn.",
    "status": "INVALID_ARGUMENT"
  }
}
]
{:message LLM response status: 400 body: [{
  "error": {
    "code": 400,
    "message": "Please ensure that function call turn comes immediately after a user turn or after a function response turn.",
    "status": "INVALID_ARGUMENT"
  }
}
]}
:db/upsert-cache 0ms
:db/upsert-cache 0ms

with:

/init
Rollback chat to before this message

Thought 0.00 s
**Formulating the Plan**

I'm currently focused on the structure of the `AGENTS.md` file. I'll need to define sections for build, lint, and test commands, especially a single test run. Code style guidelines are also a priority. I'm exploring the codebase to understand the build process.




Listing file tree ✅ 0.02 s

Tool: directory_tree
Server: eca
Arguments: 
  path: "/home/alex/mysrc/hubi.git"
  max_depth: 2

Output: /home/alex/mysrc/hubi.git
 Makefile
 README.org
 TODO.org
 docs
  screenshot.png
 hubi-configure.el
 hubi-ert.el
 hubi.el

1 directories, 7 files


LLM response status: 400 body: [{
  "error": {
    "code": 400,
    "message": "Please ensure that function call turn comes immediately after a user turn or after a function response turn.",
    "status": "INVALID_ARGUMENT"
  }
}
]

it looks like its the latest binary:

➜  ps ax | grep eca
2526115 ?        Ssl    0:00 /home/alex/.emacs.d/eca/eca server
2528005 pts/9    S+     0:00 grep --color=auto eca
🕙13:41:22 alex@draig:qemu.git/builds/all  on  cpus/cleanup-cpu-reset [$!?] 
➜  /home/alex/.emacs.d/eca/eca --version
eca 0.90.1
🕙13:41:28 alex@draig:qemu.git

[zikajk]

[ericdallo]

@zikajk does /init works for you?

[zikajk]

@ericdallo It doesn't :-(. Interesting...

[zikajk]

I will try to take a look soon why is that.