Dispatch tools when code mode is not awaited directly#14437
Merged
pakrym-oai merged 5 commits intomainfrom Mar 12, 2026
Merged
Conversation
pakrym-oai
force-pushed
the
pakrym/code-mode-worker-dispatch
branch
from
c472815 to
9c4ffb8
Compare
pakrym-oai
changed the title
cconger
reviewed
Mar 12, 2026
codex-rs/core/src/tools/code_mode.rs
Outdated
| section.push_str("- Import nested tools from `tools.js`, for example `import { exec_command } from \"tools.js\"` or `import { ALL_TOOLS } from \"tools.js\"` to inspect the available `{ module, name, description }` entries. Namespaced tools are also available from `tools/<namespace...>.js`; MCP tools use `tools/mcp/<server>.js`, for example `import { append_notebook_logs_chart } from \"tools/mcp/ologs.js\"`. Nested tool calls resolve to their code-mode result values.\n"); | ||
| section.push_str(&format!( | ||
| "- Import `{{ output_text, output_image, set_max_output_tokens_per_exec_call, set_yield_time, store, load }}` from `@openai/code_mode` (or `\"openai/code_mode\"`). `output_text(value)` surfaces text back to the model and stringifies non-string objects with `JSON.stringify(...)` when possible. `output_image(imageUrl)` appends an `input_image` content item for `http(s)` or `data:` URLs. `store(key, value)` persists JSON-serializable values across `{PUBLIC_TOOL_NAME}` calls in the current session, and `load(key)` returns a cloned stored value or `undefined`. `set_max_output_tokens_per_exec_call(value)` sets the token budget used to truncate direct `{PUBLIC_TOOL_NAME}` returns; `{WAIT_TOOL_NAME}` uses its own `max_tokens` argument instead and defaults to `10000`. `set_yield_time(value)` asks `{PUBLIC_TOOL_NAME}` to return early if the script is still running after that many milliseconds so `{WAIT_TOOL_NAME}` can resume it later. The returned content starts with a separate `Script completed`, `Script failed`, or `Script running with session ID …` text item that includes wall time. When truncation happens, the final text may include `Total output lines:` and the usual `…N tokens truncated…` marker.\n", | ||
| "- Import `{{ output_text, output_image, set_max_output_tokens_per_exec_call, set_yield_time, store, load, yield_control }}` from `@openai/code_mode` (or `\"openai/code_mode\"`). `output_text(value)` surfaces text back to the model and stringifies non-string objects with `JSON.stringify(...)` when possible. `output_image(imageUrl)` appends an `input_image` content item for `http(s)` or `data:` URLs. `store(key, value)` persists JSON-serializable values across `{PUBLIC_TOOL_NAME}` calls in the current session, and `load(key)` returns a cloned stored value or `undefined`. `set_max_output_tokens_per_exec_call(value)` sets the token budget used to truncate direct `{PUBLIC_TOOL_NAME}` returns; `{WAIT_TOOL_NAME}` uses its own `max_tokens` argument instead and defaults to `10000`. `set_yield_time(value)` asks `{PUBLIC_TOOL_NAME}` to return early if the script is still running after that many milliseconds so `{WAIT_TOOL_NAME}` can resume it later. `yield_control()` returns a yielded `{PUBLIC_TOOL_NAME}` response immediately while the script keeps running in the background. The returned content starts with a separate `Script completed`, `Script failed`, or `Script running with session ID …` text item that includes wall time. When truncation happens, the final text may include `Total output lines:` and the usual `…N tokens truncated…` marker.\n", |
Contributor
There was a problem hiding this comment.
I'm realizing that yield doesn't actually feel quite right here, its more about "detaching" focus or backgrounding. The yield is actually just fire and forget and has no impact on the control flow in repl. This is all the more apparent when we're allowing tool calls to run backgrounded too.
cconger
approved these changes
Mar 12, 2026
Contributor
cconger
left a comment
cconger
left a comment
There was a problem hiding this comment.
Naming nit, curious your thoughts.
cconger
reviewed
Mar 12, 2026
codex-rs/core/src/tools/code_mode.rs
Outdated
| tool_call.name, | ||
| tool_call.input, | ||
| ) | ||
| .await, |
Contributor
There was a problem hiding this comment.
We actually have head of line blocking here in this loop. Since our tool calls are async its possible that multiple could be in flight from js at a time, and here we are serially executing so a slow one blocks the rest.
Contributor
There was a problem hiding this comment.
Its even more exacerbated by this being shared across the process.
Collaborator
Author
There was a problem hiding this comment.
I agree but one step at a time. Want to keep individual changes small.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Testing