Repository navigation
Studio: keep lone asterisks and backticks in finished chat replies - #14391
NilayYadav wants to merge 6 commits into
Conversation
[skip ci]
|
@codex review |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Codex Review: Didn't find any major issues. Keep it up! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
@codex review |
|
Codex Review: Didn't find any major issues. Another round soon, please! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
|
@codex review |
|
Codex Review: Didn't find any major issues. 👍 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Review + reproduction evidencePinned review: I reproduced a normal chat with a connected OpenAI-compatible model on the merge base
Conclusion: real defect on main, fixed; waiting only on the PR's CI run for |
|
That run merged into |
## Problem `studio/backend/tests/test_text_io_encoding.py::test_text_io_names_its_encoding` fails on main, so the `(Python 3.13, rest)` backend shard is red on every PR built on top of #14374 (for example #14391): ``` AssertionError: Text I/O without an explicit encoding falls back to locale.getencoding(), ... hardware.py:165: subprocess(text = True) without encoding ``` #14374 added `_running_under_rosetta()`, which runs `sysctl -n hw.optional.arm64` with `text = True` and no encoding. The guard is right. Every other `subprocess` call in `hardware.py` names its encoding, and a bare `text = True` decodes with the locale codec (ASCII under a C or POSIX locale). ## Fix Pass `encoding = "utf-8", errors = "replace"`, as the other probes in this file do. The probe only reads `0` or `1`, so behaviour is unchanged. ## Verification - `studio/backend/tests/test_text_io_encoding.py`: fails on main (the line above); with this change, 857 passed. - `test_chat_only_reason.py`, `test_export_capability.py` and `test_video_capability.py`, which exercise the Rosetta check: 64 passed.
|
@codex review |
1 similar comment
|
@codex review |
|
Codex Review: Didn't find any major issues. 🎉 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |

Once a chat reply finished, a lone asterisk, double asterisk or backtick in ordinary text was deleted and the rest of the sentence was restyled: "Add .pyc to .gitignore." showed as "Add .pyc to .gitignore." in italics, and "Press the ` key" turned the rest of the line into code. When the reply ended with a line break, a stray "" or "**" was added at the end instead: "Use **kwargs to forward them." showed as "Use **kwargs to forward them. **". It stayed that way after reopening the chat, while Copy still gave the original text.
The repair that closes half-written Markdown now runs only on a reply that was cut short (Stop, the Max Tokens limit, or an error), right away and after a reload. A reply that finished normally is shown exactly as written.
Known limits, unchanged from before: a cut reply can still restyle a lone marker earlier in it, and thinking text over about 16,000 characters is still repaired piece by piece.