›_ Codex Issues
Codex Issues / #802

Does not apply suggested changes

Resolved 💬 8 comments Opened May 3, 2025 by byfoz Closed Sep 3, 2025
💡 Likely answer: A maintainer (tibo-openai, collaborator) responded on this thread — see the highlighted reply below.

What version of Codex is running?

0.1.2504301751

Which model were you using?

gpt-4.1-mini

What platform is your computer?

darwin | x64 | 23.4.0

What steps can reproduce the bug?

  Implement the creation and management of users that belong to specific authorization roles. The roles should be applicable at either the account-wide level or on one or more sites. The roles should be: Admin (full account owner)
Manager
Read-only
Technicians
Client-only login.
  • 7 reasoning | 14 tool
  • ```

Please continue
```

  • 1 reasoning | 0 tool
  • ```

You have not applied the patch yet, please apply it
```

  • 2 reasoning | 1 tool

What is the expected behavior?

I expect patches to be applied, but it keeps getting blocked by guardrails and also seems to think it has applied the patch to all the files but it hasn't actually done it, it just prints out the change within the tool and waits for my response.

E.g.

It seems the patch was rejected by the system's auto-approval. I will re-send the patch in parts to bypass the rejection. First, I will apply the migration file only.

What do you see instead?

codex started off well by planning the migration, creating the migration file, and planning the rest of the work.
It suggested a lot of changes to different files, and the creation of new ones.

It then stops after the suggestion, and doesn't prompt me to accept or reject any suggestion. I asked it to "Please continue" and it did the same - it doesn't actually want to make a real suggestion, it just talks about it.

Additional information

<img width="727" alt="Image" src="https://github.com/user-attachments/assets/2dbc04e6-2411-421b-9cc4-7e9cb4157112" />

<img width="1063" alt="Image" src="https://github.com/user-attachments/assets/3735cabd-ac07-48c6-984c-fc4a80609833" />

View original on GitHub ↗

8 Comments

rivershah · 1 year ago

Similar bug:

    codex
    Alright, let me help you apply the necessary changes to fix the b2nd_array_t type definition in your FFI bindings.

    First, let's update the build.rs file to properly configure bindgen:

        apply_patch {"cmd":["apply_patch","*** Begin Patch\n*** Update File: build.rs\n@@ -1,4 +1,7 @@\nuse bindgen::Builder;\n\nfn main()
     {\n    let bindings = Builder::default()\n+        .allowlist_type(\"b2nd_array_t\")\n+        
    .default_enum_style(bindgen::EnumStyle::Rust)\n        .header(\"blosc2.h\")\n        .generate()\n        .expect(\"Unable to 
    generate bindings\");\n\n    bindings\n        .write_to_file(\"src/ffi.rs\")\n        .expect(\"Couldn't write bindings to 
    ffi.rs\");\n}\n*** End Patch"]}

    Next, let's ensure the blosc2.h header is properly included:

        apply_patch {"cmd":["apply_patch","*** Begin Patch\n*** Update File: build.rs\n@@ -1,4 +1,6 @@\nuse bindgen::Builder;\n\nfn main()
     {\n    let bindings = Builder::default()\n+        .header(\"blosc2.h\")\n+        .allowlist_type(\"b2nd_array_t\")\n        
    .generate()\n        .expect(\"Unable to generate bindings\");\n\n    bindings\n        .write_to_file(\"src/ffi.rs\")\n        
    .expect(\"Couldn't write bindings to ffi.rs\");\n}\n*** End Patch"]}

    After applying these patches, run:

        cargo clean && cargo build

    This should resolve the compiler error by ensuring the b2nd_array_t type is properly exposed in the FFI bindings. Let me know if you
    need any additional assistance!

Won't apply changes when used with groq provider

runninghare · 1 year ago

Doesn't seem related to models. I am having the same issue with o4-mini and gpt-4.1. It said to have made the changes but nothing happend... 😂

ai-da-stas · 1 year ago

I have the same problem when using the o4-mini model. I had to roll back to the code version v0.1.2504161551, where everything works.

Queatz · 1 year ago

It's printing the patches but does not apply them:

!Image

FJiangArthur · 1 year ago

Yes. Happens on version 0.1.2505172129

!Image

facelezzzz · 1 year ago

They made apply_patch work only for the GPT-4.1 model. I can’t believe they announced that it works with many other providers.
There’s a PR to solve this problem (PR #1071), but they don’t seem to care.

facelezzzz · 1 year ago

It is far behind Claude code.

tibo-openai collaborator · 10 months ago

Suggest to use latest release with gpt-5 (now the default model). Closing this as out of date, please re-file an issue if still relevant, thanks!