# CMake -Werror flag removal logic may break ABI detection

**URL:** https://discourse.cmake.org/t/cmake-werror-flag-removal-logic-may-break-abi-detection/8230
**Category:** Usage
**Created:** [May 29, 2023, 8:46am UTC](https://discourse.cmake.org/t/cmake-werror-flag-removal-logic-may-break-abi-detection/8230 "2023-05-29T08:46:24Z")
**Posts on this page:** 3
**Page:** 1

<div class="post-metadata">

### Author: ![nowitshistory](https://discourse.cmake.org/letter_avatar_proxy/v4/letter/n/bc8723/32.png) [@nowitshistory](https://discourse.cmake.org/u/nowitshistory)
#### Post date: [May 29, 2023, 8:46am UTC](https://discourse.cmake.org/t/cmake-werror-flag-removal-logic-may-break-abi-detection/8230/1 "2023-05-29T08:46:25Z")

</div>

CMake compiles a temporary C program to detect values for `CMAKE_SIZEOF_VOID_P` and other variables.

I have a toolchain file that defined the linker flag `-pie` for executable targets and for the temporary program to compile and set `CMAKE_SIZEOF_VOID_P` correctly it requires the compiler flags `-fPIC` or `-fPIE`.  
But if I put `-fPIC` immediately after `-Werror` in `CMAKE_C_FLAGS` in the toolchain file then it would have no effect on the temporary program. It would simply get removed along with `-Werror` thus causing the temporary program fails to compile and `CMAKE_SIZEOF_VOID_P` left unset.

The minimum reproducible example:

```cmake
# CMakeLists.txt
cmake_minimum_required(VERSION 3.14.1)
project(test)
message("CMAKE_SIZEOF_VOID_P: ${CMAKE_SIZEOF_VOID_P}")

```

```cmake
# toolchain.cmake
set(CMAKE_C_FLAGS "-Werror -fPIC")
set(CMAKE_EXE_LINKER_FLAGS "-pie")

```

By executing `cmake -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake .` with CMake 3.20.5 I get the following output which shows that `CMAKE_SIZEOF_VOID_P` is unset:

```auto
CMAKE_SIZEOF_VOID_P:

```

I expect the following output on my 64 bit machine as a lower version of CMake (such as CMake 3.16.5) would correctly output:

```auto
CMAKE_SIZEOF_VOID_P: 8

```

I believe the problem is introduced by the following change to “Handle NVCC-style -Werror flags” which was first released in CMake 3.19 or so. While handling the “NVCC-style flags” it also erroneously removes other unrelated flags which can cause problems.

[https://gitlab.kitware.com/cmake/cmake/-/commit/079ea66468a6ffe0b02c3d6622bc0230fdf455b0](https://gitlab.kitware.com/cmake/cmake/-/commit/079ea66468a6ffe0b02c3d6622bc0230fdf455b0)

I fixed it by setting `-Werror` as the last flag of `CMAKE_C_FLAGS` so no flags will be removed along with it but it is rather hackish. I hope that CMake could do a better job at handling `-Werror` removal possibly by modifying the regular expression to treat flags immediately after `-Werror` that start with a `-` differently so that they aren’t mistaken as NVCC Werror arguments and therefore not removed incorrectly.

---

<div class="post-metadata">

### Author: ![ben.boeckel](https://discourse.cmake.org/letter_avatar_proxy/v4/letter/b/ea5d25/32.png) [@ben.boeckel](https://discourse.cmake.org/u/ben.boeckel)
#### Post date: [May 29, 2023, 2:02pm UTC](https://discourse.cmake.org/t/cmake-werror-flag-removal-logic-may-break-abi-detection/8230/2 "2023-05-29T14:02:46Z")

</div>

This indeed sounds like a regression. Could you please [file an issue](https://gitlab.kitware.com/cmake/cmake/-/issues)?

Cc: @brad.king @tambre

---

<div class="post-metadata">

### Author: ![tambre](https://discourse.cmake.org/user_avatar/discourse.cmake.org/tambre/32/5127_2.png) [@tambre](https://discourse.cmake.org/u/tambre)
#### Post date: [May 29, 2023, 2:29pm UTC](https://discourse.cmake.org/t/cmake-werror-flag-removal-logic-may-break-abi-detection/8230/3 "2023-05-29T14:29:23Z")

</div>

Oops!  
Surprising this went unnoticed for so long.

Fix:  
[https://gitlab.kitware.com/cmake/cmake/-/merge\_requests/8518](https://gitlab.kitware.com/cmake/cmake/-/merge_requests/8518)