# Help with unexpected behavior when installing

**URL:** https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709
**Category:** Usage
**Created:** [July 7, 2021, 9:37pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709 "2021-07-07T21:37:02Z")
**Posts on this page:** 11
**Page:** 1

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 7, 2021, 9:37pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/1 "2021-07-07T21:37:02Z")

</div>

I have some really weird behavior happening. I have a CMake build that creates a dynamic library and a static one (which is used by Google Test). I instruct `install` to only install the dynamic library. So that works, but what is weird is that it compiles the static one as well (and correctly does not install it).

Here is the setup:

```auto
  add_library("native-build" SHARED ${ARG_BUILD_SOURCES} ${ARG_NATIVE_BUILD_SOURCES})
  target_link_libraries("native-build" PUBLIC z-re-sdk-lib)
  set_target_properties("native-build" PROPERTIES OUTPUT_NAME ${RE_ID})
  # ...

  add_library("native-test-lib" STATIC ${ARG_BUILD_SOURCES} ${ARG_NATIVE_BUILD_SOURCES})
  set_target_properties("native-test-lib" PROPERTIES OUTPUT_NAME ${RE_ID})
  # ...

  install(
      TARGETS "native-build"
      DESTINATION ${INSTALL_DIR}
  )

```

and here are some runs:

```auto
# cmake --build . --config Debug --target native-build
[50%] Built target z-re-sdk-lib
[100%] Built target native-build

# cmake --build . --config Debug --target native-test-lib
[100%] Built target native-test-lib

# cmake --build . --config Debug --target install
[30%] Built target z-re-sdk-lib
[61%] Built target native-test-lib
[100%] Built target native-build
Install the project...
-- Install configuration: "Debug"
-- Up-to-date: /Users/ypujante/Library/Application Support/Propellerhead Software/RackExtensions_Dev/TestInstal/TestInstal.dylib
...

```

Any idea what is going on? It seems that somehow `native-test-lib` gets added to the dependencies for install yet does not get installed (as can be seen from the output of install and I checked the installation folder and it is not present).

How can I troubleshoot this issue?

Thanks  
Yan

---

<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: [July 7, 2021, 10:06pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/2 "2021-07-07T22:06:31Z")

</div>

The `install` target depends on `all` by default. You can set [`CMAKE_SKIP_INSTALL_ALL_DEPENDENCY`](https://cmake.org/cmake/help/latest/variable/CMAKE_SKIP_INSTALL_ALL_DEPENDENCY.html) to skip this link or use the [`EXCLUDE_FROM_ALL`](https://cmake.org/cmake/help/latest/prop_tgt/EXCLUDE_FROM_ALL.html) property to remove `native-test-lib` from the `all` set.

---

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 7, 2021, 10:18pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/3 "2021-07-07T22:18:17Z")

</div>

I tried `CMAKE_SKIP_INSTALL_ALL_DEPENDENCY` but then it does not build anything at all so install fails. Using `EXCLUDE_FROM_ALL` on the dependencies that need to be excluded work.

I guess this is something that is really not intuitive and behaves backward from what you expect: you define a list of things to install… and they get installed properly but the rest still builds. If you remove the dependency on all (`CMAKE_SKIP_INSTALL_ALL_DEPENDENCY`), then nothing gets built even what is specified in `install()`.

Thanks for the help anyway.

---

<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: [July 7, 2021, 11:26pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/4 "2021-07-07T23:26:53Z")

</div>

The built-in targets have very course dependencies. There’s [an issue](https://gitlab.kitware.com/cmake/cmake/issues/8438) to fix that, but it hasn’t seen work. Component installations also make the graph more complicated. The first problem that comes to my mind is knowing which commands make what is needed for `install(FILES)` (or `install(DIRECTORY)`) when the paths are side effects of other commands. Long ago, CMake made the decision to just say "`install` will first build the project" and now has some release valves on that, but there hasn’t been any concerted effort to meaningfully improve the situation.

---

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 8, 2021, 12:21pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/5 "2021-07-08T12:21:53Z")

</div>

Thanks for the detailed explanation. I think without doing any work to change how CMake behaves in regards to `install`, it would be useful to change the documentation to make it more clear, as the behavior is counter intuitive. For example, if you look at [https://cmake.org/cmake/help/latest/command/install.html](https://cmake.org/cmake/help/latest/command/install.html), there isn’t even a mention about `CMAKE_SKIP_INSTALL_ALL_DEPENDENCY` and `EXCLUDE_FROM_ALL` is always discussed in regards to the `install` calls, not what happens if you set it as a property of a target.

---

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 11, 2021, 2:39pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/6 "2021-07-11T14:39:56Z")

</div>

After simplifying my build and using `FetchContent_MakeAvailable(googletest)` as suggested in the GoogleTest documentation, to my horror, running the install target for my project, also ended up installing GoogleTest library (includes and libs) locally (under /usr/local)!!!

This is so NOT the expected behavior.

I can see that the generated `cmake_install.cmake` now contains:

```auto
if(NOT CMAKE_INSTALL_LOCAL_ONLY)
  # Include the install script for each subdirectory.
  include("/Volumes/Vault/Downloads/com.acme.Kooza2-plugin/build/googletest-build/cmake_install.cmake")

endif()

```

How do I tell CMake to not even include this section in the `cmake_install.cmake`?

---

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 11, 2021, 3:53pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/7 "2021-07-11T15:53:51Z")

</div>

I find out you can set the option `INSTALL_GTEST` to `OFF` to disable installing the tests, but then I am back with the same issue I was having originally which is that all GoogleTest targets ends up being part of the `All` target and so they get built on install…

I suppose there is no `FetchContent_MakeAvailable(googletest EXCLUDE_FROM_ALL)` version of the call so I am reverting to including the directory myself since this way offers the option to exclude it:

```auto
add_subdirectory(${googletest_SOURCE_DIR} ${googletest_BINARY_DIR} EXCLUDE_FROM_ALL)

```

---

<div class="post-metadata">

### Author: ![craig.scott](https://discourse.cmake.org/user_avatar/discourse.cmake.org/craig.scott/32/20_2.png) [@craig.scott](https://discourse.cmake.org/u/craig.scott)
#### Post date: [July 11, 2021, 10:54pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/8 "2021-07-11T22:54:23Z")

</div>

You have at least a couple of choices. There is an `EXCLUDE_FROM_ALL` directory property that you can set before calling `FetchContent_MakeAvailable()`. It will affect all directories you pull in from the current directory scope, but if you’re adding all your dependencies from a dedicated directory (e.g. `<top-of-source-tree>/dependencies`), this should be workable. Do not use `EXCLUDE_FROM_ALL` in the top level directory due to limitations discussed in [issue 20168](https://gitlab.kitware.com/cmake/cmake/-/issues/20168). [Issue 20167](https://gitlab.kitware.com/cmake/cmake/-/issues/20167) is also relevant.

The method I prefer though is to solve this issue using install components. Change the default install component just before you call `FetchContent_MakeAvailable(googletest)` using `CMAKE_INSTALL_DEFAULT_COMPONENT_NAME`. Then, simply don’t include that component in your packages.

You could combine the two methods I guess. Personally, I never felt the building of googletest targets was a significant contributor to the overall build time of our projects, but it will obviously depend on your own project.

---

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 12, 2021, 12:54pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/9 "2021-07-12T12:54:32Z")

</div>

It is not so much about the build time. My projects are audio plugins, so they need to be installed to be manually tested (copying the macOS bundle or windows dll to a specific location so that the DAW can load them). Sometimes I modify code in a way that breaks the compilation of the tests and I just want to manually test because I am not sure that the changes are worth it before I go and fix the tests. Having the tests unnecessarily compiled as part of the install phase, breaks this workflow because now install cannot happen…

The solution I mentioned works fine (`add_subdirectory ... EXCLUDE_FROM_ALL`). It is just very unfortunate that the default behavior of CMake combined with the example provided on the GoogleTest CMake [documentation](https://google.github.io/googletest/quickstart-cmake.html) page leads to having GoogleTest installed when that should pretty much never be the case…

---

<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: [July 13, 2021, 2:12pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/10 "2021-07-13T14:12:57Z")

</div>

> [@fry](#):
>
> the default behavior of CMake combined with the example provided on the GoogleTest CMake [documentation](https://google.github.io/googletest/quickstart-cmake.html) page leads to having GoogleTest installed when that should pretty much never be the case…

I would suggest contacting the GoogleTest maintainers to improve the documentation there to mention this behavior.

---

<div class="post-metadata">

### Author: ![fry](https://discourse.cmake.org/user_avatar/discourse.cmake.org/fry/32/613_2.png) [@fry](https://discourse.cmake.org/u/fry)
#### Post date: [July 13, 2021, 2:14pm UTC](https://discourse.cmake.org/t/help-with-unexpected-behavior-when-installing/3709/11 "2021-07-13T14:14:14Z")

</div>

Already done [Fix CMake documentation to prevent GoogleTest install · Issue #3486 · google/googletest · GitHub](https://github.com/google/googletest/issues/3486) 😉