# install(EXPORT) does not properly take into account of paths containing '..'
**URL:** https://discourse.cmake.org/t/install-export-does-not-properly-take-into-account-of-paths-containing/14408
**Category:** Code
**Tags:** os:linux
**Created:** [July 22, 2025, 12:50am UTC](https://discourse.cmake.org/t/install-export-does-not-properly-take-into-account-of-paths-containing/14408 "2025-07-22T00:50:11Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@WildCard65](https://discourse.cmake.org/u/WildCard65)
#### Post date: [July 22, 2025, 12:50am UTC](https://discourse.cmake.org/t/install-export-does-not-properly-take-into-account-of-paths-containing/14408/1 "2025-07-22T00:50:11Z")
The command install(EXPORT) in its current state is vastly incompatible with making packages with non-standard install directories (for example /lib instead of /usr/lib) resulting in extra `get_filename_component(_IMPORT_PREFIX "${_IMPORT_PREFIX}" PATH)` being generated for every ‘..’ used instead of correctly calculating the appropriate path.
For example: If its installed to /lib/cmake//.cmake and the install prefix is /usr, it will generate an extra `get_filename_component(_IMPORT_PREFIX "${_IMPORT_PREFIX}" PATH)` for the ‘..’ inside of CMAKE\_INSTALL\_LIBDIR.
Why does this matter to me? Well I’m trying my hand of making my own Linux OS that uses a seperated /lib and /usr/lib instead of the standard symlink of /lib → /usr/lib.
---
### Author:  [@ClausKlein](https://discourse.cmake.org/u/ClausKlein)
#### Post date: [July 22, 2025, 5:44am UTC](https://discourse.cmake.org/t/install-export-does-not-properly-take-into-account-of-paths-containing/14408/2 "2025-07-22T05:44:38Z")
I can’t understand what you try to do?
You may always install to any prefix independent of the default settings:
```bash
bash-5.2$ cmake --install . --prefix /tmp
-- Install configuration: "RelWithDebInfo"
-- Installing: /tmp/lib/libbeman.exemplar.a
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: file: /tmp/lib/libbeman.exemplar.a(identity.cpp.o) has no symbols
warning: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: archive library: /tmp/lib/libbeman.exemplar.a the table of contents is empty (no object file members in the library define global symbols)
-- Installing: /tmp/include/beman/exemplar/identity.hpp
-- Installing: /tmp/lib/cmake/beman.exemplar/beman.exemplar-config.cmake
-- Installing: /tmp/lib/cmake/beman.exemplar/beman.exemplar-config-version.cmake
-- Installing: /tmp/lib/cmake/beman.exemplar/beman.exemplar-targets.cmake
-- Installing: /tmp/lib/cmake/beman.exemplar/beman.exemplar-targets-relwithdebinfo.cmake
bash-5.2$ cmake --install .
-- Install configuration: "RelWithDebInfo"
-- Installing: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/libbeman.exemplar.a
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: file: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/libbeman.exemplar.a(identity.cpp.o) has no symbols
warning: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: archive library: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/libbeman.exemplar.a the table of contents is empty (no object file members in the library define global symbols)
-- Up-to-date: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/include/beman/exemplar/identity.hpp
-- Up-to-date: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/cmake/beman.exemplar/beman.exemplar-config.cmake
-- Up-to-date: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/cmake/beman.exemplar/beman.exemplar-config-version.cmake
-- Up-to-date: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/cmake/beman.exemplar/beman.exemplar-targets.cmake
-- Up-to-date: /Users/clausklein/Workspace/cpp/beman-project/exemplar/stagedir/lib/cmake/beman.exemplar/beman.exemplar-targets-relwithdebinfo.cmake
bash-5.2$
```
---
### Author:  [@WildCard65](https://discourse.cmake.org/u/WildCard65)
#### Post date: [July 22, 2025, 12:59pm UTC](https://discourse.cmake.org/t/install-export-does-not-properly-take-into-account-of-paths-containing/14408/3 "2025-07-22T12:59:50Z")
GNUInstallDirs normalizes the variable CMAKE\_INSTALL\_LIBDIR to be under /usr/, I have to force it to go to / by using ../ which results in an extra get\_filename\_componet() call for \_IMPORT\_PREFIX resulting in the final import prefix being outside of where “root” is.
This also results in /usr being dropped from the final auto determined import prefix as well.
---
### Author:  [@vito.gamberini](https://discourse.cmake.org/u/vito.gamberini)
#### Post date: [July 22, 2025, 1:18pm UTC](https://discourse.cmake.org/t/install-export-does-not-properly-take-into-account-of-paths-containing/14408/4 "2025-07-22T13:18:10Z")
> [@WildCard65](#):
>
> GNUInstallDirs normalizes the variable CMAKE\_INSTALL\_LIBDIR to be under /usr/
This is the real bug, and the documentation added in CMake 3.4 to justify it is wrong:
```log
/
For other than the SYSCONFDIR, LOCALSTATEDIR and
RUNSTATEDIR, the value of CMAKE_INSTALL_ is prefixed
with usr/ if it is not user-specified as an absolute path.
```
It justifies this with:
```log
This is required by the GNU Coding Standards, which state:
When building the complete GNU system, the prefix will be empty
and /usr will be a symbolic link to /.
```
But that’s just, not what that sentence says at all? It’s saying if the prefix is _empty_ (not `"/"`), then everything should be installed to `/`, and anything that is installed to `/usr` should instead end up in `/`.
I would open up an issue. CMake should absolutely support the old-school usage of root as the “essentials” install tree.