[question] Mimicking CMake Find Modules

[Morwenn]

When creating a package that already exists in standard CMake Find Modules it is advised to have a cpp_info.name corresponding to the name expected by the equivalent call to find_package to make sure that the cmake_find_package* generators work smoothly.

However some of these standard CMake recipes provide more CMake variables than those provided by the Conan generators, and libraries using them through find_package tend to make use of these additional variables. The example I had in mind was HDF5 for which I added a recipe to conan-center-index a few weeks ago: it provides more CMake variables that the bare minimum such as HDF5_HL_LIBRARIES or HDF5_IS_PARALLEL.

When adding recipes to conan-center-index for such packages that already standard in CMake, should we also provide those additional variables through cpp_info.build_modules to make sure that the cmake_* generators work with already existing projects in the wild to the best of our ability?

[madebr]

Other packages have the same problem:

• openssl: missing OpenSSL::Crypto and OpenSSL::SSL targets, missing OPENSSL_CRYPTO_LIBRARY, OPENSSL_CRYPTO_LIBRARIES, OPENSSL_SSL_LIBRARY and OPENSSL_SSL_LIBRARIES variable
• libxml2: missing LIBXML2_XMLLINT_EXECUTABLE and LIBXML2_VERSION_STRING variable (if the executable is shipped)
• zlib: missing *_VERSION_* variables

[madebr]

I've created a PR for adding the variables for libpng: #434

[jgsogo]

As we were discussing here, sometimes it would be better not to provide all the CMake variables if they are not consistent, it can be better to patch the build scripts of the consumers (and work to improve the C++ ecosystem).

Other variables could go in the cpp_info.defines and others will be available with the components feature (getting delayed more than expected). IMO the build_modules should only be used as a last resort (or for famous libraries where it is not possible to patch/educate all the consumers).

[Morwenn]

I wish education would be enough, but on the other hand I have to redefine the exact same variable in the CMake wrapper whenever I package a new library that uses HDF5 through CMake and can't just rely on cmake_find_package :/

[madebr]

Imho, the FindXXX.cmake documentation at cmake.org should be considered as an API.
Any deviation of it must be considered a bug.

The recipes created by this repo are targeted to the complete conan community, not only us (=packagers).
If some conan find_package script behaves different then the cmake one,
users would consider conan as non-functional or broken.
The conan find_package script should just work interchangeably with the cmake one.

[gocarlos]

Any deviation of it must be considered a bug.

agree, currently we build with conan for development and with yocto for production (regulated industries...)
that we cannot interchangeable use the standard find_package() functions makes a lot of issues...

actually its not only the imported targets ala OpenSSL::SSL vs. OpenSSL::OpenSSL but also result variables e.g. https://cmake.org/cmake/help/v3.9/module/FindGTest.html#result-variables defines:

GTEST_LIBRARIES, conan defines GTest_LIBRARIES

those smaller bugs make the adoptions in some big coorporations pretty hard/slow

[madebr]

I've opened a RFC at conan-io/conan#6587
Please comment

[kyllingstad]

Any deviation of it must be considered a bug.

agree, currently we build with conan for development and with yocto for production (regulated industries...)
that we cannot interchangeable use the standard find_package() functions makes a lot of issues...

I agree, and would like to add that using Conan for development but not for production is going to be a very common use case. For example, I don't foresee that any of the big Linux distros are going to adopt Conan for dependency management in the near future, so anyone who wants their software to be distributed with Debian, Red Hat, etc. are going to be in this situation.

Any deviation from the CMake-bundled FindXxx.cmake modules or upstream XxxConfig.cmake files is effectively locking developers into the Conan ecosystem or forcing them to jump through hoops to break out of it.

[jgsogo]

Linking here this issue (conan-io/conan#7254), we will be adding to 1.28 a way to define the name of the FindXXXX.cmake file which should provide the last bit needed to mimic existing CMake modules. It will, for example, allow the protobuf package to provide targets like protobuf:: from a FindProtobuf.cmake file.

[jgsogo]

And finally this one: conan-io/conan#7261, it should be possible to make any adjustments needed.

Anyone is missing something?

[ericLemanissier]

qt (since 5.15.0) creates targets both in Qt5 and Qt namespace : https://doc.qt.io/qt-5/cmake-get-started.html#imported-library-targets. I guess this is not possible to mimic this yet ?

[franramirez688]

This is already solved by the generator CMakeDeps.