[fmt] conan v2 compatibility

[SSE4]

Specify library name and version: fmt/all

a port from czoido@5708ca4
/cc @czoido

• Fix for hooks (v2 failure): [conan-v2] hooks arguments changed in beta1 hooks#436
  • Requires 2.0.0beta2: Available folders in post_export hook conan#11643

[jgsogo]

IMHO we should keep a working test_xxxxx/conanfile.py using stable generators (cmake + cmake_find_package[_multi]). Would you mind keeping that example too?

[SSE4]

IMHO we should keep a working test_xxxxx/conanfile.py using stable generators (cmake + cmake_find_package[_multi]). Would you mind keeping that example too?

these generators aren't stable, they are marked as "experimental": https://docs.conan.io/en/latest/reference/generators/cmake_find_package_multi.html

I would keep only examples using current best practices going forward (CMakeDeps/CMakeToolchain).

UPD: last year I've tried to promote these generators conan-io/docs#2269, but seems like there was not enough interest to make them officially recommended. anyway, given now we have at least 7 different CMake generators available, it's easy for newcomers to get confused and completely lost. I think going forward, it only makes sense to promote CMakeDeps/CMakeToolchain as recommended, as others will just disappear soon.

[ghost]

I detected other pull requests that are modifying fmt/all recipe:

• fmt: Add version 9.0.0 #11554

This message is automatically generated by https://github.com/ericLemanissier/conan-center-conflicting-prs so don't hesitate to report issues/improvements there.

[jgsogo]

@SSE4 , please, fix conflicts... let's see if this one works with v2

[czoido]

@SSE4 the package_info I proposed was failing because the components in Conan 1.50 don't have default libdirs, includedirs, etc. and the hook 54 failed. It would be fixed with this updated package_info:

    def package_info(self):
        self.cpp_info.names["cmake_find_package"] = "fmt"
        self.cpp_info.names["cmake_find_package_multi"] = "fmt"
        self.cpp_info.names["pkg_config"] = "fmt"

        target = "fmt-header-only" if self.options.header_only else "fmt"
        self.cpp_info.set_property("cmake_target_name", "fmt::{}".format(target))

        # TODO: back to global scope in conan v2 once cmake_find_package* generators removed
        self.cpp_info.components["_fmt"].includedirs.extend(["include"])
        if self.options.header_only:
            self.cpp_info.components["_fmt"].defines.append("FMT_HEADER_ONLY=1")
            if self.options.with_fmt_alias:
                self.cpp_info.components["_fmt"].defines.append("FMT_STRING_ALIAS=1")
        else:
            postfix = "d" if self.settings.build_type == "Debug" else ""
            libname = "fmt" + postfix
            self.cpp_info.components["_fmt"].libs = [libname]
            # FIXME: remove when Conan 1.50 is used in c3i and update the Conan required version
            # from that version components don't have empty libdirs by default
            self.cpp_info.components["_fmt"].libdirs.extend(["lib"])
            if self.options.with_fmt_alias:
                self.cpp_info.components["_fmt"].defines.append("FMT_STRING_ALIAS=1")
            if self.options.shared:
                self.cpp_info.components["_fmt"].defines.append("FMT_SHARED")

        # TODO: to remove in conan v2 once cmake_find_package* generators removed
        self.cpp_info.components["_fmt"].names["cmake_find_package"] = target
        self.cpp_info.components["_fmt"].names["cmake_find_package_multi"] = target
        self.cpp_info.components["_fmt"].set_property("cmake_target_name", "fmt::{}".format(target))

This is more aligned with what we do in other recipes, but I would be ok with merging the current package_info as is but adding a FIXME or some comments saying that the reason of having a component there is just to create a new target name for header-only.

[SSE4]

@czoido seems like it still fails for Windows/shared configs. might it be due to missing default bindirs on components?

[czoido]

@czoido seems like it still fails for Windows/shared configs. might it be due to missing default bindirs on components?

Yes, I think that could be the reason....

[github-actions]

Hooks produced the following warnings for commit 48eecaf

fmt/5.3.0

post_package(): WARN: [MISSING SYSTEM LIBS (KB-H043)] Library './lib/libfmt.so' links to system library 'm' but it is not in cpp_info.system_libs.
post_package(): WARN: [MISSING SYSTEM LIBS (KB-H043)] Library './lib/libfmtd.so' links to system library 'm' but it is not in cpp_info.system_libs.

fmt/9.0.0

post_package(): WARN: [MISSING SYSTEM LIBS (KB-H043)] Library './lib/libfmtd.so' links to system library 'm' but it is not in cpp_info.system_libs.
post_package(): WARN: [MISSING SYSTEM LIBS (KB-H043)] Library './lib/libfmt.so' links to system library 'm' but it is not in cpp_info.system_libs.

[prince-chrismc]

Super excited seeing this working!!!

[prince-chrismc]

I am curious about the other variable mechanics https://docs.conan.io/en/latest/reference/conanfile/tools/cmake/cmaketoolchain.html#variables

If the toolchain sets the value before the options are declared then they do nothing according to https://cmake.org/cmake/help/latest/command/option.html

If this is set to use 1.49 then I think we could get away for using just tc.variables however I agree the cache ones are better UX this seems to be working for my project

[conan-center-bot]

All green in build 30 (32eced70afe823b2204860fd4f4ed25ad1d5a90f):

• fmt/5.3.0@:
  All packages built successfully! (All logs)

• fmt/7.1.3@:
  All packages built successfully! (All logs)

• fmt/9.0.0@:
  All packages built successfully! (All logs)

• fmt/8.0.1@:
  All packages built successfully! (All logs)

• fmt/6.2.1@:
  All packages built successfully! (All logs)

• fmt/8.1.1@:
  All packages built successfully! (All logs)

[jgsogo]

LGTM! 🎉

This is a good starting point for a working v2 recipe, some details can be polished later.