[bug] PkgConfigDeps: -F <frameworkdir> on macOS leads to "clang: error: no such file or directory:" at link time #11867
Comments
|
Hi @SpaceIm Thanks for reporting the issue! In a nutshell, the explanation is as follows:
@property
def framework_paths(self):
if self._framework_paths is None:
self._framework_paths = self._filter_paths(self.frameworkdirs)
return self._framework_paths
Given that, the remaining question is, why is So, I think that the workaround could come in two ways:
def package_info(self):
self.cpp_info.frameworkdirs = []
I was wondering if it could be possible to declare in Conan 1.x |
It means updating 1300+ recipes in conan-center :/
zlib recipe has a layout: |
We can start with the recipes causing failures, I bet there won't be 1300. The recipes were always declaring by default |
All recipes potentially cause failures for consumers, since it affects a generator. For the specific case of glib recipe migration to conan v2, indeed we have to fix few recipes only (at least pcre2, pcre & zlib). But hardcoding I still don't understand: Do we need to hardcode |
I don't understand what you mean. For me it is not strange at all because Conan 1.X always had a default to
It is needed because the consumers using v2 generators won't skip the non-existing folders. |
When you say "In Conan v2 (OR recipes using the layout())", does it mean that in conan v1 mode, a recipe with a layout like cmake_layout shouldn't populate frameworksdirs by default? Because latest RREV of zlib has a cmake_layout, but still, its frameworksdirs is not empty. |
|
Not True. With Conan 1.51.3, zlib/1.2.12 with layout, using set(zlib_FRAMEWORK_DIRS_RELEASE )Without layout(): set(zlib_FRAMEWORK_DIRS_RELEASE "${zlib_PACKAGE_FOLDER_RELEASE}/Frameworks") |
|
Indeed, I've checked in the latest log of conan-io/conan-center-index#12221, and I don't see zlib frameworks dirs injection anymore (I see bzip2 & pcre2 Frameworks dir. bzip2 recipe has been updated 2 days ago, so it should be fine now. Remains pcre2 recipe to update). |
|
@franramirez688 I think we talked about it before. I have experienced the same problem already in conan-io/conan-center-index#11139 |
|
@SSE4 Maybe a CCI hook should be implemented to enforce |
|
yes, I think it can be enforced. I believe we already have layouts for most of the build helpers (AutTools, CMake, MSBuild and Meson). for others, hand-crafted layout could be used. |
|
Note: In case you consider it. To activate the "2.0" mode an empty |
for the record (probably, irrelevant for this issue), this behavior might be consuming for newcomers: having an empty method doing essentially nothing, but has some important side-effects. reference also currently says nothing about that "side-effect" :https://docs.conan.io/en/latest/reference/conanfile/methods.html#layout |
|
I don't think the bug here is what it seems it is. Simply providing non-existing directory for "-F" won't cause an error. The bug here is that "-F" flag and the proceeding directory are treated as different. This should be The first line will only cause a warning, but the third line will cause the error, because it should be preceeded by "-F" flag. This happens because autotools (in my case with imagemagick), pkgconfig and apparently meson reorder the flags and remove duplicates. |
|
In my case, doing a workaround patch like this, which just removed the space in between fixed the issue: if self.settings.os == "Macos":
# FIXME workaround bug with pkgconfig files and apple frameworks
# see https://github.com/conan-io/conan/pull/12307
for pc in glob.glob(os.path.join(self.generators_folder, "*.pc")):
files.replace_in_file(self, pc, "-F ", "-F", strict=False)
files.replace_in_file(self,
os.path.join(self.generators_folder,
"conanautotoolsdeps.sh"),
"-F ", "-F", strict=False) |
Environment Details (include every applicable attribute)
Steps to reproduce (Include if Applicable)
conan create . glib/2.73.3@ -o "*":shared=True -o libelf:shared=Falseglib build system is Meson, but it can be reproduced as well with a simple CMakeLists relying on pkg_check_modules() and a shared dependency (not tried with static).
If I tweak this PR by replacing PkgConfigDeps by pkg_config generator (also need to move these .pc files in generator folder expected by layout), the build succeeds. It fails afterwards in test_package since it also relies on PkgConfigDeps (with CMake build system).
Logs (Executed commands with output) (Include/Attach if Applicable)
The text was updated successfully, but these errors were encountered: