Reland Print library module manifest path again #84881
Conversation
…2160) This implements a way for the compiler to find the modules.json associated with the C++23 Standard library modules. This is based on a discussion in SG15. At the moment no Standard library installs this manifest. llvm#75741 adds this feature in libc++. This reverts commit 82f424f. Disables the tests on non-X86 platforms as suggested.
llvmbot
added
clang
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-clang-modules Author: Chuanqi Xu (ChuanqiXu9) ChangesFollowing of #82160 The reason why the above PR fails is that the So for patch it self, I hacked it by using resource dir in the test since the resource dir has the higher priority, which is not strongly correct since we won't do that in practice. @kaz7 would you like to test on your environment to avoid this get reverted again? On the libc++ side, it shows that it lacks a Also if we don't like the hacked test in the current patch, we must wait for libc++ to fix this to proceed. But I feel this is somewhat odd since the test of clang shouldn't dependent on libc++. CC: @mordante Full diff: https://github.com/llvm/llvm-project/pull/84881.diff 4 Files Affected:
diff --git a/clang/include/clang/Driver/Driver.h b/clang/include/clang/Driver/Driver.h
index c4cab360bab3bb..bcf8c1295f2ddf 100644
--- a/clang/include/clang/Driver/Driver.h
+++ b/clang/include/clang/Driver/Driver.h
@@ -615,6 +615,16 @@ class Driver {
// FIXME: This should be in CompilationInfo.
std::string GetProgramPath(StringRef Name, const ToolChain &TC) const;
+ /// Lookup the path to the Standard library module manifest.
+ ///
+ /// \param C - The compilation.
+ /// \param TC - The tool chain for additional information on
+ /// directories to search.
+ //
+ // FIXME: This should be in CompilationInfo.
+ std::string GetStdModuleManifestPath(const Compilation &C,
+ const ToolChain &TC) const;
+
/// HandleAutocompletions - Handle --autocomplete by searching and printing
/// possible flags, descriptions, and its arguments.
void HandleAutocompletions(StringRef PassedFlags) const;
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index aca8c9b0d5487a..6e7f142789c994 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5374,6 +5374,9 @@ def print_resource_dir : Flag<["-", "--"], "print-resource-dir">,
def print_search_dirs : Flag<["-", "--"], "print-search-dirs">,
HelpText<"Print the paths used for finding libraries and programs">,
Visibility<[ClangOption, CLOption]>;
+def print_std_module_manifest_path : Flag<["-", "--"], "print-library-module-manifest-path">,
+ HelpText<"Print the path for the C++ Standard library module manifest">,
+ Visibility<[ClangOption, CLOption]>;
def print_targets : Flag<["-", "--"], "print-targets">,
HelpText<"Print the registered targets">,
Visibility<[ClangOption, CLOption]>;
diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index 190782a79a2456..eae27769b73ec0 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -2196,6 +2196,12 @@ bool Driver::HandleImmediateArgs(const Compilation &C) {
return false;
}
+ if (C.getArgs().hasArg(options::OPT_print_std_module_manifest_path)) {
+ llvm::outs() << GetStdModuleManifestPath(C, C.getDefaultToolChain())
+ << '\n';
+ return false;
+ }
+
if (C.getArgs().hasArg(options::OPT_print_runtime_dir)) {
if (std::optional<std::string> RuntimePath = TC.getRuntimePath())
llvm::outs() << *RuntimePath << '\n';
@@ -6185,6 +6191,44 @@ std::string Driver::GetProgramPath(StringRef Name, const ToolChain &TC) const {
return std::string(Name);
}
+std::string Driver::GetStdModuleManifestPath(const Compilation &C,
+ const ToolChain &TC) const {
+ std::string error = "<NOT PRESENT>";
+
+ switch (TC.GetCXXStdlibType(C.getArgs())) {
+ case ToolChain::CST_Libcxx: {
+ std::string lib = GetFilePath("libc++.so", TC);
+
+ // Note when there are multiple flavours of libc++ the module json needs to
+ // look at the command-line arguments for the proper json.
+ // These flavours do not exist at the moment, but there are plans to
+ // provide a variant that is built with sanitizer instrumentation enabled.
+
+ // For example
+ // StringRef modules = [&] {
+ // const SanitizerArgs &Sanitize = TC.getSanitizerArgs(C.getArgs());
+ // if (Sanitize.needsAsanRt())
+ // return "modules-asan.json";
+ // return "modules.json";
+ // }();
+
+ SmallString<128> path(lib.begin(), lib.end());
+ llvm::sys::path::remove_filename(path);
+ llvm::sys::path::append(path, "modules.json");
+ if (TC.getVFS().exists(path))
+ return static_cast<std::string>(path);
+
+ return error;
+ }
+
+ case ToolChain::CST_Libstdcxx:
+ // libstdc++ does not provide Standard library modules yet.
+ return error;
+ }
+
+ return error;
+}
+
std::string Driver::GetTemporaryPath(StringRef Prefix, StringRef Suffix) const {
SmallString<128> Path;
std::error_code EC = llvm::sys::fs::createTemporaryFile(Prefix, Suffix, Path);
diff --git a/clang/test/Driver/modules-print-library-module-manifest-path.cpp b/clang/test/Driver/modules-print-library-module-manifest-path.cpp
new file mode 100644
index 00000000000000..24797002b80f53
--- /dev/null
+++ b/clang/test/Driver/modules-print-library-module-manifest-path.cpp
@@ -0,0 +1,36 @@
+// Test that -print-library-module-manifest-path finds the correct file.
+
+// RUN: rm -rf %t && split-file %s %t && cd %t
+// RUN: mkdir -p %t/Inputs/usr/lib/x86_64-linux-gnu
+// RUN: touch %t/Inputs/usr/lib/x86_64-linux-gnu/libc++.so
+
+// RUN: %clang -print-library-module-manifest-path \
+// RUN: -stdlib=libc++ \
+// RUN: -resource-dir=%t/Inputs/usr/lib/x86_64-linux-gnu \
+// RUN: --target=x86_64-linux-gnu 2>&1 \
+// RUN: | FileCheck libcxx-no-module-json.cpp
+
+// RUN: touch %t/Inputs/usr/lib/x86_64-linux-gnu/modules.json
+// RUN: %clang -print-library-module-manifest-path \
+// RUN: -stdlib=libc++ \
+// RUN: -resource-dir=%t/Inputs/usr/lib/x86_64-linux-gnu \
+// RUN: --target=x86_64-linux-gnu 2>&1 \
+// RUN: | FileCheck libcxx.cpp
+
+// RUN: %clang -print-library-module-manifest-path \
+// RUN: -stdlib=libstdc++ \
+// RUN: -resource-dir=%t/Inputs/usr/lib/x86_64-linux-gnu \
+// RUN: --target=x86_64-linux-gnu 2>&1 \
+// RUN: | FileCheck libstdcxx.cpp
+
+//--- libcxx-no-module-json.cpp
+
+// CHECK: <NOT PRESENT>
+
+//--- libcxx.cpp
+
+// CHECK: {{.*}}/Inputs/usr/lib/x86_64-linux-gnu{{/|\\}}modules.json
+
+//--- libstdcxx.cpp
+
+// CHECK: <NOT PRESENT>
|
There was a problem hiding this comment.
Thanks for picking it up @ChuanqiXu9 and thanks for testing @kaz7.
|
I'll backport this tomorrow if no revert request shows up. |
|
How do we use this ? and but no manifest path is returned |
|
The file we're looking for is |
BTW, how do you feel about the paragraph. I thought it as a defect in some level. |
|
/cherry-pick 0e1e1fc |
but i didn't rename it, it was with the libc++ prefix directly :/ |
Following of llvm#82160 The reason why the above PR fails is that the `--sysroot` has lower priority than the libc++ built from the same source. On the one hand, it matches the codes behavior. We will add the built libc++ project paths in the ToolChain class. But we will only add the path related to sysroot in Linux class, which is derived from the ToolChain classes. So the paths of just built libc++ is in the front of the paths relative to sysroot. On the other hand, the behavior should be good from the higher level. Since the just built libc++ has the same version number with the just built clang, so it makes sense that these 2 compilers just matches. So for patch it self, I hacked it by using resource dir in the test since the resource dir has the higher priority, which is not strongly correct since we won't do that in practice. @kaz7 would you like to test on your environment to avoid this get reverted again? On the libc++ side, it shows that it lacks a `modules.json` file for the just built libc++ directory. If we don't have that, it will be problematic to use std modules from the just built clang and libc++ pair. Then it is not good. And I feel it may be problematic for future compiler/standard library developers. So I feel this is somewhat a libc++ issue that need to be fixed. Also if we don't like the hacked test in the current patch, we must wait for libc++ to fix this to proceed. But I feel this is somewhat odd since the test of clang shouldn't dependent on libc++. CC: @mordante --------- Co-authored-by: Mark de Wever <[email protected]> (cherry picked from commit 0e1e1fc)
|
/pull-request #85637 |
Good point I did :-/ It seems we originally talked about |
Could you elaborate the reason more? I feel like renaming |
It would allow to install libstdc++ and libc++ in the same lib directory without "fighting" who owns |
hmmmm, I can accept that. Then the patch itself is not incorrect... Since it looks like we'd never look for Would you like to pick up this? I feel you may be more appropriate since you understand the environment better. I'm also like to take it if you don't have the time. |
|
Build system don't use this yet (since it wasn't available). I know CMake is working on implementing modules using libc++. Looking at the stage of the release I feel less comfortable to propose new features. So I think the feature should not be back ported. I'm happy to pick this up, I expect to have time for it tomorrow. |
|
Can we also have a fallback check for stdlibs that don't have a std::string lib = GetFilePath("libc++.so", TC);
if (lib.empty())
lib = GetFilePath("libc++.a", TC); |
Excellent point! thanks! |
Following of llvm#82160 The reason why the above PR fails is that the `--sysroot` has lower priority than the libc++ built from the same source. On the one hand, it matches the codes behavior. We will add the built libc++ project paths in the ToolChain class. But we will only add the path related to sysroot in Linux class, which is derived from the ToolChain classes. So the paths of just built libc++ is in the front of the paths relative to sysroot. On the other hand, the behavior should be good from the higher level. Since the just built libc++ has the same version number with the just built clang, so it makes sense that these 2 compilers just matches. So for patch it self, I hacked it by using resource dir in the test since the resource dir has the higher priority, which is not strongly correct since we won't do that in practice. @kaz7 would you like to test on your environment to avoid this get reverted again? On the libc++ side, it shows that it lacks a `modules.json` file for the just built libc++ directory. If we don't have that, it will be problematic to use std modules from the just built clang and libc++ pair. Then it is not good. And I feel it may be problematic for future compiler/standard library developers. So I feel this is somewhat a libc++ issue that need to be fixed. Also if we don't like the hacked test in the current patch, we must wait for libc++ to fix this to proceed. But I feel this is somewhat odd since the test of clang shouldn't dependent on libc++. CC: @mordante --------- Co-authored-by: Mark de Wever <[email protected]> (cherry picked from commit 0e1e1fc)
Following of #82160
The reason why the above PR fails is that the
--sysroothas lower priority than the libc++ built from the same source. On the one hand, it matches the codes behavior. We will add the built libc++ project paths in the ToolChain class. But we will only add the path related to sysroot in Linux class, which is derived from the ToolChain classes. So the paths of just built libc++ is in the front of the paths relative to sysroot. On the other hand, the behavior should be good from the higher level. Since the just built libc++ has the same version number with the just built clang, so it makes sense that these 2 compilers just matches.So for patch it self, I hacked it by using resource dir in the test since the resource dir has the higher priority, which is not strongly correct since we won't do that in practice.
@kaz7 would you like to test on your environment to avoid this get reverted again?
On the libc++ side, it shows that it lacks a
modules.jsonfile for the just built libc++ directory. If we don't have that, it will be problematic to use std modules from the just built clang and libc++ pair. Then it is not good. And I feel it may be problematic for future compiler/standard library developers. So I feel this is somewhat a libc++ issue that need to be fixed.Also if we don't like the hacked test in the current patch, we must wait for libc++ to fix this to proceed. But I feel this is somewhat odd since the test of clang shouldn't dependent on libc++.
CC: @mordante