[libc++] Make sure flat_{multi}map::key_compare handle boolean-testable correctly

[hewillk]

This is sibling of #69378.

[llvmbot]

@llvm/pr-subscribers-libcxx

Author: Hewill Kang (hewillk)

Changes

This is sibling of #69378.

---

Full diff: https://github.com/llvm/llvm-project/pull/132621.diff

1 Files Affected:

• (modified) libcxx/include/__flat_map/flat_map.h (+2-2)

diff --git a/libcxx/include/__flat_map/flat_map.h b/libcxx/include/__flat_map/flat_map.h
index a0594ed9dc411..2f32497ab20d3 100644
--- a/libcxx/include/__flat_map/flat_map.h
+++ b/libcxx/include/__flat_map/flat_map.h
@@ -846,7 +846,7 @@ class flat_map {
         __compare_(std::forward<_CompArg>(__comp)...) {}
 
   _LIBCPP_HIDE_FROM_ABI bool __is_sorted_and_unique(auto&& __key_container) const {
-    auto __greater_or_equal_to = [this](const auto& __x, const auto& __y) { return !__compare_(__x, __y); };
+    auto __greater_or_equal_to = [this](const auto& __x, const auto& __y) -> bool { return !__compare_(__x, __y); };
     return ranges::adjacent_find(__key_container, __greater_or_equal_to) == ranges::end(__key_container);
   }
 
@@ -870,7 +870,7 @@ class flat_map {
       auto __zv                  = ranges::views::zip(__containers_.keys, __containers_.values);
       auto __append_start_offset = __containers_.keys.size() - __num_of_appended;
       auto __end                 = __zv.end();
-      auto __compare_key         = [this](const auto& __p1, const auto& __p2) {
+      auto __compare_key         = [this](const auto& __p1, const auto& __p2) -> bool {
         return __compare_(std::get<0>(__p1), std::get<0>(__p2));
       };
       if constexpr (!_WasSorted) {

[frederick-vs-ja]

I think it's expected that decltype(__compare_(__x, __y)) already model boolean-testable (or even is exactly bool in most normal usages).

If the standard were already so requiring, we wouldn't need to change these lambdas. But I haven't found such a requirement...

[frederick-vs-ja]

Oh, I was likely to be wrong. But are such conversions already performed in algorithms?

[hewillk]

Oh, I was likely to be wrong. But are such conversions already performed in algorithms?

https://godbolt.org/z/nbG9v4zqY

[frederick-vs-ja]

Oh, I see. It's the nature of lambda expression without trailing return type that unfortunately decay-copies the result.

Perhaps there should be some test cases for this, with types in boolean_testable.h reused?

[hewillk]

Perhaps there should be some test cases for this, with types in boolean_testable.h reused?

I don't have a local repository for llvm, co-working is welcome.

[ldionne]

Can we add tests similarly to what we do in libcxx/test/std/algorithms/ranges_robust_against_nonbool.compile.pass.cpp?

[hewillk]

Can we add tests similarly to what we do in libcxx/test/std/algorithms/ranges_robust_against_nonbool.compile.pass.cpp?

No problem, I can handle this when I have the time .

[huixie90]

thanks for the patch. if you are not familiar with libc++ tests, i can certainly help with the testing for this patch. let me know if you want me to add the test to your branch