New issue from Jiang An: "Should swapping a built-in array or std::array with itself result in UB?"

Author: Dani-Hub

xml/issue4165.xml

@@ -0,0 +1,30 @@
+<?xml version='1.0' encoding='utf-8' standalone='no'?>
+<!DOCTYPE issue SYSTEM "lwg-issue.dtd">
+
+<issue num="4165" status="New">
+<title>Should swapping a built-in array or `std::array` with itself result in UB?</title>
+<section><sref ref="[concept.swappable]"/><sref ref="[utility.swap]"/><sref ref="[array.members]"/></section>
+<submitter>Jiang An</submitter>
+<date>13 Oct 2024</date>
+<priority>99</priority>
+
+<discussion>
+<p>
+Currently, the `std::swap` overload for built-in arrays, the `swap` member function of 
+`std::array`, the `std::swap` overload for `std::array`, and the expression-equivalent-to 
+operation of `std::ranges::swap` for built-in arrays are specified to be equivalent to performing 
+`std::(ranges::)swap_ranges`. As `swap_ranges` functions require that the swapped ranges shall 
+not overlap (<sref ref="[alg.swap]"/>/2), swapping a built-in array or a nonzero-length 
+<tt>std::array</tt> with itself results in undefined behavior.
+<p/>
+It seems that no other swap operation in the standard library has precondition that both arguments 
+don't refer to the same object. Should we make self-swapping of arrays well-defined? If so, 
+perhaps we should change the semantics to plain loops that execute `swap` or `ranges::swap` on 
+elements.
+</p>
+</discussion>
+
+<resolution>
+</resolution>
+
+</issue>