From 3ba196391f2eca99d709d688787779252536c785 Mon Sep 17 00:00:00 2001
From: Arthur O'Dwyer <arthur.j.odwyer@gmail.com>
Date: Sun, 31 Dec 2023 19:32:32 -0500
Subject: [PATCH] D1144R10: Relevance of operator= (this should become a blog
 post, too)

---
 d1144-object-relocation.bs   | 108 +++++++++++++++++++++++----------
 d1144-object-relocation.html | 112 ++++++++++++++++++++++++-----------
 2 files changed, 156 insertions(+), 64 deletions(-)

diff --git a/d1144-object-relocation.bs b/d1144-object-relocation.bs
index 6766ec1b36d..e01b5ebcf85 100644
--- a/d1144-object-relocation.bs
+++ b/d1144-object-relocation.bs
@@ -2,7 +2,7 @@
 Title: std::is_trivially_relocatable
 Shortname: D1144
 Revision: 10
-!Draft Revision: 59
+!Draft Revision: 60
 Audience: LEWG, EWG
 Status: D
 Group: WG21
@@ -16,7 +16,7 @@ Abstract:
   We propose a compiler rule that propagates trivial relocatability among Rule-of-Zero types.
   Finally, we propose a standard syntax for a user-defined type (e.g. <code>boost::container::static_vector</code>) to warrant
   to the implementation that it is trivially relocatable.
-Date: 2023-12-22
+Date: 2023-12-31
 </pre>
 
 <style>
@@ -29,7 +29,8 @@ del  {background-color: #FFCACA; text-decoration: line-through;}
 
 - R10:
 
-    - Added implementation experience with [[HPX]], [[ParlayLib]], and [[StdxError]].
+    - Added implementation experience with [[HPX]], [[ParlayLib]], [[AMC]], and [[StdxError]].
+            Added <a href="#relevance-of-assignment">§5.2 "Relevance of `operator=`"</a>.
 
     - Wording for `ExecutionPolicy` overloads of the specialized algorithms. Explain why
             there are no `ranges` overloads.
@@ -46,7 +47,7 @@ del  {background-color: #FFCACA; text-decoration: line-through;}
     - Replaced [[#non-trivial-samples|Appendix C "Examples of non-trivially relocatable class types"]]
             with a shorter summary.
 
-    - Removed Appendix E "Open questions" as basically redundant with <a href="#pmr-concerns">§5.2 "Confusing interactions with `std::pmr`"</a>.
+    - Removed Appendix E "Open questions" as basically redundant with <a href="#pmr-concerns">§5.3 "Confusing interactions with `std::pmr`"</a>.
             See ["P1144 PMR koans"](https://quuxplusone.github.io/blog/2023/06/03/p1144-pmr-koans/) (June 2023).
 
 - R8 (pre-Varna 2023):
@@ -61,11 +62,7 @@ del  {background-color: #FFCACA; text-decoration: line-through;}
 
     - Marked `std::relocate` as `[[nodiscard]]`; changed its return type to `remove_cv_t<T>`.
 
-    - Resolved the Appendix E question about potentially overlapping subobjects;
-            we have implementation experience now that proves this is not a problem. Expanded discussion
-            of `pmr` types.
-
-    - Replaced [[#design-goals|§3 "Design goals"]] with a shorter summary.
+    - Rewrote [[#design-goals|§3 "Design goals."]]
 
 - R7 (post-Issaquah 2023):
 
@@ -79,19 +76,6 @@ del  {background-color: #FFCACA; text-decoration: line-through;}
     - Added straw poll results from Issaquah; removed much background and discussion
             (leaving only notes for the interested reader to consult [[P1144R6]]).
 
-    - Removed mentions of EASTL.
-
-    - Added Appendix E "Open questions."
-
-- R6:
-
-    - Added `T std::relocate(T *source)`.
-
-    - Changed the exception guarantee of `uninitialized_relocate` and `uninitialized_relocate_n`
-            in accord with feedback from David Stone.
-
-    - Removed `std::ranges::relocate_at` (but kept `std::relocate_at`).
-
 
 # Introduction and motivation # {#intro}
 
@@ -268,7 +252,10 @@ the programmer to warrant the trivial relocatability of any type stored in a `fo
 
 This merely encouraged the programmer to add the warrant and continue. An incorrect
 warrant was discovered only at runtime, via undefined behavior. See
-[[FollyIssue889]] and (most importantly) [[CppNow]] [@27:26–31:47](https://www.youtube.com/watch?v=SGdfPextuAU&t=27m26s).
+[[CppNow]] [@27:26–31:47](https://www.youtube.com/watch?v=SGdfPextuAU&t=27m26s);
+for real-world examples see Folly issue [#889](https://github.com/facebook/folly/issues/889)
+([#35](https://github.com/facebook/folly/issues/35), [#316](https://github.com/facebook/folly/issues/316),
+[#498](https://github.com/facebook/folly/issues/498)).
 
 ```c++
     class Gadget {
@@ -303,7 +290,7 @@ can even find real bugs. [Example:](https://p1144.godbolt.org/z/7b8893sjr)
     static_assert(std::is_trivially_relocatable_v<Gadget>);  // correctly ERRORS
 ```
 
-The improvement in user experience and safety in real-world codebases ([[Folly]], [[BSL]], [[Qt]], [[HPX]], [[ParlayLib]], etc.)
+The improvement in user experience and safety in real-world codebases ([[Folly]], [[BSL]], [[Qt]], [[HPX]], [[ParlayLib]], [[AMC]], etc.)
 is the most important benefit to be gained by this proposal.
 
 # Design goals # {#design-goals}
@@ -894,6 +881,64 @@ EWGI took a three-way straw poll on `[[trivially_relocatable]]` versus `[[maybe_
 with an inconclusive 7–5–6 vote (the author of P1144 voting "For" and the three representatives of P2786
 voting "Against"; i.e. the vote sans authors was 6–5–3).
 
+## Relevance of `operator=` ## {#relevance-of-assignment}
+
+Here's how current implementations define "is trivially relocatable":
+
+- Mainline Clang `__is_trivially_relocatable(T)`: Trivially copyable, or has `[[clang::trivial_abi]]`
+- Arthur's Clang (P1144) `__is_trivially_relocatable(T)`: Trivially copyable, or has `[[clang::trivial_abi]]`, or has `[[trivially_relocatable]]`
+- Amadeus [[AMC]]'s `amc::is_trivially_relocatable<T>`: `is_trivially_copyable<T>` or has member typedef `T::trivially_relocatable`
+- Facebook [[Folly]]'s `folly::IsRelocatable<T>`: `is_trivially_copyable<T>` or has member typedef `T::IsRelocatable`
+- Bloomberg [[BSL]]'s `bslmf::IsBitwiseMoveable<T>`: Has a conversion to `bslmf::NestedTraitDeclaration<T, bslmf::IsBitwiseMoveable, true>`,
+    or has a conversion to `bslmf::NestedTraitDeclaration<T, bslmf::IsBitwiseCopyable, true>`, or `is_trivially_copyable<T>`,
+    or `sizeof(T) == 1`
+- [[HPX]]'s `hpx::experimental::is_trivially_relocatable<T>`: `is_trivially_copyable<T>`, or the trait is specialized for `T`
+- [[ParlayLib]]'s `parlay::is_trivially_relocatable<T>`: currently "`is_trivially_move_constructible_v<T> && is_trivially_destructible_v<T>`,"
+    but after [#67](https://github.com/cmuparlay/parlaylib/pull/67) it'll be "`is_trivially_copyable<T>`, or the trait is specialized for `T`"
+
+Notice that all implementors (except ParlayLib trunk) agree that a type with trivial move-constructor, trivial destructor,
+*and non-trivial assignment operator* is considered non-trivially relocatable. That is, the assignment operator is relevant!
+P1144 preserves that behavior, and justifies it by pointing out that relocation
+can replace assignment in `vector::erase` and `std::swap`, among other algorithms. Existing codebases have written a lot
+of code that branches on `is_trivially_relocatable`, taking an optimized codepath when `is_trivially_relocatable` and an
+unoptimized codepath otherwise. P1144 carefully proposes a definition for `std::is_trivially_relocatable` that keeps
+those codebases safe, and never unexpectedly reports a type `T` as `is_trivially_relocatable` when it's not physically safe
+to use `T` with such an optimized codepath.
+
+```c++
+    #include <bsl_vector.h>
+    using namespace BloombergLP::bslmf;
+
+    struct T : NestedTraitDeclaration<T, IsBitwiseMoveable> {
+        int i_;
+        T(int i) : i_(i) {}
+        T(const T&) = default;
+        void operator=(const T&) noexcept { puts("Called operator="); }
+        ~T() = default;
+    };
+    int main() {
+        bsl::vector<T> v = {1,2};
+        v.erase(v.begin());
+          // prints nothing; bsl::vector::erase is optimized
+    }
+
+    struct U {
+        int i_;
+        U(int i) : i_(i) {}
+        U(const U&) = default;
+        void operator=(const U&) noexcept { puts("Called operator="); }
+        ~U() = default;
+    };
+    int main() {
+        bsl::vector<U> v = {1,2};
+        v.erase(v.begin());
+          // prints "Called operator="; BSL believes that a non-defaulted
+          // assignment operator makes U non-trivially relocatable.
+          // P1144 agrees. std::is_trivially_relocatable_v<U>
+          // must remain false, so as not to silently change the
+          // behavior of bsl::vector<U>.
+    }
+```
 
 ## Confusing interactions with `std::pmr` ## {#pmr-concerns}
 
@@ -1196,6 +1241,14 @@ See [[#applications|table §2.1]] for more details on [[Qt]], [[Folly]], and [[B
 
 <pre class=biblio>
 {
+  "AMC": {
+    "authors": [
+      "Stephane Janel"
+    ],
+    "title": "AMadeus (C++) Containers",
+    "href": "https://github.com/AmadeusITGroup/amc",
+    "date": "April 2021"
+  },
   "Announcing": {
     "authors": [
       "Arthur O'Dwyer"
@@ -1285,13 +1338,6 @@ See [[#applications|table §2.1]] for more details on [[Qt]], [[Folly]], and [[B
     "title": "Folly documentation on \"Object Relocation\"",
     "href": "https://github.com/facebook/folly/blob/master/folly/docs/FBVector.md#object-relocation"
   },
-  "FollyIssue889": {
-    "authors": [
-      "Arthur O'Dwyer"
-    ],
-    "title": "Traits.h marks std::list as trivially relocatable, but in fact it is not",
-    "href": "https://github.com/facebook/folly/issues/889"
-  },
   "HPX": {
     "authors": [
       "Isidoros Tsaousis-Seiras"
diff --git a/d1144-object-relocation.html b/d1144-object-relocation.html
index 07ed1f10536..e36b0658e11 100644
--- a/d1144-object-relocation.html
+++ b/d1144-object-relocation.html
@@ -2094,7 +2094,7 @@
   <div class="head">
    <p data-fill-with="logo"></p>
    <h1 class="p-name no-ref" id="title">D1144R10<br>std::is_trivially_relocatable</h1>
-   <h2 class="no-num no-toc no-ref heading settled" id="profile-and-date"><span class="content">Draft Proposal, <time class="dt-updated" datetime="2023-12-22">2023-12-22</time></span></h2>
+   <h2 class="no-num no-toc no-ref heading settled" id="profile-and-date"><span class="content">Draft Proposal, <time class="dt-updated" datetime="2023-12-31">2023-12-31</time></span></h2>
    <div data-fill-with="spec-metadata">
     <dl>
      <dt>Issue Tracking:
@@ -2106,7 +2106,7 @@ <h2 class="no-num no-toc no-ref heading settled" id="profile-and-date"><span cla
      <dt>Project:
      <dd>ISO/IEC 14882 Programming Languages — C++, ISO/IEC JTC1/SC22/WG21
      <dt>Draft Revision:
-     <dd>59
+     <dd>60
     </dl>
    </div>
    <div data-fill-with="warning"></div>
@@ -2169,7 +2169,8 @@ <h2 class="no-num no-toc no-ref" id="contents">Table of Contents</h2>
      <a href="#alternatives"><span class="secno">5</span> <span class="content">Rationale and alternatives</span></a>
      <ol class="toc">
       <li><a href="#maybe-trivially-relocatable"><span class="secno">5.1</span> <span class="content">Attribute <code class="highlight"><c- p>[[</c-><c- n>maybe_trivially_relocatable</c-><c- p>]]</c-></code></span></a>
-      <li><a href="#pmr-concerns"><span class="secno">5.2</span> <span class="content">Confusing interactions with <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>pmr</c-></code></span></a>
+      <li><a href="#relevance-of-assignment"><span class="secno">5.2</span> <span class="content">Relevance of <code class="highlight"><c- k>operator</c-><c- o>=</c-></code></span></a>
+      <li><a href="#pmr-concerns"><span class="secno">5.3</span> <span class="content">Confusing interactions with <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>pmr</c-></code></span></a>
      </ol>
     <li><a href="#acknowledgements"><span class="secno">6</span> <span class="content">Acknowledgements</span></a>
     <li>
@@ -2201,7 +2202,8 @@ <h2 class="heading settled" data-level="1" id="changelog"><span class="secno">1.
      <p>R10:</p>
      <ul>
       <li data-md>
-       <p>Added implementation experience with <a data-link-type="biblio" href="#biblio-hpx" title="Relocation Semantics in the HPX Library">[HPX]</a>, <a data-link-type="biblio" href="#biblio-parlaylib" title="ParlayLib, a toolkit for programming parallel algorithms on shared-memory multicore machines">[ParlayLib]</a>, and <a data-link-type="biblio" href="#biblio-stdxerror" title="Implementation of std::error as proposed by Herb Sutter in P0709R0">[StdxError]</a>.</p>
+       <p>Added implementation experience with <a data-link-type="biblio" href="#biblio-hpx" title="Relocation Semantics in the HPX Library">[HPX]</a>, <a data-link-type="biblio" href="#biblio-parlaylib" title="ParlayLib, a toolkit for programming parallel algorithms on shared-memory multicore machines">[ParlayLib]</a>, <a data-link-type="biblio" href="#biblio-amc" title="AMadeus (C++) Containers">[AMC]</a>, and <a data-link-type="biblio" href="#biblio-stdxerror" title="Implementation of std::error as proposed by Herb Sutter in P0709R0">[StdxError]</a>.
+    Added <a href="#relevance-of-assignment">§5.2 "Relevance of <code class="highlight"><c- k>operator</c-><c- o>=</c-></code>"</a>.</p>
       <li data-md>
        <p>Wording for <code class="highlight"><c- n>ExecutionPolicy</c-></code> overloads of the specialized algorithms. Explain why
     there are no <code class="highlight"><c- n>ranges</c-></code> overloads.</p>
@@ -2219,7 +2221,7 @@ <h2 class="heading settled" data-level="1" id="changelog"><span class="secno">1.
       <li data-md>
        <p>Replaced <a href="#non-trivial-samples">Appendix C "Examples of non-trivially relocatable class types"</a> with a shorter summary.</p>
       <li data-md>
-       <p>Removed Appendix E "Open questions" as basically redundant with <a href="#pmr-concerns">§5.2 "Confusing interactions with <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>pmr</c-></code>"</a>.
+       <p>Removed Appendix E "Open questions" as basically redundant with <a href="#pmr-concerns">§5.3 "Confusing interactions with <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>pmr</c-></code>"</a>.
     See <a href="https://quuxplusone.github.io/blog/2023/06/03/p1144-pmr-koans/">"P1144 PMR koans"</a> (June 2023).</p>
      </ul>
     <li data-md>
@@ -2235,11 +2237,7 @@ <h2 class="heading settled" data-level="1" id="changelog"><span class="secno">1.
       <li data-md>
        <p>Marked <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>relocate</c-></code> as <code class="highlight"><c- p>[[</c-><c- n>nodiscard</c-><c- p>]]</c-></code>; changed its return type to <code class="highlight"><c- n>remove_cv_t</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>.</p>
       <li data-md>
-       <p>Resolved the Appendix E question about potentially overlapping subobjects;
-    we have implementation experience now that proves this is not a problem. Expanded discussion
-    of <code class="highlight"><c- n>pmr</c-></code> types.</p>
-      <li data-md>
-       <p>Replaced <a href="#design-goals">§3 "Design goals"</a> with a shorter summary.</p>
+       <p>Rewrote <a href="#design-goals">§3 "Design goals."</a></p>
      </ul>
     <li data-md>
      <p>R7 (post-Issaquah 2023):</p>
@@ -2254,20 +2252,6 @@ <h2 class="heading settled" data-level="1" id="changelog"><span class="secno">1.
       <li data-md>
        <p>Added straw poll results from Issaquah; removed much background and discussion
     (leaving only notes for the interested reader to consult <a data-link-type="biblio" href="#biblio-p1144r6" title="Object relocation in terms of move plus destroy">[P1144R6]</a>).</p>
-      <li data-md>
-       <p>Removed mentions of EASTL.</p>
-      <li data-md>
-       <p>Added Appendix E "Open questions."</p>
-     </ul>
-    <li data-md>
-     <p>R6:</p>
-     <ul>
-      <li data-md>
-       <p>Added <code class="highlight"><c- n>T</c-> <c- n>std</c-><c- o>::</c-><c- n>relocate</c-><c- p>(</c-><c- n>T</c-> <c- o>*</c-><c- n>source</c-><c- p>)</c-></code>.</p>
-      <li data-md>
-       <p>Changed the exception guarantee of <code class="highlight"><c- n>uninitialized_relocate</c-></code> and <code class="highlight"><c- n>uninitialized_relocate_n</c-></code> in accord with feedback from David Stone.</p>
-      <li data-md>
-       <p>Removed <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>ranges</c-><c- o>::</c-><c- n>relocate_at</c-></code> (but kept <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>relocate_at</c-></code>).</p>
      </ul>
    </ul>
    <h2 class="heading settled" data-level="2" id="intro"><span class="secno">2. </span><span class="content">Introduction and motivation</span><a class="self-link" href="#intro"></a></h2>
@@ -2365,7 +2349,7 @@ <h4 class="heading settled" data-level="2.1.4" id="benefit-insert-erase"><span c
 then construct the new elements in place). <a data-link-type="biblio" href="#biblio-folly" title="Folly documentation on &quot;Object Relocation&quot;">[Folly]</a>'s <code class="highlight"><c- n>fbvector</c-></code> does
 this optimization; see <a href="https://github.com/facebook/folly/blob/1d4690d0a3/folly/FBVector.h#L1273-L1308"><code class="highlight"><c- n>fbvector</c-><c- o>::</c-><c- n>make_window</c-></code></a>.
 Bloomberg’s <a data-link-type="biblio" href="#biblio-bsl" title="bslmf::IsBitwiseMoveable: bitwise moveable trait metafunction">[BSL]</a> also does this optimization.</p>
-   <p>But <a href="#pmr-concerns">§ 5.2 Confusing interactions with std::pmr</a> determines whether this optimization will be possible for certain types.</p>
+   <p>But <a href="#pmr-concerns">§ 5.3 Confusing interactions with std::pmr</a> determines whether this optimization will be possible for certain types.</p>
    <h4 class="heading settled" data-level="2.1.5" id="benefit-swap"><span class="secno">2.1.5. </span><span class="content">Swap</span><a class="self-link" href="#benefit-swap"></a></h4>
    <p>Given a reliable way of detecting trivial relocatability,
 we can optimize any routine that uses the moral equivalent of <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>swap</c-></code>, such as <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>rotate</c-></code>, <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>partition</c-></code>, or <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>sort</c-></code>.</p>
@@ -2374,7 +2358,7 @@ <h4 class="heading settled" data-level="2.1.5" id="benefit-swap"><span class="se
 See <a data-link-type="biblio" href="#biblio-cppnow" title="Trivially Relocatable (C++Now 2019)">[CppNow]</a> <a href="https://www.youtube.com/watch?v=SGdfPextuAU&amp;t=19m56s">@19:56–21:22</a>,
 and see <a href="https://p1144.godbolt.org/z/PPYhcYd8d">this Godbolt</a>. <a href="https://quick-bench.com/q/suAss7ElmwgLDbOJQGnlEseheSQ">This Quick-Bench</a> shows a 25% speedup
 in <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>rotate</c-></code> when it is allowed to use bitwise swap on a Rule-of-Zero type.</p>
-   <p>But <a href="#pmr-concerns">§ 5.2 Confusing interactions with std::pmr</a> determines whether this optimization will be possible for certain types.</p>
+   <p>But <a href="#pmr-concerns">§ 5.3 Confusing interactions with std::pmr</a> determines whether this optimization will be possible for certain types.</p>
    <h3 class="heading settled" data-level="2.2" id="benefit-assertions"><span class="secno">2.2. </span><span class="content">Assertions, not assumptions</span><a class="self-link" href="#benefit-assertions"></a></h3>
    <p>Some data structures might reasonably assert the trivial relocatability of
 their elements, just as they sometimes assert the stronger property of trivial <em>copyability</em> today.
@@ -2391,7 +2375,8 @@ <h3 class="heading settled" data-level="2.3" id="doing-it-wrong"><span class="se
 <c- n>folly</c-><c- o>::</c-><c- n>fbvector</c-><c- o>&lt;</c-><c- n>Widget</c-><c- o>></c-> <c- n>vec</c-><c- p>;</c->  <c- c1>// FAIL AT COMPILE TIME for lack of warrant</c->
 </pre>
    <p>This merely encouraged the programmer to add the warrant and continue. An incorrect
-warrant was discovered only at runtime, via undefined behavior. See <a data-link-type="biblio" href="#biblio-follyissue889" title="Traits.h marks std::list as trivially relocatable, but in fact it is not">[FollyIssue889]</a> and (most importantly) <a data-link-type="biblio" href="#biblio-cppnow" title="Trivially Relocatable (C++Now 2019)">[CppNow]</a> <a href="https://www.youtube.com/watch?v=SGdfPextuAU&amp;t=27m26s">@27:26–31:47</a>.</p>
+warrant was discovered only at runtime, via undefined behavior. See <a data-link-type="biblio" href="#biblio-cppnow" title="Trivially Relocatable (C++Now 2019)">[CppNow]</a> <a href="https://www.youtube.com/watch?v=SGdfPextuAU&amp;t=27m26s">@27:26–31:47</a>;
+for real-world examples see Folly issue <a href="https://github.com/facebook/folly/issues/889">#889</a> (<a href="https://github.com/facebook/folly/issues/35">#35</a>, <a href="https://github.com/facebook/folly/issues/316">#316</a>, <a href="https://github.com/facebook/folly/issues/498">#498</a>).</p>
 <pre class="language-c++ highlight"><c- n>class</c-> <c- n>Gadget</c-> <c- p>{</c->
     <c- n>std</c-><c- o>::</c-><c- n>list</c-><c- o>&lt;</c-><c- b>int</c-><c- o>></c-> <c- n>lst_</c-><c- p>;</c->
 <c- p>};</c->
@@ -2418,7 +2403,7 @@ <h3 class="heading settled" data-level="2.3" id="doing-it-wrong"><span class="se
 <c- p>};</c->
 <c- k>static_assert</c-><c- p>(</c-><c- n>std</c-><c- o>::</c-><c- n>is_trivially_relocatable_v</c-><c- o>&lt;</c-><c- n>Gadget</c-><c- o>></c-><c- p>);</c->  <c- c1>// correctly ERRORS</c->
 </pre>
-   <p>The improvement in user experience and safety in real-world codebases (<a data-link-type="biblio" href="#biblio-folly" title="Folly documentation on &quot;Object Relocation&quot;">[Folly]</a>, <a data-link-type="biblio" href="#biblio-bsl" title="bslmf::IsBitwiseMoveable: bitwise moveable trait metafunction">[BSL]</a>, <a data-link-type="biblio" href="#biblio-qt" title="Qt Base">[Qt]</a>, <a data-link-type="biblio" href="#biblio-hpx" title="Relocation Semantics in the HPX Library">[HPX]</a>, <a data-link-type="biblio" href="#biblio-parlaylib" title="ParlayLib, a toolkit for programming parallel algorithms on shared-memory multicore machines">[ParlayLib]</a>, etc.)
+   <p>The improvement in user experience and safety in real-world codebases (<a data-link-type="biblio" href="#biblio-folly" title="Folly documentation on &quot;Object Relocation&quot;">[Folly]</a>, <a data-link-type="biblio" href="#biblio-bsl" title="bslmf::IsBitwiseMoveable: bitwise moveable trait metafunction">[BSL]</a>, <a data-link-type="biblio" href="#biblio-qt" title="Qt Base">[Qt]</a>, <a data-link-type="biblio" href="#biblio-hpx" title="Relocation Semantics in the HPX Library">[HPX]</a>, <a data-link-type="biblio" href="#biblio-parlaylib" title="ParlayLib, a toolkit for programming parallel algorithms on shared-memory multicore machines">[ParlayLib]</a>, <a data-link-type="biblio" href="#biblio-amc" title="AMadeus (C++) Containers">[AMC]</a>, etc.)
 is the most important benefit to be gained by this proposal.</p>
    <h2 class="heading settled" data-level="3" id="design-goals"><span class="secno">3. </span><span class="content">Design goals</span><a class="self-link" href="#design-goals"></a></h2>
    <p>For the former, longer, contents of this section, see <a href="https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2023/p1144r7.html#design-goals">P1144R7 §3</a>.</p>
@@ -2795,7 +2780,68 @@ <h3 class="heading settled" data-level="5.1" id="maybe-trivially-relocatable"><s
 EWGI took a three-way straw poll on <code class="highlight"><c- p>[[</c-><c- n>trivially_relocatable</c-><c- p>]]</c-></code> versus <code class="highlight"><c- p>[[</c-><c- n>maybe_trivially_relocatable</c-><c- p>]]</c-></code>,
 with an inconclusive 7–5–6 vote (the author of P1144 voting "For" and the three representatives of P2786
 voting "Against"; i.e. the vote sans authors was 6–5–3).</p>
-   <h3 class="heading settled" data-level="5.2" id="pmr-concerns"><span class="secno">5.2. </span><span class="content">Confusing interactions with <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>pmr</c-></code></span><a class="self-link" href="#pmr-concerns"></a></h3>
+   <h3 class="heading settled" data-level="5.2" id="relevance-of-assignment"><span class="secno">5.2. </span><span class="content">Relevance of <code class="highlight"><c- k>operator</c-><c- o>=</c-></code></span><a class="self-link" href="#relevance-of-assignment"></a></h3>
+   <p>Here’s how current implementations define "is trivially relocatable":</p>
+   <ul>
+    <li data-md>
+     <p>Mainline Clang <code class="highlight"><c- n>__is_trivially_relocatable</c-><c- p>(</c-><c- n>T</c-><c- p>)</c-></code>: Trivially copyable, or has <code class="highlight"><c- p>[[</c-><c- n>clang</c-><c- o>::</c-><c- n>trivial_abi</c-><c- p>]]</c-></code></p>
+    <li data-md>
+     <p>Arthur’s Clang (P1144) <code class="highlight"><c- n>__is_trivially_relocatable</c-><c- p>(</c-><c- n>T</c-><c- p>)</c-></code>: Trivially copyable, or has <code class="highlight"><c- p>[[</c-><c- n>clang</c-><c- o>::</c-><c- n>trivial_abi</c-><c- p>]]</c-></code>, or has <code class="highlight"><c- p>[[</c-><c- n>trivially_relocatable</c-><c- p>]]</c-></code></p>
+    <li data-md>
+     <p>Amadeus <a data-link-type="biblio" href="#biblio-amc" title="AMadeus (C++) Containers">[AMC]</a>'s <code class="highlight"><c- n>amc</c-><c- o>::</c-><c- n>is_trivially_relocatable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>: <code class="highlight"><c- n>is_trivially_copyable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code> or has member typedef <code class="highlight"><c- n>T</c-><c- o>::</c-><c- n>trivially_relocatable</c-></code></p>
+    <li data-md>
+     <p>Facebook <a data-link-type="biblio" href="#biblio-folly" title="Folly documentation on &quot;Object Relocation&quot;">[Folly]</a>'s <code class="highlight"><c- n>folly</c-><c- o>::</c-><c- n>IsRelocatable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>: <code class="highlight"><c- n>is_trivially_copyable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code> or has member typedef <code class="highlight"><c- n>T</c-><c- o>::</c-><c- n>IsRelocatable</c-></code></p>
+    <li data-md>
+     <p>Bloomberg <a data-link-type="biblio" href="#biblio-bsl" title="bslmf::IsBitwiseMoveable: bitwise moveable trait metafunction">[BSL]</a>'s <code class="highlight"><c- n>bslmf</c-><c- o>::</c-><c- n>IsBitwiseMoveable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>: Has a conversion to <code class="highlight"><c- n>bslmf</c-><c- o>::</c-><c- n>NestedTraitDeclaration</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>bslmf</c-><c- o>::</c-><c- n>IsBitwiseMoveable</c-><c- p>,</c-> true<c- o>></c-></code>,
+or has a conversion to <code class="highlight"><c- n>bslmf</c-><c- o>::</c-><c- n>NestedTraitDeclaration</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>bslmf</c-><c- o>::</c-><c- n>IsBitwiseCopyable</c-><c- p>,</c-> true<c- o>></c-></code>, or <code class="highlight"><c- n>is_trivially_copyable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>,
+or <code class="highlight"><c- k>sizeof</c-><c- p>(</c-><c- n>T</c-><c- p>)</c-> <c- o>==</c-> <c- mi>1</c-></code></p>
+    <li data-md>
+     <p><a data-link-type="biblio" href="#biblio-hpx" title="Relocation Semantics in the HPX Library">[HPX]</a>'s <code class="highlight"><c- n>hpx</c-><c- o>::</c-><c- n>experimental</c-><c- o>::</c-><c- n>is_trivially_relocatable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>: <code class="highlight"><c- n>is_trivially_copyable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>, or the trait is specialized for <code class="highlight"><c- n>T</c-></code></p>
+    <li data-md>
+     <p><a data-link-type="biblio" href="#biblio-parlaylib" title="ParlayLib, a toolkit for programming parallel algorithms on shared-memory multicore machines">[ParlayLib]</a>'s <code class="highlight"><c- n>parlay</c-><c- o>::</c-><c- n>is_trivially_relocatable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>: currently "<code class="highlight"><c- n>is_trivially_move_constructible_v</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-> <c- o>&amp;&amp;</c-> <c- n>is_trivially_destructible_v</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>,"
+but after <a href="https://github.com/cmuparlay/parlaylib/pull/67">#67</a> it’ll be "<code class="highlight"><c- n>is_trivially_copyable</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-></code>, or the trait is specialized for <code class="highlight"><c- n>T</c-></code>"</p>
+   </ul>
+   <p>Notice that all implementors (except ParlayLib trunk) agree that a type with trivial move-constructor, trivial destructor, <em>and non-trivial assignment operator</em> is considered non-trivially relocatable. That is, the assignment operator is relevant!
+P1144 preserves that behavior, and justifies it by pointing out that relocation
+can replace assignment in <code class="highlight"><c- n>vector</c-><c- o>::</c-><c- n>erase</c-></code> and <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>swap</c-></code>, among other algorithms. Existing codebases have written a lot
+of code that branches on <code class="highlight"><c- n>is_trivially_relocatable</c-></code>, taking an optimized codepath when <code class="highlight"><c- n>is_trivially_relocatable</c-></code> and an
+unoptimized codepath otherwise. P1144 carefully proposes a definition for <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>is_trivially_relocatable</c-></code> that keeps
+those codebases safe, and never unexpectedly reports a type <code class="highlight"><c- n>T</c-></code> as <code class="highlight"><c- n>is_trivially_relocatable</c-></code> when it’s not physically safe
+to use <code class="highlight"><c- n>T</c-></code> with such an optimized codepath.</p>
+<pre class="language-c++ highlight"><c- cp>#include</c-> &lt;bsl_vector.h>
+<c- n>using</c-> <c- n>namespace</c-> <c- n>BloombergLP</c-><c- o>::</c-><c- n>bslmf</c-><c- p>;</c->
+
+<c- k>struct</c-> <c- nc>T</c-> <c- o>:</c-> <c- n>NestedTraitDeclaration</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>IsBitwiseMoveable</c-><c- o>></c-> <c- p>{</c->
+    <c- b>int</c-> <c- n>i_</c-><c- p>;</c->
+    <c- n>T</c-><c- p>(</c-><c- b>int</c-> <c- n>i</c-><c- p>)</c-> <c- o>:</c-> <c- n>i_</c-><c- p>(</c-><c- n>i</c-><c- p>)</c-> <c- p>{}</c->
+    <c- n>T</c-><c- p>(</c-><c- k>const</c-> <c- n>T</c-><c- o>&amp;</c-><c- p>)</c-> <c- o>=</c-> <c- k>default</c-><c- p>;</c->
+    <c- b>void</c-> <c- n>operator</c-><c- o>=</c-><c- p>(</c-><c- k>const</c-> <c- n>T</c-><c- o>&amp;</c-><c- p>)</c-> <c- n>noexcept</c-> <c- p>{</c-> <c- n>puts</c-><c- p>(</c-><c- s>"Called operator="</c-><c- p>);</c-> <c- p>}</c->
+    <c- o>~</c-><c- n>T</c-><c- p>()</c-> <c- o>=</c-> <c- k>default</c-><c- p>;</c->
+<c- p>};</c->
+<c- b>int</c-> <c- nf>main</c-><c- p>()</c-> <c- p>{</c->
+    <c- n>bsl</c-><c- o>::</c-><c- n>vector</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c-> <c- n>v</c-> <c- o>=</c-> <c- p>{</c-><c- mi>1</c-><c- p>,</c-><c- mi>2</c-><c- p>};</c->
+    <c- n>v</c-><c- p>.</c-><c- n>erase</c-><c- p>(</c-><c- n>v</c-><c- p>.</c-><c- n>begin</c-><c- p>());</c->
+      <c- c1>// prints nothing; bsl::vector::erase is optimized</c->
+<c- p>}</c->
+
+<c- k>struct</c-> <c- nc>U</c-> <c- p>{</c->
+    <c- b>int</c-> <c- n>i_</c-><c- p>;</c->
+    <c- n>U</c-><c- p>(</c-><c- b>int</c-> <c- n>i</c-><c- p>)</c-> <c- o>:</c-> <c- n>i_</c-><c- p>(</c-><c- n>i</c-><c- p>)</c-> <c- p>{}</c->
+    <c- n>U</c-><c- p>(</c-><c- k>const</c-> <c- n>U</c-><c- o>&amp;</c-><c- p>)</c-> <c- o>=</c-> <c- k>default</c-><c- p>;</c->
+    <c- b>void</c-> <c- n>operator</c-><c- o>=</c-><c- p>(</c-><c- k>const</c-> <c- n>U</c-><c- o>&amp;</c-><c- p>)</c-> <c- n>noexcept</c-> <c- p>{</c-> <c- n>puts</c-><c- p>(</c-><c- s>"Called operator="</c-><c- p>);</c-> <c- p>}</c->
+    <c- o>~</c-><c- n>U</c-><c- p>()</c-> <c- o>=</c-> <c- k>default</c-><c- p>;</c->
+<c- p>};</c->
+<c- b>int</c-> <c- nf>main</c-><c- p>()</c-> <c- p>{</c->
+    <c- n>bsl</c-><c- o>::</c-><c- n>vector</c-><c- o>&lt;</c-><c- n>U</c-><c- o>></c-> <c- n>v</c-> <c- o>=</c-> <c- p>{</c-><c- mi>1</c-><c- p>,</c-><c- mi>2</c-><c- p>};</c->
+    <c- n>v</c-><c- p>.</c-><c- n>erase</c-><c- p>(</c-><c- n>v</c-><c- p>.</c-><c- n>begin</c-><c- p>());</c->
+      <c- c1>// prints "Called operator="; BSL believes that a non-defaulted</c->
+      <c- c1>// assignment operator makes U non-trivially relocatable.</c->
+      <c- c1>// P1144 agrees. std::is_trivially_relocatable_v&lt;U></c->
+      <c- c1>// must remain false, so as not to silently change the</c->
+      <c- c1>// behavior of bsl::vector&lt;U>.</c->
+<c- p>}</c->
+</pre>
+   <h3 class="heading settled" data-level="5.3" id="pmr-concerns"><span class="secno">5.3. </span><span class="content">Confusing interactions with <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>pmr</c-></code></span><a class="self-link" href="#pmr-concerns"></a></h3>
    <p class="note" role="note"><span class="marker">Note:</span> See <a href="https://quuxplusone.github.io/blog/2023/06/03/p1144-pmr-koans/">"P1144 PMR koans"</a> (June 2023)
 for additional analysis of the problem cases.</p>
    <p class="note" role="note"><span class="marker">Note:</span> This section was added in P1144R5, revised in P1144R7, and again in P1144R8.
@@ -3027,10 +3073,10 @@ <h3 class="heading settled" id="implementation-core"><span class="content">Core-
    <p><a href="https://github.com/Quuxplusone/llvm-project/tree/trivially-relocatable">Arthur’s fork of Clang</a> implements
 all of P1144 (since 2018), and has been kept up-to-date with the latest P1144. See it in action <a href="https://p1144.godbolt.org/z/jf67zx5hq">on godbolt.org</a>, under the name "x86-64 clang (experimental P1144)."</p>
    <h3 class="heading settled" id="implementation-lib"><span class="content">Library implementations</span><a class="self-link" href="#implementation-lib"></a></h3>
-   <p>Since 2018, <a href="https://github.com/Quuxplusone/llvm-project/tree/trivially-relocatable">Arthur maintains a fork of libc++</a> with a fully optimizing implementation of P1144. See it in action <a href="https://p1144.godbolt.org/z/PzfW8azj5">on godbolt.org</a>, under the name "x86-64 clang (experimental P1144)."
+   <p>Since 2018, <a href="https://github.com/Quuxplusone/llvm-project/tree/trivially-relocatable">Arthur maintains a fork of libc++</a> with a fully optimizing implementation of P1144. See it in action <a href="https://p1144.godbolt.org/z/KasGGoE64">on godbolt.org</a>, under the name "x86-64 clang (experimental P1144)."
 (<a href="https://p1144.godbolt.org/z/t28saC">Here</a> is another example showing P1144’s interaction with fancy allocators.)</p>
    <p>Since 2023, <a href="https://github.com/Quuxplusone/gcc/tree/trivially-relocatable">Arthur maintains a fork of libstdc++</a> with a conforming implementation of P1144, although it omits some optimizations. For example, <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>relocate</c-></code> never
-lowers to <code class="highlight"><c- n>memcpy</c-></code>, and (as of December 2023) <code class="highlight"><c- n>vector</c-></code> reallocation never relocates. See it in action <a href="https://p1144.godbolt.org/z/bn4qsj48G">on godbolt.org</a>, under the name "x86-64 clang (experimental P1144),"
+lowers to <code class="highlight"><c- n>memcpy</c-></code>, and (as of December 2023) <code class="highlight"><c- n>vector</c-></code> reallocation never relocates. See it in action <a href="https://p1144.godbolt.org/z/zTvKPzxer">on godbolt.org</a>, under the name "x86-64 clang (experimental P1144),"
 by passing <code class="highlight"><c- o>-</c-><c- n>stdlib</c-><c- o>=</c-><c- n>libstdc</c-><c- o>++</c-></code> on the command line.</p>
    <p>Since October 2020, Carnegie Mellon’s <a data-link-type="biblio" href="#biblio-parlaylib" title="ParlayLib, a toolkit for programming parallel algorithms on shared-memory multicore machines">[ParlayLib]</a> implements <code class="highlight"><c- n>uninitialized_relocate</c-><c- p>{,</c-><c- n>_n</c-><c- p>}</c-></code> in the <code class="highlight"><c- n>parlay</c-></code> namespace.</p>
    <p>Since September 2023, Stellar <a data-link-type="biblio" href="#biblio-hpx" title="Relocation Semantics in the HPX Library">[HPX]</a> implements <code class="highlight"><c- n>relocate_at</c-></code> and <code class="highlight"><c- n>uninitialized_relocate</c-><c- p>{,</c-><c- n>_n</c-><c- p>}</c-></code> in the <code class="highlight"><c- n>hpx</c-><c- o>::</c-><c- n>experimental</c-></code> namespace. This was a GSoC project.</p>
@@ -3177,6 +3223,8 @@ <h3 class="heading settled" id="implementation-lib"><span class="content">Librar
   <h2 class="no-num no-ref heading settled" id="references"><span class="content">References</span><a class="self-link" href="#references"></a></h2>
   <h3 class="no-num no-ref heading settled" id="informative"><span class="content">Informative References</span><a class="self-link" href="#informative"></a></h3>
   <dl>
+   <dt id="biblio-amc">[AMC]
+   <dd>Stephane Janel. <a href="https://github.com/AmadeusITGroup/amc"><cite>AMadeus (C++) Containers</cite></a>. April 2021. URL: <a href="https://github.com/AmadeusITGroup/amc">https://github.com/AmadeusITGroup/amc</a>
    <dt id="biblio-announcing">[Announcing]
    <dd>Arthur O'Dwyer. <a href="https://quuxplusone.github.io/blog/2018/07/18/announcing-trivially-relocatable/"><cite>Announcing "trivially relocatable"</cite></a>. July 2018. URL: <a href="https://quuxplusone.github.io/blog/2018/07/18/announcing-trivially-relocatable/">https://quuxplusone.github.io/blog/2018/07/18/announcing-trivially-relocatable/</a>
    <dt id="biblio-bench">[Bench]
@@ -3197,8 +3245,6 @@ <h3 class="no-num no-ref heading settled" id="informative"><span class="content"
    <dd>Arthur O'Dwyer. <a href="https://quuxplusone.github.io/blog/2019/02/22/p1144-fixed-capacity-vector/"><cite>P1144 case study: Moving a `fixed_capacity_vector`</cite></a>. URL: <a href="https://quuxplusone.github.io/blog/2019/02/22/p1144-fixed-capacity-vector/">https://quuxplusone.github.io/blog/2019/02/22/p1144-fixed-capacity-vector/</a>
    <dt id="biblio-folly">[Folly]
    <dd>Facebook. <a href="https://github.com/facebook/folly/blob/master/folly/docs/FBVector.md#object-relocation"><cite>Folly documentation on "Object Relocation"</cite></a>. URL: <a href="https://github.com/facebook/folly/blob/master/folly/docs/FBVector.md#object-relocation">https://github.com/facebook/folly/blob/master/folly/docs/FBVector.md#object-relocation</a>
-   <dt id="biblio-follyissue889">[FollyIssue889]
-   <dd>Arthur O'Dwyer. <a href="https://github.com/facebook/folly/issues/889"><cite>Traits.h marks std::list as trivially relocatable, but in fact it is not</cite></a>. URL: <a href="https://github.com/facebook/folly/issues/889">https://github.com/facebook/folly/issues/889</a>
    <dt id="biblio-hpx">[HPX]
    <dd>Isidoros Tsaousis-Seiras. <a href="https://isidorostsa.github.io/gsoc2023/"><cite>Relocation Semantics in the HPX Library</cite></a>. August 2023. URL: <a href="https://isidorostsa.github.io/gsoc2023/">https://isidorostsa.github.io/gsoc2023/</a>
    <dt id="biblio-inpractice">[InPractice]