Skip to content

Commit

Permalink
Update paper
Browse files Browse the repository at this point in the history
  • Loading branch information
@vitaut
vitaut committed Apr 20, 2024
1 parent d27beb5 commit 3003252
Show file tree
Hide file tree
Showing 2 changed files with 58 additions and 25 deletions.
44 changes: 31 additions & 13 deletions papers/p3235.bs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -28,8 +28,19 @@ important optimization to other standard types.
Proposal {#proposal}
========

<!-- TODO: motivate summarizing the benefits of P3107 and possibly mention that this
is a follow-up requested in LEWG discussion -->
[[P3107]] "Permit an efficient implementation of std::print" brought
significant speedups (more than 20% in some cases) to `std::print` and
eliminated the neeed for dynamic memory allocations in the common case by
enabling direct writes into the stream buffer. To speed up the adoption of the
fix, [[P3107]] limited the scope to fundamental and string types but it is, of
course, beneficial to enable this optimization for other standard types that
have formatters. This was even discussed in LEWG that encouraged writing
another paper:

> LEWG requests for an additional paper to fix formatters for ranges

The current paper proposes opting in formatters for ranges and other standard
types into this optimization.

Here is a list of standard formatters that are not yet opted into the
`std::print` optimization.
Expand Down Expand Up @@ -78,14 +89,14 @@ Date and time formatters [[time.syn](https://eel.is/c++draft/time.syn)]:

`Rep` is an arithmetic type, `Period` is `std::ratio<...>`, `Duration` is
`std::duration<...>` and `charT` is `char` or `wchar_t` so all chrono formatters
except the one for `zoned_time` can be unconditionally opted into the
optimization. The formatter for `zoned_time` can be opted in for the default
except the one for `std::zoned_time` can be unconditionally opted into the
optimization. The formatter for `std::zoned_time` can be opted in for the default
`TimeZonePtr` (`const std::chrono::time_zone*`) but not arbitrary user-provided
`TimeZonePtr` that can be potentially locking.

<!-- https://www.godbolt.org/z/v87aY37GG -->

`thread::id` formatter [[thread.thread.id](
`std::thread::id` formatter [[thread.thread.id](
https://eel.is/c++draft/thread.thread.id)]:

```
Expand All @@ -101,7 +112,7 @@ https://eel.is/c++draft/stacktrace.syn)]:
template<class Allocator> struct formatter<basic_stacktrace<Allocator>>;
```

`vector<bool>` formatter [[vector.syn](
`std::vector<bool>` formatter [[vector.syn](
https://eel.is/c++draft/vector.syn)]:

```
Expand All @@ -118,8 +129,8 @@ the process of being merged into the standard draft:
template<class charT> struct formatter<filesystem::path, charT>;
```

`thread::id`, stacktrace, `vector<bool>` and `std::filesystem::path` formatters
don't invoke any user code and can be opted into the optimization.
`std::thread::id`, stacktrace, `std::vector<bool>` and `std::filesystem::path`
formatters don't invoke any user code and can be opted into the optimization.

Tuple formatter [[format.tuple](https://eel.is/c++draft/format.tuple)]:

Expand All @@ -142,7 +153,7 @@ Range formatter [[format.syn](https://eel.is/c++draft/format.syn)]:
struct formatter<R, charT> : range-default-formatter<format_kind<R>, R, charT> { };
```

`queue` and `priority_queue` formatters [[queue.syn](
`std::queue` and `std::priority_queue` formatters [[queue.syn](
https://eel.is/c++draft/queue.syn)]:

```
Expand All @@ -157,7 +168,7 @@ https://eel.is/c++draft/queue.syn)]:
struct formatter<priority_queue<T, Container, Compare>, charT>;
```

`stack` formatter [[stack.syn](https://eel.is/c++draft/stack.syn)]:
`std::stack` formatter [[stack.syn](https://eel.is/c++draft/stack.syn)]:

```
// [container.adaptors.format], formatter specialization for stack
Expand All @@ -166,10 +177,13 @@ https://eel.is/c++draft/queue.syn)]:
```

Range and container adaptor formatters are the most interesting case because
formatting requires iterating and user-defined iterators can potentially be
locking.
formatting requires iterating and user-defined iterators can be locking, at
least in principle. None of the standard containers, ranges and container
adaptors and even common concurrent containers such as `concurrent_vector` from
[[TBB]] provide locking iterators. For this reason, the current paper proposes
opting range and adaptor formatters into the optimization by default.

TODO: check existing concurrent containers
<!-- TODO: wording -->

<pre class=biblio>
{
Expand All @@ -182,6 +196,10 @@ TODO: check existing concurrent containers
"title": "Permit an efficient implementation of std::print",
"authors": ["Victor Zverovich"],
"href": "https://wg21.link/p3107"
},
"TBB": {
"title": "oneAPI Threading Building Blocks (oneTBB)",
"href": "https://oneapi-src.github.io/oneTBB/"
}
}
</pre>
39 changes: 27 additions & 12 deletions papers/p3235.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1563,7 +1563,7 @@
</style>
<meta content="Bikeshed version 4416b18d5, updated Tue Jan 2 15:52:39 2024 -0800" name="generator">
<link href="https://isocpp.org/favicon.ico" rel="icon">
<meta content="f4d944279c43baba24123058f6c290cfaa3810a0" name="revision">
<meta content="36b6b51283da017ef5047ba63cc725bb8ef30418" name="revision">
<style>/* Boilerplate: style-autolinks */
.css.css, .property.property, .descriptor.descriptor {
color: var(--a-normal-text);
Expand Down Expand Up @@ -2122,6 +2122,19 @@ <h2 class="heading settled" data-level="1" id="intro"><span class="secno">1. </s
optimization to fundamental and string types. The current paper applies this
important optimization to other standard types.</p>
<h2 class="heading settled" data-level="2" id="proposal"><span class="secno">2. </span><span class="content">Proposal</span><a class="self-link" href="#proposal"></a></h2>
<p><a data-link-type="biblio" href="#biblio-p3107" title="Permit an efficient implementation of std::print">[P3107]</a> "Permit an efficient implementation of std::print" brought
significant speedups (more than 20% in some cases) to <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>print</c-></code> and
eliminated the neeed for dynamic memory allocations in the common case by
enabling direct writes into the stream buffer. To speed up the adoption of the
fix, <a data-link-type="biblio" href="#biblio-p3107" title="Permit an efficient implementation of std::print">[P3107]</a> limited the scope to fundamental and string types but it is, of
course, beneficial to enable this optimization for other standard types that
have formatters. This was even discussed in LEWG that encouraged writing
another paper:</p>
<blockquote>
<p>LEWG requests for an additional paper to fix formatters for ranges</p>
</blockquote>
<p>The current paper proposes opting in formatters for ranges and other standard
types into this optimization.</p>
<p>Here is a list of standard formatters that are not yet opted into the <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>print</c-></code> optimization.</p>
<p>Date and time formatters <a href="https://eel.is/c++draft/time.syn">[time.syn</a>]:</p>
<pre class="highlight"><c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>Rep</c-><c- p>,</c-> <c- k>class</c-> <c- nc>Period</c-><c- p>,</c-> <c- k>class</c-> <c- nc>charT</c-><c- o>></c->
Expand Down Expand Up @@ -2163,17 +2176,17 @@ <h2 class="heading settled" data-level="2" id="proposal"><span class="secno">2.
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>chrono</c-><c- o>::</c-><c- n>zoned_time</c-><c- o>&lt;</c-><c- n>Duration</c-><c- p>,</c-> <c- n>TimeZonePtr</c-><c- o>></c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
</pre>
<p><code class="highlight"><c- n>Rep</c-></code> is an arithmetic type, <code class="highlight"><c- n>Period</c-></code> is <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>ratio</c-><c- o>&lt;</c-><c- p>...</c-><c- o>></c-></code>, <code class="highlight"><c- n>Duration</c-></code> is <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>duration</c-><c- o>&lt;</c-><c- p>...</c-><c- o>></c-></code> and <code class="highlight"><c- n>charT</c-></code> is <code class="highlight"><c- b>char</c-></code> or <code class="highlight"><c- b>wchar_t</c-></code> so all chrono formatters
except the one for <code class="highlight"><c- n>zoned_time</c-></code> can be unconditionally opted into the
optimization. The formatter for <code class="highlight"><c- n>zoned_time</c-></code> can be opted in for the default <code class="highlight"><c- n>TimeZonePtr</c-></code> (<code class="highlight"><c- k>const</c-> <c- n>std</c-><c- o>::</c-><c- n>chrono</c-><c- o>::</c-><c- n>time_zone</c-><c- o>*</c-></code>) but not arbitrary user-provided <code class="highlight"><c- n>TimeZonePtr</c-></code> that can be potentially locking.</p>
<p><code class="highlight"><c- kr>thread</c-><c- o>::</c-><c- n>id</c-></code> formatter <a href="https://eel.is/c++draft/thread.thread.id">[thread.thread.id</a>]:</p>
except the one for <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>zoned_time</c-></code> can be unconditionally opted into the
optimization. The formatter for <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>zoned_time</c-></code> can be opted in for the default <code class="highlight"><c- n>TimeZonePtr</c-></code> (<code class="highlight"><c- k>const</c-> <c- n>std</c-><c- o>::</c-><c- n>chrono</c-><c- o>::</c-><c- n>time_zone</c-><c- o>*</c-></code>) but not arbitrary user-provided <code class="highlight"><c- n>TimeZonePtr</c-></code> that can be potentially locking.</p>
<p><code class="highlight"><c- n>std</c-><c- o>::</c-><c- kr>thread</c-><c- o>::</c-><c- n>id</c-></code> formatter <a href="https://eel.is/c++draft/thread.thread.id">[thread.thread.id</a>]:</p>
<pre class="highlight"><c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>charT</c-><c- o>></c-> <c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- kr>thread</c-><c- o>::</c-><c- n>id</c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
</pre>
<p>Stacktrace formatters <a href="https://eel.is/c++draft/stacktrace.syn">[stacktrace.syn</a>]:</p>
<pre class="highlight"><c- c1>// [stacktrace.format], formatting support</c->
<c- k>template</c-><c- o>&lt;></c-> <c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>stacktrace_entry</c-><c- o>></c-><c- p>;</c->
<c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>Allocator</c-><c- o>></c-> <c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>basic_stacktrace</c-><c- o>&lt;</c-><c- n>Allocator</c-><c- o>>></c-><c- p>;</c->
</pre>
<p><code class="highlight"><c- n>vector</c-><c- o>&lt;</c-><c- b>bool</c-><c- o>></c-></code> formatter <a href="https://eel.is/c++draft/vector.syn">[vector.syn</a>]:</p>
<p><code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>vector</c-><c- o>&lt;</c-><c- b>bool</c-><c- o>></c-></code> formatter <a href="https://eel.is/c++draft/vector.syn">[vector.syn</a>]:</p>
<pre class="highlight"><c- c1>// [vector.bool.fmt], formatter specialization for vector&lt;bool></c->
<c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>T</c-><c- p>,</c-> <c- k>class</c-> <c- nc>charT</c-><c- o>></c-> <c- k>requires</c-> <c- n>is</c-><c- o>-</c-><c- n>vector</c-><c- o>-</c-><c- b>bool</c-><c- o>-</c-><c- n>reference</c-><c- o>&lt;</c-><c- n>T</c-><c- o>></c->
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
Expand All @@ -2183,8 +2196,7 @@ <h2 class="heading settled" data-level="2" id="proposal"><span class="secno">2.
<pre class="highlight"><c- c1>// [fs.path.fmt], formatter</c->
<c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>charT</c-><c- o>></c-> <c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>filesystem</c-><c- o>::</c-><c- n>path</c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
</pre>
<p><code class="highlight"><c- kr>thread</c-><c- o>::</c-><c- n>id</c-></code>, stacktrace, <code class="highlight"><c- n>vector</c-><c- o>&lt;</c-><c- b>bool</c-><c- o>></c-></code> and <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>filesystem</c-><c- o>::</c-><c- n>path</c-></code> formatters
don’t invoke any user code and can be opted into the optimization.</p>
<p><code class="highlight"><c- n>std</c-><c- o>::</c-><c- kr>thread</c-><c- o>::</c-><c- n>id</c-></code>, stacktrace, <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>vector</c-><c- o>&lt;</c-><c- b>bool</c-><c- o>></c-></code> and <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>filesystem</c-><c- o>::</c-><c- n>path</c-></code> formatters don’t invoke any user code and can be opted into the optimization.</p>
<p>Tuple formatter <a href="https://eel.is/c++draft/format.tuple">[format.tuple</a>]:</p>
<pre class="highlight"><c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>charT</c-><c- p>,</c-> <c- n>formattable</c-><c- o>&lt;</c-><c- n>charT</c-><c- o>></c-><c- p>...</c-> <c- n>Ts</c-><c- o>></c->
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>pair</c-><c- o>-</c-><c- k>or</c-><c- o>-</c-><c- n>tuple</c-><c- o>&lt;</c-><c- n>Ts</c-><c- p>...</c-><c- o>></c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-> <c- p>{</c->
Expand All @@ -2199,7 +2211,7 @@ <h2 class="heading settled" data-level="2" id="proposal"><span class="secno">2.
<c- n>formattable</c-><c- o>&lt;</c-><c- n>ranges</c-><c- o>::</c-><c- n>range_reference_t</c-><c- o>&lt;</c-><c- n>R</c-><c- o>></c-><c- p>,</c-> <c- n>charT</c-><c- o>></c->
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>R</c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-> <c- o>:</c-> <c- n>range</c-><c- o>-</c-><c- k>default</c-><c- o>-</c-><c- n>formatter</c-><c- o>&lt;</c-><c- n>format_kind</c-><c- o>&lt;</c-><c- n>R</c-><c- o>></c-><c- p>,</c-> <c- n>R</c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-> <c- p>{</c-> <c- p>};</c->
</pre>
<p><code class="highlight"><c- n>queue</c-></code> and <code class="highlight"><c- n>priority_queue</c-></code> formatters <a href="https://eel.is/c++draft/queue.syn">[queue.syn</a>]:</p>
<p><code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>queue</c-></code> and <code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>priority_queue</c-></code> formatters <a href="https://eel.is/c++draft/queue.syn">[queue.syn</a>]:</p>
<pre class="highlight"><c- c1>// [container.adaptors.format], formatter specialization for queue</c->
<c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>charT</c-><c- p>,</c-> <c- k>class</c-> <c- nc>T</c-><c- p>,</c-> <c- n>formattable</c-><c- o>&lt;</c-><c- n>charT</c-><c- o>></c-> <c- n>Container</c-><c- o>></c->
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>queue</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>Container</c-><c- o>></c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
Expand All @@ -2210,15 +2222,16 @@ <h2 class="heading settled" data-level="2" id="proposal"><span class="secno">2.
<c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>charT</c-><c- p>,</c-> <c- k>class</c-> <c- nc>T</c-><c- p>,</c-> <c- n>formattable</c-><c- o>&lt;</c-><c- n>charT</c-><c- o>></c-> <c- n>Container</c-><c- p>,</c-> <c- k>class</c-> <c- nc>Compare</c-><c- o>></c->
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>priority_queue</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>Container</c-><c- p>,</c-> <c- n>Compare</c-><c- o>></c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
</pre>
<p><code class="highlight"><c- n>stack</c-></code> formatter <a href="https://eel.is/c++draft/stack.syn">[stack.syn</a>]:</p>
<p><code class="highlight"><c- n>std</c-><c- o>::</c-><c- n>stack</c-></code> formatter <a href="https://eel.is/c++draft/stack.syn">[stack.syn</a>]:</p>
<pre class="highlight"><c- c1>// [container.adaptors.format], formatter specialization for stack</c->
<c- k>template</c-><c- o>&lt;</c-><c- k>class</c-> <c- nc>charT</c-><c- p>,</c-> <c- k>class</c-> <c- nc>T</c-><c- p>,</c-> <c- n>formattable</c-><c- o>&lt;</c-><c- n>charT</c-><c- o>></c-> <c- n>Container</c-><c- o>></c->
<c- k>struct</c-> <c- nc>formatter</c-><c- o>&lt;</c-><c- n>stack</c-><c- o>&lt;</c-><c- n>T</c-><c- p>,</c-> <c- n>Container</c-><c- o>></c-><c- p>,</c-> <c- n>charT</c-><c- o>></c-><c- p>;</c->
</pre>
<p>Range and container adaptor formatters are the most interesting case because
formatting requires iterating and user-defined iterators can potentially be
locking.</p>
<p>TODO: check existing concurrent containers</p>
formatting requires iterating and user-defined iterators can be locking, at
least in principle. None of the standard containers, ranges and container
adaptors and even common concurrent containers such as <code class="highlight"><c- n>concurrent_vector</c-></code> from <a data-link-type="biblio" href="#biblio-tbb" title="oneAPI Threading Building Blocks (oneTBB)">[TBB]</a> provide locking iterators. For this reason, the current paper proposes
opting range and adaptor formatters into the optimization by default.</p>
</main>
<script>
(function() {
Expand Down Expand Up @@ -2356,4 +2369,6 @@ <h3 class="no-num no-ref heading settled" id="informative"><span class="content"
<dd>Victor Zverovich. <a href="https://wg21.link/p2845"><cite>Formatting of std::filesystem::path</cite></a>. URL: <a href="https://wg21.link/p2845">https://wg21.link/p2845</a>
<dt id="biblio-p3107">[P3107]
<dd>Victor Zverovich. <a href="https://wg21.link/p3107"><cite>Permit an efficient implementation of std::print</cite></a>. URL: <a href="https://wg21.link/p3107">https://wg21.link/p3107</a>
<dt id="biblio-tbb">[TBB]
<dd><a href="https://oneapi-src.github.io/oneTBB/"><cite>oneAPI Threading Building Blocks (oneTBB)</cite></a>. URL: <a href="https://oneapi-src.github.io/oneTBB/">https://oneapi-src.github.io/oneTBB/</a>
</dl>

0 comments on commit 3003252

Please sign in to comment.