New issue from Jiang An: "std::formatter specializations should be co…

…nsistently restricted to supported character types"

xml/issue3997.xml

@@ -0,0 +1,33 @@
+<?xml version='1.0' encoding='utf-8' standalone='no'?>
+<!DOCTYPE issue SYSTEM "lwg-issue.dtd">
+
+<issue num="3997" status="New">
+<title><tt>std::formatter</tt> specializations should be consistently restricted to supported character types</title>
+<section><sref ref="[format.syn]"/><sref ref="[time.syn]"/><sref ref="[thread.thread.id]"/></section>
+<submitter>Jiang An</submitter>
+<date>15 Oct 2023</date>
+<priority>99</priority>
+
+<discussion>
+<p>
+<sref ref="[format.formatter.spec]"/>/2 requires some <tt>std::formatter</tt> specializations are only enabled 
+for character types <tt>char</tt> and <tt>wchar_t</tt>. But for specializations that are explicitly shown in 
+the synopses (<sref ref="[format.syn]"/>, <sref ref="[time.syn]"/>, and <sref ref="[thread.thread.id]"/>), 
+there is no such restrictions for <tt>charT</tt>. The difference may be observable via the <tt>std::formattable</tt> 
+concept, e.g. <tt>std::formattable&lt;std::chrono::second, char8_t&gt;</tt> is currently satisfied, even though 
+there would be a hard error during formatting due to <tt><i>STATICALLY-WIDEN</i></tt>.
+<p/>
+Perhaps we should introduce an exposition-only concept that is only satisfied (and modeled) by <tt>char</tt> and 
+<tt>wchar_t</tt> for now, and constrain the <tt>charT</tt> template parameter of all <tt>std::formatter</tt> 
+specializations that are explicitly mentioned in synopses with that concept. This is what libc++ currently 
+implements (e.g. the 
+<a href="https://github.com/llvm/llvm-project/blob/6dfea561ba96974b205c31546c5e2069429c75b1/libcxx/include/__format/concepts.h#L32-L38"><tt>__fmt_char_type</tt></a> 
+concept and <a href="https://github.com/llvm/llvm-project/blob/main/libcxx/include/__chrono/formatter.h#L580-L825">its 
+usage in <tt>&lt;chrono&gt;</tt></a>).
+</p>
+</discussion>
+
+<resolution>
+</resolution>
+
+</issue>