|
| 1 | +<?xml version='1.0' encoding='utf-8' standalone='no'?> |
| 2 | +<!DOCTYPE issue SYSTEM "lwg-issue.dtd"> |
| 3 | + |
| 4 | +<issue num="4216" status="New"> |
| 5 | +<title>`num_put::do_put` and `void` pointers</title> |
| 6 | +<section> |
| 7 | +<sref ref="[facet.num.put.virtuals]"/> |
| 8 | +</section> |
| 9 | +<submitter>Nikolas Klauser</submitter> |
| 10 | +<date>26 Feb 2025</date> |
| 11 | +<priority>99</priority> |
| 12 | + |
| 13 | +<discussion> |
| 14 | +<p> |
| 15 | +The `num_put::do_put` overloads are defined in terms of `printf`. |
| 16 | +However, it is not clear what the intended behaviour of pointers is. |
| 17 | +<p/> |
| 18 | +While the `num_put` wording makes a quite clear statement that it should |
| 19 | +be whatever `printf("%p", ptr)` would be, `num_get` is entirely silent |
| 20 | +on which function should be used. This makes it entirely unclear whether |
| 21 | +round-tripping is supposed to work. It's also not clear whether |
| 22 | +`num_put` was just simple to specify via `printf` or whether the intent |
| 23 | +was that the output matches in all cases. Round-tripping between |
| 24 | +`num_put` and `num_get` was broken in libc++ until recently. However, to |
| 25 | +fix that, the output of `num_put::do_put` no longer matches the libc's |
| 26 | +`printf` in all cases. libstdc++ had this behaviour since at least two |
| 27 | +decades, indicating that nobody seems to have a problem with |
| 28 | +`num_put::do_put` and `printf` having different results in some rare cases. |
| 29 | +</p> |
| 30 | +</discussion> |
| 31 | + |
| 32 | +<resolution> |
| 33 | +<p> |
| 34 | +This wording is relative to <paper num="N5001"/>. |
| 35 | +</p> |
| 36 | + |
| 37 | +<ol> |
| 38 | +<li><p>Modify <sref ref="[facet.num.put.virtuals]"/> as indicated:</p> |
| 39 | + |
| 40 | +<blockquote> |
| 41 | +<pre> |
| 42 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, long val) const; |
| 43 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, long long val) const; |
| 44 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, unsigned long val) const; |
| 45 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, unsigned long long val) const; |
| 46 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, double val) const; |
| 47 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, long double val) const; |
| 48 | +iter_type do_put(iter_type out, ios_base& str, char_type fill, const void* val) const; |
| 49 | +</pre> |
| 50 | +<blockquote> |
| 51 | +<p> |
| 52 | +-1- <i>Effects</i>: […] |
| 53 | +<p/> |
| 54 | +-2- The details of this operation occur in several stages: […] |
| 55 | +<p/> |
| 56 | +-3- Detailed descriptions of each stage follow. |
| 57 | +<p/> |
| 58 | +-4- <i>Returns</i>: <tt>out</tt>. |
| 59 | +<p/> |
| 60 | +<b>Stage 1</b>: The first action of stage 1 is to determine a conversion specifier. […] |
| 61 | +<p/> |
| 62 | +For conversion from `void*` the specifier is `%p`. |
| 63 | +<p/> |
| 64 | +The representations at the end of stage 1 consists of the `char`'s that would be printed by a call of |
| 65 | +`printf(s, val)` where `s` is the conversion specifier determined above<ins>, except that any |
| 66 | +implementation-defined behavior of `printf` may be different from a call to `printf`</ins>. |
| 67 | +</p> |
| 68 | +</blockquote> |
| 69 | +</blockquote> |
| 70 | +</li> |
| 71 | + |
| 72 | +</ol> |
| 73 | + |
| 74 | +</resolution> |
| 75 | + |
| 76 | +</issue> |
0 commit comments