Planet
navi homePPSaboutscreenshotsdownloaddevelopmentforum

source: downloads/boost_1_34_1/libs/format/doc/choices.html @ 29

Last change on this file since 29 was 29, checked in by landauf, 16 years ago

updated boost from 1_33_1 to 1_34_1
File size: 13.2 KB
Line 
1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
2
3<html>
4<head>
5  <meta http-equiv="Content-Language" content="en-us">
6  <meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
7
8  <title>Type-safe 'printf-like' format class</title>
9</head>
10
11<body bgcolor="#FFFFFF" text="#000000">
12  <h1><img align="middle" alt="boost.png (6897 bytes)" height="86" src=
13  "../../../boost.png" width="277">Type-safe 'printf-like' <b>format
14  class</b></h1>
15
16  <h2>Choices made</h2>
17
18  <p>"Le pourquoi du comment" ( - "the why of the how")</p>
19  <hr>
20
21  <h3>The syntax of the format-string</h3>
22
23  <p>Format is a new library. One of its goal is to provide a replacement for
24  printf, that means format can parse a format-string designed for printf,
25  apply it to the given arguments, and produce the same result as printf
26  would have.<br>
27  With this constraint, there were roughly 3 possible choices for the syntax
28  of the format-string :</p>
29
30  <ol>
31    <li>Use the exact same syntax of printf. It's well known by many
32    experienced users, and fits almost all needs. But with C++ streams, the
33    type-conversion character, crucial to determine the end of a directive,
34    is only useful to set some associated formatting options, in a C++
35    streams context (%x for setting hexa, etc..) It would be better to make
36    this obligatory type-conversion character, with modified meaning,
37    optional.</li>
38
39    <li>extend printf syntax while maintaining compatibility, by using
40    characters and constructs not yet valid as printf syntax. e.g. : "%1%",
41    "%[1]", "%|1$d|", .. Using begin / end marks, all sort of extension can
42    be considered.</li>
43
44    <li>Provide a non-legacy mode, in parallel of the printf-compatible one,
45    that can be designed to fit other objectives without constraints of
46    compatibilty with the existing printf syntax.<br>
47    But Designing a replacement to printf's syntax, that would be clearly
48    better, and as much powerful, is yet another task than building a format
49    class. When such a syntax is designed, we should consider splitting
50    Boost.format into 2 separate libraries : one working hand in hand with
51    this new syntax, and another supporting the legacy syntax (possibly a
52    fast version, built with safety improvement above snprintf or the
53    like).</li>
54  </ol>In the absence of a full, clever, new syntax clearly better adapted to
55  C++ streams than printf, the second approach was chosen. Boost.format uses
56  printf's syntax, with extensions (tabulations, centered alignements) that
57  can be expressed using extensions to this syntax.<br>
58  And alternate compatible notations are provided to address the weaknesses
59  of printf's :
60
61  <ul>
62    <li><i>"%<b>N</b>%"</i> as a simpler positional, typeless and optionless
63    notation.</li>
64
65    <li><i>%|spec|</i> as a way to encapsulate printf directive in movre
66    visually evident structures, at the same time making printf's
67    'type-conversion character' optional.</li>
68  </ul>
69  <hr>
70
71  <h3>Why are arguments passed through an operator rather than a function
72  call ?</h3><br>
73  The inconvenience of the operator approach (for some people) is that it
74  might be confusing. It's a usual warning that too much of overloading
75  operators gets people real confused.<br>
76  Since the use of format objects will be in specific contexts ( most often
77  right after a "cout &lt;&lt; ") and look like a formatting string followed
78  by arguments indeed :
79
80  <blockquote>
81    <pre>
82format(" %s at %s  with %s\n") % x % y % z;
83</pre>
84  </blockquote>we can hope it wont confuse people that much.
85
86  <p>An other fear about operators, is precedence problems. What if I someday
87  write <b>format("%s") % x+y</b><br>
88  instead of <i>format("%s") % (x+y)</i> ??<br>
89  It will make a mistake at compile-time, so the error will be immediately
90  detected.<br>
91  indeed, this line calls <i>tmp = operator%( format("%s"), x)</i><br>
92  and then <i>operator+(tmp, y)</i><br>
93  tmp will be a format object, for which no implicit conversion is defined,
94  and thus the call to operator+ will fail. (except if you define such an
95  operator, of course). So you can safely assume precedence mistakes will be
96  noticed at compilation.</p>
97
98  <p><br>
99  On the other hand, the function approach has a true inconvenience. It needs
100  to define lots of template function like :</p>
101
102  <blockquote>
103    <pre>
104template &lt;class T1, class T2,  .., class TN&gt; 
105string format(string s,  const T1&amp; x1, .... , const T1&amp; xN);
106
107</pre>
108  </blockquote>and even if we define those for N up to 500, that is still a
109  limitation, that C's printf does not have.<br>
110  Also, since format somehow emulates printf in some cases, but is far from
111  being fully equivalent to printf, it's best to use a radically different
112  appearance, and using operator calls succeeds very well in that !
113
114  <p><br>
115  Anyhow, if we actually chose the formal function call templates system, it
116  would only be able to print Classes T for which there is an</p>
117
118  <blockquote>
119    <pre>
120operator&lt;&lt; ( stream,   const T&amp;)
121</pre>
122  </blockquote>Because allowing both const and non const produces a
123  combinatorics explosion - if we go up to 10 arguments, we need 2^10
124  functions.<br>
125  (providing overloads on T&amp; / const T&amp; is at the frontier of defects
126  of the C++ standard, and thus is far from guaranteed to be supported. But
127  right now several compilers support those overloads)<br>
128  There is a lot of chances that a class which only provides the non-const
129  equivalent is badly designed, but yet it is another unjustified restriction
130  to the user.<br>
131  Also, some manipulators are functions, and can not be passed as const
132  references. The function call approach thus does not support manipulators
133  well.
134
135  <p>In conclusion, using a dedicated binary operator is the simplest, most
136  robust, and least restrictive mechanism to pass arguments when you can't
137  know the number of arguments at compile-time.</p>
138  <hr>
139
140  <h3>Why operator% rather than a member function 'with(..)'
141  ??</h3>technically,
142
143  <blockquote>
144    <pre>
145format(fstr) % x1 % x2 % x3;
146</pre>
147  </blockquote>has the same structure as
148
149  <blockquote>
150    <pre>
151format(fstr).with( x1 ).with( x2 ).with( x3 );
152</pre>
153  </blockquote>which does not have any precedence problem. The only drawback,
154  is it's harder for the eye to catch what is done in this line, than when we
155  are using operators. calling .with(..), it looks just like any other line
156  of code. So it may be a better solution, depending on tastes. The extra
157  characters, and overall cluttered aspect of the line of code using
158  'with(..)' were enough for me to opt for a true operator.
159  <hr>
160
161  <h3>Why operator% rather than usual formatting operator&lt;&lt; ??</h3>
162
163  <ul>
164    <li>because passing arguments to a format object is *not* the same as
165    sending variables, sequentially, into a stream, and because a format
166    object is not a stream, nor a manipulator.<br>
167      We use an operator to pass arguments. format will use them as a
168      function would, it simply takes arguments one by one.<br>
169      format objects can not provide stream-like behaviour. When you try to
170      implement a format object that acts like a manipulator, returning a
171      stream, you make the user beleive it is completely like a
172      stream-manipulator. And sooner or later, the user is deceived by this
173      point of view.<br>
174      The most obvious example of that difference in behaviour is
175
176      <blockquote>
177        <pre>
178cout &lt;&lt; format("%s %s ") &lt;&lt; x;
179cout &lt;&lt; y ;  // uh-oh, format is not really a stream manipulator
180</pre>
181      </blockquote>
182    </li>
183
184    <li>precedence of % is higher than that of &lt;&lt;. It can be viewd as a
185    problem, because + and - thus needs to be grouped inside parentheses,
186    while it is not necessary with '&lt;&lt;'. But if the user forgets, the
187    mistake is catched at compilation, and hopefully he won't forget
188    again.<br>
189      On the other hand, the higher precedence makes format's behaviour very
190      straight-forward.
191
192      <blockquote>
193        <pre>
194cout &lt;&lt; format("%s %s ") % x % y &lt;&lt; endl;
195</pre>
196      </blockquote>is treated exaclt like :
197
198      <blockquote>
199        <pre>
200cout &lt;&lt; ( format("%s %s ") % x % y ) &lt;&lt; endl;
201</pre>
202      </blockquote>So using %, the life of a format object does not interfere
203      with the surrounding stream context. This is the simplest possible
204      behaviour, and thus the user is able to continue using the stream after
205      the format object.<br>
206      <br>
207      With operator&lt;&lt;, things are much more problematic in this
208      situation. This line :
209
210      <blockquote>
211        <pre>
212cout &lt;&lt; format("%s %s ") &lt;&lt;  x  &lt;&lt;  y &lt;&lt; endl;
213</pre>
214      </blockquote>is understood as :
215
216      <blockquote>
217        <pre>
218( ( ( cout &lt;&lt; format("%s %s ") ) &lt;&lt; x ) &lt;&lt;  y ) &lt;&lt; endl;
219</pre>
220      </blockquote>Several alternative implementations chose
221      operator&lt;&lt;, and there is only one way to make it work :<br>
222      the first call to
223
224      <blockquote>
225        <pre>
226operator&lt;&lt;( ostream&amp;, format const&amp;)
227</pre>
228      </blockquote>returns a proxy, encapsulating both the final destination
229      (cout) and the format-string information<br>
230      Passing arguments to format, or to the final destination after
231      completion of the format are indistinguishable. This is a problem.
232
233      <p>I examined several possible implementations, and none is completely
234      satsifying.<br>
235      E.g. : In order to catch users mistake, it makes sense to raise
236      exceptions when the user passes too many arguments. But in this
237      context, supplementary arguments are most certainly aimed at the final
238      destination. There are several choices here :</p>
239
240      <ul>
241        <li>You can give-up detection of arity excess, and have the proxy's
242        template member operator&lt;&lt;( const T&amp;) simply forward all
243        supplementary arguments to cout.</li>
244
245        <li>Require the user to close the format arguments with a special
246        manipulator, 'endf', in this way :
247
248          <blockquote>
249            <pre>
250cout &lt;&lt; format("%s %s ") &lt;&lt;  x  &lt;&lt;  y &lt;&lt; endf &lt;&lt; endl;
251</pre>
252          </blockquote>You can define endf to be a function that returns the
253          final destination stored inside the proxy. Then it's okay, after
254          endf the user is calling &lt;&lt; on cout again.
255        </li>
256
257        <li>An intermediate solution, is to adress the most frequent use,
258        where the user simply wants to output one more manipulator item to
259        cout (a std::flush, or endl, ..)
260
261          <blockquote>
262            <pre>
263cout &lt;&lt; format("%s %s \n") &lt;&lt;  x  &lt;&lt;  y &lt;&lt; flush ;
264</pre>
265          </blockquote>Then, the solution is to overload the operator&lt;&lt;
266          for manipulators. This way You don't need endf, but outputting a
267          non-manipulator item right after the format arguments is a mistake.
268        </li>
269      </ul><br>
270      The most complete solution is the one with the endf manipualtor. With
271      operator%, there is no need for this end-format function, plus you
272      instantly see which arguments are going into the format object, and
273      which are going to the stream.
274    </li>
275
276    <li>Esthetically : '%' is the same letter as used inside the
277    format-string. That is quite nice to have the same letter used for
278    passing each argument. '&lt;&lt;' is 2 letters, '%' is one. '%' is also
279    smaller in size. It overall improves visualisation (we see what goes with
280    what) :
281
282      <blockquote>
283        <pre>
284cout &lt;&lt; format("%s %s %s") %x %y %z &lt;&lt; "And  avg is" &lt;&lt; format("%s\n") %avg;
285</pre>
286      </blockquote>compared to :
287
288      <blockquote>
289        <pre>
290cout &lt;&lt; format("%s %s %s") &lt;&lt; x &lt;&lt; y &lt;&lt; z &lt;&lt; endf &lt;&lt;"And avg is" &lt;&lt; format("%s\n") &lt;&lt; avg;
291</pre>
292      </blockquote>"&lt;&lt;" misleadingly puts the arguments at the same
293      level as any object passed to the stream.
294    </li>
295
296    <li>python also uses % for formatting, so you see it's not so "unheard
297    of" ;-)</li>
298  </ul>
299  <hr>
300
301  <h3>Why operator% rather than operator(), or operator[] ??</h3>
302
303  <p>operator() has the merit of being the natural way to send an argument
304  into a function. And some think that operator[] 's meaning apply well to
305  the usage in format.<br>
306  They're as good as operator% technically, but quite ugly. (that's a matter
307  of taste)<br>
308  And deepd down, using operator% for passing arguments that were referred to
309  by "%" in the format string seems much more natural to me than using those
310  operators.</p>
311  <hr>
312
313  <p><a href="http://validator.w3.org/check?uri=referer"><img border="0" src=
314  "http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Transitional"
315  height="31" width="88"></a></p>
316
317  <p>Revised
318  <!--webbot bot="Timestamp" s-type="EDITED" s-format="%d %B, %Y" startspan -->02 December, 2006<!--webbot bot="Timestamp" endspan i-checksum="38510" --></p>
319
320  <p><i>Copyright &copy; 2001 Samuel Krempp</i></p>
321
322  <p><i>Distributed under the Boost Software License, Version 1.0. (See
323  accompanying file <a href="../../../LICENSE_1_0.txt">LICENSE_1_0.txt</a> or
324  copy at <a href=
325  "http://www.boost.org/LICENSE_1_0.txt">http://www.boost.org/LICENSE_1_0.txt</a>)</i></p>
326</body>
327</html>
Note: See TracBrowser for help on using the repository browser.