-
Notifications
You must be signed in to change notification settings - Fork 0
/
p3311r0.html
543 lines (473 loc) · 19.5 KB
/
p3311r0.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
<!doctype html public "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
<meta http-equiv="Content-Type" content="text/html;charset=UTF-8">
<head>
<title>
An opt-in approach for integration of traditional assert facilities in C++ contracts
</title>
<style type="text/css">
pre {
display: inline;
}
table#header th,
table#header td
{
text-align: left;
}
table#references th,
table#references td
{
vertical-align: top;
}
#hideins:checked ~ * ins, #hideins:checked ~ * ins * { display:none; visibility:hidden }
#hidedel:checked ~ * del, #hidedel:checked ~ * del * { display:none; visibility:hidden }
ins, ins *
{
text-decoration: underline;
color: #000000;
background-color:#C8FFC8
}
del, del *
{
text-decoration: line-through;
color: #000000;
background-color:#FFA0A0
}
nop, nop *
{
color: #000000;
background-color:#B0B0FF
}
blockquote
{
color: #000000;
background-color: #F1F1F1;
border: 1px solid #D1D1D1;
padding-left: 0.5em;
padding-right: 0.5em;
}
blockquote.code
{
white-space: pre;
font-family: monospace;
}
blockquote.stdins
{
/* text-decoration: underline; */
color: #000000;
background-color: #C8FFC8;
border: 1px solid #B3EBB3;
padding: 0.5em;
}
blockquote.stddel
{
text-decoration: line-through;
color: #000000;
background-color: #FFA0A0;
border: 1px solid #ECD7EC;
padding-left: 0.5empadding-right: 0.5em;
}
blockquote.stdnop
{
color: #000000;
background-color: #B0B0FF;
border: 1px solid #ECD7EC;
padding-left: 0.5empadding-right: 0.5em;
}
</style>
</head>
<body style="max-width: 8.5in">
<table id="header">
<tr>
<th>Document Number:</th>
<td>P3311R0</td>
</tr>
<tr>
<th>Date:</th>
<td>2024-05-22</td>
</tr>
<tr>
<th>Audience:</th>
<td>SG21</td>
</tr>
<tr>
<th>Reply-to:</th>
<td>Tom Honermann <[email protected]></td>
</tr>
</table>
<h1>An opt-in approach for integration of traditional assert facilities in C++ contracts</h1>
<ul>
<li><a href="#introduction">Introduction</a></li>
<li><a href="#motivation">Motivation</a></li>
<li><a href="#design">Design considerations</a>
<ul>
<li><a href="#design-handler">Invoking the contract violation handler</a></li>
<li><a href="#design-ndebug">Interaction with <tt>NDEBUG</tt></a></li>
</ul>
</li>
<li><a href="#proposal">Proposal</a>
<ul>
<li><a href="#proposal-traditional-semantics">Extend traditional semantics
to <tt>contract_assert</tt></a></li>
<li><a href="#proposal-assert-opt-in">Enable optional C++ contracts
integration for C assert</a></li>
</ul>
</li>
<li><a href="#implementation-exp">Implementation experience</a></li>
<li><a href="#ack">Acknowledgements</a></li>
<li><a href="#references">References</a></li>
<li><a href="#wording">Wording</a></li>
</ul>
<h1 id="introduction">Introduction</h1>
<p>
SG21 attained strong concensus to forward
<a title="Contracts for C++"
href="https://wg21.link/p2900r6">
P2900R6 (Contracts for C++)</a>
<sup><a title="Contracts for C++"
href="#ref_p2900r6">[P2900R6]</a></sup>
to EWG during the
<a href="https://wiki.edg.com/bin/view/Wg21telecons2024/Teleconference2024-02-29">2024-02-29 SG21 telecon</a>
as its design for a minimally viable solution for contracts in C++.
This design does not integrate the traditional C assert facility, nor does it
expose interfaces to enable other traditional or ad hoc assert facilities to
integrate with the new contracts features.
</p>
<p>
The P2900R6 design includes a new <tt>contract_assert</tt> statement that avoids
many of the well-known pitfalls inherent in macro based traditional assert
facilities.
Despite considerable desire and effort to enable use of the <tt>assert</tt>
identifier in the contracts design, no method to do so was identified that both
avoided backward compatibility problems and used a syntax that gained
consensus.
<a title="assert Should Be A Keyword In C++26"
href="https://wg21.link/p2884r0">
P2884R0 (<tt>assert</tt> Should Be A Keyword In C++26)</a>
<sup><a title="assert Should Be A Keyword In C++26"
href="#ref_p2884r0">[P2884R0]</a></sup>
attempted to promote <tt>assert</tt> to a keyword, but failed to get consensus.
<a title="A natural syntax for Contracts"
href="https://wg21.link/p2961r2">
P2961R2 (A natural syntax for Contracts)</a>
<sup><a title="A natural syntax for Contracts"
href="#ref_p2961r2">[P2961R2]</a></sup>,
was approved by SG21 during the
<a href="https://wiki.edg.com/bin/view/Wg21kona2023/SG21">Kona, 2023 meeting</a>.
Section 5.2, "The <tt>assert</tt> name clash", details the challenges faced with
trying to repurpose that identifier.
</p
<p>
As a core language feature, <tt>contract_assert</tt> does not provoke ODR
violations due to <tt>NDEBUG</tt> being inconsistently defined across
translation units, is not subject to being redefined during translation as is
permissible for macros (note that <tt>contract_assert</tt> is proposed as a
keyword and is therefore not available for use as an identifier), and is not
confused by the presence of a comma that is not contained within a pair of
balanced parenthesis (an issue addressed for C assert by
<a title="Make assert() macro user friendly for C and C++"
href="https://wg21.link/p2264r7">
P2264R7 (Make assert() macro user friendly for C and C++)</a>
<sup><a title="Make assert() macro user friendly for C and C++"
href="#ref_p2264r7">[P2264R7]</a></sup>
for C++26).
<tt>contract_assert</tt> also guards against some cases of possibly unintended
behavior that may otherwise occur when its associated
<a href="http://eel.is/c++draft/expr.cond"><em>conditional-expression</em></a>
is evaluated.
It guards against possibly unintended mutation of some objects by implicitly
applying a const qualifier to the result of
<a href="http://eel.is/c++draft/expr.prim.id.general"><em>id-expression</em></a>s
that name a parameter or variable with automatic storage duration and to the
pointee type of
<a href="http://eel.is/c++draft/expr.prim.this"><em>this</em> expression</a>s.
It guards against unintended dependence on side effects by allowing an
implementation to elide evaluation when the result of the evaluation can be
predetermined or to perform the evaluation multiple times (with an expectation
that multiple evaluations will make such side effects more noticeable).
</p>
<p>
The differences between <tt>contract_assert</tt> and traditional assert
facilities make it impossible to directly reimplement traditional assert
facilities using <tt>contract_assert</tt> as currently proposed without imposing
potentially extensive changes to code that uses those facilities.
Further, the latter two behaviors (implicit application of a const qualifier
and evaluation elision or duplication) have become contentious due to a lack of
confidence that <tt>contract_assert</tt>, as currently proposed, suffices as
a modern alternative to traditional assert facilities like C assert.
</p>
<p>
Proposed are extensions to the P2900R6 design that are intended to enable
existing assert facilities, including C assert, to be optionally integrated
with C++ contracts, with minimal changes, such that assertion failures that
occur through use of those facilities are also reported through the C++
contract violation handler.
In doing so, the author hopes to reduce the current contention over the design
of <tt>contract_assert</tt> by allowing those that prefer the traditional
evaluation and side effect guarantees to use interfaces that behave as they
desire without compromising on the protections that <tt>contract_assert</tt>, as
currently proposed, provides.
</p>
<h1 id="motivation">Motivation</h1>
<p>
Vast amounts of code have been written over the lifetime of the C and C++
languages that utilize C assert or other traditional or ad hoc assertion
facilities.
It is not realistic to expect or require existing uses of traditional assert
facilities to be rewritten to use C++ contracts when such code is migrated to
C++26.
The limitations and pitfalls of traditional assert facilities provides good
motivation for a new modern facility.
However, greater motivation for use of a new facility is produced if it enables
integration with existing facilities (such integration also helps to reduce the
negative consequences of the infamous
<a href="https://xkcd.com/927">"there are N+1 competing standards"</a>
situation).
Specifically, the consequences of use of multiple assertion facilities, a
persistent reality for many programmers, is reduced if assertion failures can
be handled in a uniform way.
</p>
<p>
While it is not feasible to unconditionally claim the <tt>assert</tt> identifier
for use with C++ contracts without unacceptable backward compatibility impact,
it is feasible to offer the ability for projects to opt-in to a mode that
handles assertion failures in traditional <tt>assert()</tt> expressions by
invoking the C++ contract violation handler.
In the long term, perhaps with assistance from code modernization tools, such
integration could become standard practice and eventually allow <tt>assert</tt>
to be more fully integrated into C++ contracts.
</p>
<h1 id="design">Design considerations</h1>
<h2 id="design-handler">Invoking the contract violation handler</h2>
<p>
There are two basic approaches to enable a traditional assert facility to
invoke the contract violation handler.
<ul>
<li>Expose an interface to call <tt>::handle_contract_violation()</tt> with
a user constructed <tt>std::contracts::contract_violation</tt> object,
or</li>
<li>Expose a core language feature that evaluates an arbitrary expression and
then calls <tt>::handle_contract_violation()</tt> with an appropriately
constructed <tt>std::contracts::contract_violation</tt> object if that
evaluation yields a false result.</li>
</ul>
</p>
<p>
Enabling user construction of <tt>std::contracts::contract_violation</tt>
objects would require exposing a suitable constructor or factory function to
specify property values to be returned by the various property access member
functions.
<tt>std::contracts::contract_violation</tt> is designed to allow these property
values to be accessed from statically allocated memory associated with each
contract assertion present in the program that has a potentially checked
semantic.
Support for user construction would require specifying an ownership model for
user provided values.
Further, it is currently implementation-defined whether the class is
polymorphic; it could be an abstract class, but the property access member
functions are non-virtual.
Enabling user construction could provide flexibility for integration of
traditional assert facilities with C++ contracts, but would also impose
responsibility for appropriate selection of the <tt>contract_kind</tt>,
<tt>contract_semantic</tt>, and <tt>detection_mode</tt> property values.
</p>
<p>
<tt>contract_assert</tt> is a core language feature that is already capable of
evaluating an expression and invoking the contract violation handler.
The inability to use it to directly implement a traditional assert interface is
superimposed by the express desire to avoid well known problems with those
traditional interfaces.
However, it is possible to relax those imposed restrictions.
</p>
<p>
The grammar for the <tt>contract_assert</tt> statement was designed with
future extension in mind.
<blockquote class="quote">
<em>assertion-statement :</em>
<div style="margin-left: 1em;">
<tt>contract_assert</tt> <em>attribute-specifier-seq<sub>opt</sub></em> <tt>(</tt> <em>conditional-expression</em> <tt>)</tt> <tt>;</tt>
</div>
</blockquote>
</p>
<p>
While use of an attribute to opt-in to alternative semantics for the
<tt>contract_assert</tt> statement would not require syntactic extension, such
use would violate the WG21 ignorability policy for standard attributes.
The ignorability policy prohibits a standard attribute from selecting a
behavior that is not one of the permissible behaviors an implementation would be
allowed to select in the absence of the attribute; an attribute cannot be used
to relax constraints.
</p>
<p>
Fortunately, there is room for context sensitive keywords to be used in between
the <tt>contract_assert</tt> keyword and the opening parenthesis (either before
or after the optional <em>attribute-specifier-seq</em>) or between the closing
parenthesis and the semicolon.
</p>
<h2 id="design-ndebug">Interaction with <tt>NDEBUG</tt></h2>
<p>
C assert supports two translation modes.
When the <tt>NDEBUG</tt> macro is defined, arguments to <tt>assert</tt>
expressions are not evaluated; this is the traditional form of the
<em>ignore</em> semantic (and a form that leads to ODR violations).
When <tt>NDEBUG</tt> is not defined, <tt>assert</tt> exhibits behavior like
that of the <em>enforce</em> semantic, but with more stringent requirements on
how termination is performed; <tt>abort()</tt> is called and any further
processing requires the installation of a signal handler for the
<tt>SIGABRT</tt> signal.
</p>
<p>
Since <tt>NDEBUG</tt> may be used to conditionally remove code that an
<tt>assert</tt> expression depends on, it isn't possible, in general, to
parse arguments passed to <tt>assert</tt> when <tt>NDEBUG</tt> is defined.
It is therefore impossible to use <tt>NDEBUG</tt> as a proxy for a contract
semantic; when <tt>NDEBUG</tt> is defined, no contract specification is
present.
However, when <tt>NDEBUG</tt> is not defined, then all contract semantics
are applicable, though maintaining conformance with C standard requirements
for the behavior of <tt>assert</tt> requires selecting the <em>enforce</em>
semantic, writing the required information to the standard error stream,
and calling <tt>abort()</tt>.
</p>
<h1 id="proposal">Proposal</h1>
<h2 id="proposal-traditional-semantics">Extend traditional semantics to
<tt>contract_assert</tt></h2>
<p>
Add a new <tt>traditional_assert</tt> enumerator to
<tt>std::contracts::contract_kind</tt>.
</p>
<p>
Extend the <tt>contract_assert</tt> syntax to allow an optional
context sensitive <tt>traditional</tt> keyword after <tt>contract_assert</tt>.
<blockquote class="quote">
<em>assertion-statement :</em>
<div style="margin-left: 1em;">
<tt>contract_assert</tt> <ins><tt>traditional</tt><sub><em>opt</em></sub> </ins><em>attribute-specifier-seq<sub>opt</sub></em> <tt>(</tt> <em>conditional-expression</em> <tt>)</tt> <tt>;</tt>
</div>
</blockquote>
</p>
<p>
Modify the <tt>contract_assert</tt> behavior as follows when
<tt>traditional</tt> is specified.
<ul>
<li>Do not implicitly apply const qualification to parameters, variables
with automatic storage duration, or the pointee type of
<tt>this</tt>.</li>
<li>Evaluate the <em>conditional-expression</em> exactly one time each time
control reaches the <tt>contract_assert</tt> statement irregardless
of implementation-defined selection of contract semantics.</li>
<li>Propagate exceptions thrown during evaluation of the
<em>conditional-expression</em>; do not catch them and invoke the
violation handler.</li>
<li>If the <em>conditional-expression</em> evaluation yields a false result,
invoke the violation handler with a reference to a
<tt>std::contracts::contract_violation</tt> object for which
its <tt>detection_mode()</tt> member will return <tt>predicate_false</tt>,
its <tt>kind()</tt> member will return <tt>traditional_assert</tt>, and
its <tt>semantic()</tt> member will return <tt>enforce</tt>.</li>
</ul>
</p>
<p>
Extend the recommended practice for the default contract violation handler
to implement the behavior required by the C standard for a failed assertion
when invoked with a <tt>contract_violation</tt> object whose <tt>kind()</tt>
member returns <tt>traditional_assert</tt>.
This includes writing select information to the standard error stream and
then calling <tt>abort()</tt>.
</p>
<p>
Note that the above calls for the violation handler to be invoked with the
<em>enforce</em> contract semantic in effect.
This is required because there is no mechanism for the <tt>contract_assert</tt>
statement to indicate whether the <em>conditional-expression</em> was
evaluated, what its result was, or whether the violation handler was
invoked and returned.
Code following the <tt>contract_assert</tt> statement therefore has no
information available to determine what action should be taken next.
This limitation can be worked around by evaluating the assertion predicate
separately and using the result value for the <em>conditional-expression</em>
of the <tt>contract_assert</tt> statement, but this effectively hides the
expression from the contracts implementation and prevents exposing
it via the <tt>comment()</tt> member of
<tt>std::contracts::contract_violation</tt>.
Should motivation arise, additional extensions to enable use of an alternative
contract semantic can be added at a later time.
Selection of an alternate semantic could be enabled by parameterizing
<tt>traditional</tt> with an argument enclosed in angle brackets (e.g.,
<tt>traditional<observe></tt>; use of parenthesis would be ambiguous due
to the parenthesis that surround the <em>conditional-expression</em>).
</p>
<h2 id="proposal-assert-opt-in">Enable optional C++ contracts integration for C
assert</h2>
<p>
Upon inclusion of the <tt>assert.h</tt> or <tt>cassert</tt> header, if
the <tt>NDEBUG</tt> macro is not defined and the
<tt>ASSERT_USES_CONTRACTS</tt> macro is defined,
define the <tt>assert</tt> macro as follows.
<blockquote class="code">
#define assert(...) \
[&] { contract_assert traditional (__VA_ARGS__); }()
</blockquote>
</p>
<p>
Note that it is not an error to define both <tt>NDEBUG</tt> and
<tt>ASSERT_USES_CONTRACTS</tt>; when <tt>NDEBUG</tt> is defined,
<tt>ASSERT_USES_CONTRACTS</tt> has no effect.
The effects of the <tt>NDEBUG</tt> macro must take precedence to preserve
backward compatibility with existing code.
</p>
<h1 id="implementation-exp">Implementation experience</h1>
<p>
None.
</p>
<h1 id="ack">Acknowledgements</h1>
<p>
Thanks to all SG21 participants for the tremendous amount of time, energy, and
expertise that has been contributed towards advancing contracts for C++.
</p>
<h1 id="references">References</h1>
<table id="references">
<tr>
<td id="ref_p2264r7"><sup>[P2264R7]</sup></td>
<td>
"Make assert() macro user friendly for C and C++", P2264R7, 2023.<br/>
<a href="https://wg21.link/p2264r7">
https://wg21.link/p2264r7</a></td>
</tr>
<tr>
<td id="ref_p2884r0"><sup>[P2884R0]</sup></td>
<td>
"<tt>assert</tt> Should Be A Keyword In C++26", P2884R0, 2023.<br/>
<a href="https://wg21.link/p2884r0">
https://wg21.link/p2884r0</a></td>
</tr>
<tr>
<td id="ref_p2900r6"><sup>[P2900R6]</sup></td>
<td>
"Contracts for C++", P2900R6, 2024.<br/>
<a href="https://wg21.link/p2900r6">
https://wg21.link/p2900r6</a></td>
</tr>
<tr>
<td id="ref_p2961r2"><sup>[P2961R2]</sup></td>
<td>
"A natural syntax for Contracts", P2961R2, 2023.<br/>
<a href="https://wg21.link/p2961r2">
https://wg21.link/p2961r2</a></td>
</tr>
</table>
<h1 id="wording">Wording</h1>
<p>
These changes are relative to
<a title="Contracts for C++"
href="https://wg21.link/p2900r6">
P2900R6</a>
<sup><a title="Contracts for C++"
href="#ref_p2900r6">[P2900R6]</a></sup>.
</p>
<input type="checkbox" id="hideins">Hide inserted text</input><br/>
<input type="checkbox" id="hidedel">Hide deleted text</input>
<p>
Wording will be provided in a future revision contingent on SG21 encouraging
further work for the proposed direction.
</p>
</body>