blob: ca43e1951d52ed76af65f8669f152c94efa14257 [file] [log] [blame]
Douglas Gregorc41b6ff2010-06-30 22:01:08 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <META http-equiv="Content-Type" content="text/html; charset=ISO-8859-1" />
6 <title>Language Compatibility</title>
7 <link type="text/css" rel="stylesheet" href="menu.css" />
8 <link type="text/css" rel="stylesheet" href="content.css" />
9 <style type="text/css">
10</style>
11</head>
12<body>
13
14<!--#include virtual="menu.html.incl"-->
15
16<div id="content">
17
18<!-- ======================================================================= -->
19<h1>Language Compatibility</h1>
20<!-- ======================================================================= -->
21
22<p>Clang strives to both conform to current language standards (C99,
23 C++98) and also to implement many widely-used extensions available
24 in other compilers, so that most correct code will "just work" when
25 compiler with Clang. However, Clang is more strict than other
26 popular compilers, and may reject incorrect code that other
27 compilers allow. This page documents common compatibility and
28 portability issues with Clang to help you understand and fix the
29 problem in your code when Clang emits an error message.</p>
30
31<ul>
32 <li><a href="#c">C compatibility</a>
33 <ul>
34 <li><a href="#inline">C99 inline functions</a></li>
Chris Lattnera02d1832010-09-16 18:17:55 +000035 <li><a href="#vector_builtins">"missing" vector __builtin functions</a></li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000036 <li><a href="#lvalue-cast">Lvalue casts</a></li>
Daniel Dunbar5a410212010-09-02 21:35:16 +000037 <li><a href="#blocks-in-protected-scope">Jumps to within <tt>__block</tt> variable scope</a></li>
Daniel Dunbar15952c92010-11-09 22:45:16 +000038 <li><a href="#block-variable-initialization">Non-initialization of <tt>__block</tt> variables</a></li>
Chris Lattner9b743f42010-11-10 23:51:50 +000039 <li><a href="#inline-asm">Inline assembly</a></li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000040 </ul>
41 </li>
42 <li><a href="#objective-c">Objective-C compatibility</a>
43 <ul>
44 <li><a href="#super-cast">Cast of super</a></li>
45 <li><a href="#sizeof-interface">Size of interfaces</a></li>
Argyrios Kyrtzidis3b5b92a2010-09-13 17:48:07 +000046 <li><a href="#objc_objs-cast">Internal Objective-C types</a></li>
Fariborz Jahanianddfa6c32010-10-22 22:35:51 +000047 <li><a href="#c_variables-class">C variables in @class or @protocol</a></li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000048 </ul>
49 </li>
50 <li><a href="#c++">C++ compatibility</a>
51 <ul>
52 <li><a href="#vla">Variable-length arrays</a></li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000053 <li><a href="#dep_lookup">Unqualified lookup in templates</a></li>
54 <li><a href="#dep_lookup_bases">Unqualified lookup into dependent bases of class templates</a></li>
55 <li><a href="#undep_incomplete">Incomplete types in templates</a></li>
56 <li><a href="#bad_templates">Templates with no valid instantiations</a></li>
57 <li><a href="#default_init_const">Default initialization of const
58 variable of a class type requires user-defined default
59 constructor</a></li>
Douglas Gregora66d3bb2010-11-10 20:24:21 +000060 <li><a href="#param_name_lookup">Parameter name lookup</a></li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000061 </ul>
62 </li>
Douglas Gregorfb2a0c52011-09-27 18:58:27 +000063 <li><a href="#c++11">C++11 compatibility</a>
64 <ul>
65 <li><a href="#deleted-special-func">Deleted special member
66 functions</a></li>
67 </ul>
68 </li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000069 <li><a href="#objective-c++">Objective-C++ compatibility</a>
70 <ul>
71 <li><a href="#implicit-downcasts">Implicit downcasts</a></li>
72 </ul>
Fariborz Jahanian36e738a2010-08-11 18:57:26 +000073 <ul>
John McCall6966c672011-02-03 11:29:18 +000074 <li><a href="#class-as-property-name">Using <code>class</code> as a property name</a></li>
Fariborz Jahanian36e738a2010-08-11 18:57:26 +000075 </ul>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000076 </li>
77</ul>
78
79<!-- ======================================================================= -->
80<h2 id="c">C compatibility</h3>
81<!-- ======================================================================= -->
82
83<!-- ======================================================================= -->
84<h3 id="inline">C99 inline functions</h3>
85<!-- ======================================================================= -->
86<p>By default, Clang builds C code according to the C99 standard,
John McCall5ae84f22011-02-03 10:16:40 +000087which provides different semantics for the <code>inline</code> keyword
88than GCC's default behavior. For example, consider the following
89code:</p>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +000090<pre>
91inline int add(int i, int j) { return i + j; }
92
93int main() {
94 int i = add(4, 5);
95 return i;
96}
97</pre>
98
John McCall5ae84f22011-02-03 10:16:40 +000099<p>In C99, <code>inline</code> means that a function's definition is
100provided only for inlining, and that there is another definition
101(without <code>inline</code>) somewhere else in the program. That
102means that this program is incomplete, because if <code>add</code>
103isn't inlined (for example, when compiling without optimization), then
104<code>main</code> will have an unresolved reference to that other
105definition. Therefore we'll get a (correct) link-time error like this:</p>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000106
107<pre>
108Undefined symbols:
109 "_add", referenced from:
110 _main in cc-y1jXIr.o
111</pre>
112
John McCall5ae84f22011-02-03 10:16:40 +0000113<p>By contrast, GCC's default behavior follows the GNU89 dialect,
114which is the C89 standard plus a lot of extensions. C89 doesn't have
115an <code>inline</code> keyword, but GCC recognizes it as an extension
116and just treats it as a hint to the optimizer.</p>
117
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000118<p>There are several ways to fix this problem:</p>
119
120<ul>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000121 <li>Change <code>add</code> to a <code>static inline</code>
John McCall5ae84f22011-02-03 10:16:40 +0000122 function. This is usually the right solution if only one
123 translation unit needs to use the function. <code>static
124 inline</code> functions are always resolved within the translation
125 unit, so you won't have to add a non-<code>inline</code> definition
126 of the function elsewhere in your program.</li>
127
128 <li>Remove the <code>inline</code> keyword from this definition of
129 <code>add</code>. The <code>inline</code> keyword is not required
130 for a function to be inlined, nor does it guarantee that it will be.
131 Some compilers ignore it completely. Clang treats it as a mild
132 suggestion from the programmer.</li>
133
134 <li>Provide an external (non-<code>inline</code>) definition
135 of <code>add</code> somewhere else in your program. The two
136 definitions must be equivalent!</li>
137
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000138 <li>Compile with the GNU89 dialect by adding
139 <code>-std=gnu89</code> to the set of Clang options. This option is
140 only recommended if the program source cannot be changed or if the
141 program also relies on additional C89-specific behavior that cannot
142 be changed.</li>
143</ul>
144
John McCall5ae84f22011-02-03 10:16:40 +0000145<p>All of this only applies to C code; the meaning of <code>inline</code>
146in C++ is very different from its meaning in either GNU89 or C99.</p>
Chris Lattnera02d1832010-09-16 18:17:55 +0000147
148<!-- ======================================================================= -->
149<h3 id="vector_builtins">"missing" vector __builtin functions</h3>
150<!-- ======================================================================= -->
151
152<p>The Intel and AMD manuals document a number "<tt>&lt;*mmintrin.h&gt;</tt>"
153header files, which define a standardized API for accessing vector operations
154on X86 CPUs. These functions have names like <tt>_mm_xor_ps</tt> and
155<tt>_mm256_addsub_pd</tt>. Compilers have leeway to implement these functions
156however they want. Since Clang supports an excellent set of <a
157href="../docs/LanguageExtensions.html#vectors">native vector operations</a>,
158the Clang headers implement these interfaces in terms of the native vector
159operations.
160</p>
161
162<p>In contrast, GCC implements these functions mostly as a 1-to-1 mapping to
163builtin function calls, like <tt>__builtin_ia32_paddw128</tt>. These builtin
164functions are an internal implementation detail of GCC, and are not portable to
165the Intel compiler, the Microsoft compiler, or Clang. If you get build errors
166mentioning these, the fix is simple: switch to the *mmintrin.h functions.</p>
167
168<p>The same issue occurs for NEON and Altivec for the ARM and PowerPC
169architectures respectively. For these, make sure to use the &lt;arm_neon.h&gt;
170and &lt;altivec.h&gt; headers.</p>
171
Eric Christophera473c952010-10-25 21:17:59 +0000172<p>For x86 architectures this <a href="builtins.py">script</a> should help with
173the manual migration process. It will rewrite your source files in place to
174use the APIs instead of builtin function calls. Just call it like this:</p>
175
176<pre>
177 builtins.py *.c *.h
178</pre>
179
180<p>and it will rewrite all of the .c and .h files in the current directory to
181use the API calls instead of calls like <tt>__builtin_ia32_paddw128</tt>.</p>
Chris Lattnera02d1832010-09-16 18:17:55 +0000182
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000183<!-- ======================================================================= -->
184<h3 id="lvalue-cast">Lvalue casts</h3>
185<!-- ======================================================================= -->
186
Douglas Gregor6f1adba2010-06-30 22:38:37 +0000187<p>Old versions of GCC permit casting the left-hand side of an assignment to a
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000188different type. Clang produces an error on similar code, e.g.,</p>
189
190<pre>
191lvalue.c:2:3: error: assignment to cast is illegal, lvalue casts are not
192 supported
193 (int*)addr = val;
194 ^~~~~~~~~~ ~
195</pre>
196
197<p>To fix this problem, move the cast to the right-hand side. In this
198example, one could use:</p>
199
200<pre>
201 addr = (float *)val;
202</pre>
203
204<!-- ======================================================================= -->
Daniel Dunbar5a410212010-09-02 21:35:16 +0000205<h3 id="blocks-in-protected-scope">Jumps to within <tt>__block</tt> variable scope</h3>
206<!-- ======================================================================= -->
207
John McCall504b3692011-02-03 10:56:31 +0000208<p>Clang disallows jumps into the scope of a <tt>__block</tt>
209variable. Variables marked with <tt>__block</tt> require special
210runtime initialization. A jump into the scope of a <tt>__block</tt>
211variable bypasses this initialization, leaving the variable's metadata
212in an invalid state. Consider the following code fragment:</p>
Daniel Dunbar5a410212010-09-02 21:35:16 +0000213
214<pre>
John McCall504b3692011-02-03 10:56:31 +0000215int fetch_object_state(struct MyObject *c) {
216 if (!c->active) goto error;
Daniel Dunbar5a410212010-09-02 21:35:16 +0000217
John McCall504b3692011-02-03 10:56:31 +0000218 __block int result;
219 run_specially_somehow(^{ result = c->state; });
220 return result;
Daniel Dunbar5a410212010-09-02 21:35:16 +0000221
222 error:
John McCall504b3692011-02-03 10:56:31 +0000223 fprintf(stderr, "error while fetching object state");
224 return -1;
Daniel Dunbar5a410212010-09-02 21:35:16 +0000225}
226</pre>
227
John McCall504b3692011-02-03 10:56:31 +0000228<p>GCC accepts this code, but it produces code that will usually crash
229when <code>result</code> goes out of scope if the jump is taken. (It's
230possible for this bug to go undetected because it often won't crash if
231the stack is fresh, i.e. still zeroed.) Therefore, Clang rejects this
232code with a hard error:</p>
Daniel Dunbar5a410212010-09-02 21:35:16 +0000233
234<pre>
235t.c:3:5: error: goto into protected scope
236 goto error;
237 ^
238t.c:5:15: note: jump bypasses setup of __block variable
John McCall504b3692011-02-03 10:56:31 +0000239 __block int result;
Daniel Dunbar5a410212010-09-02 21:35:16 +0000240 ^
241</pre>
242
John McCall504b3692011-02-03 10:56:31 +0000243<p>The fix is to rewrite the code to not require jumping into a
244<tt>__block</tt> variable's scope, e.g. by limiting that scope:</p>
245
246<pre>
247 {
248 __block int result;
249 run_specially_somehow(^{ result = c->state; });
250 return result;
251 }
252</pre>
Daniel Dunbar5a410212010-09-02 21:35:16 +0000253
254<!-- ======================================================================= -->
Daniel Dunbar15952c92010-11-09 22:45:16 +0000255<h3 id="block-variable-initialization">Non-initialization of <tt>__block</tt>
256variables</h3>
257<!-- ======================================================================= -->
258
259<p>In the following example code, the <tt>x</tt> variable is used before it is
260defined:</p>
261<pre>
262int f0() {
263 __block int x;
264 return ^(){ return x; }();
265}
266</pre>
267
268<p>By an accident of implementation, GCC and llvm-gcc unintentionally always
269zero initialized <tt>__block</tt> variables. However, any program which depends
270on this behavior is relying on unspecified compiler behavior. Programs must
271explicitly initialize all local block variables before they are used, as with
272other local variables.</p>
273
274<p>Clang does not zero initialize local block variables, and programs which rely
275on such behavior will most likely break when built with Clang.</p>
276
Chris Lattner9b743f42010-11-10 23:51:50 +0000277
278<!-- ======================================================================= -->
279<h3 id="inline-asm">Inline assembly</h3>
280<!-- ======================================================================= -->
281
282<p>In general, Clang is highly compatible with the GCC inline assembly
283extensions, allowing the same set of constraints, modifiers and operands as GCC
284inline assembly.</p>
285
286<p>On targets that use the integrated assembler (such as most X86 targets),
287inline assembly is run through the integrated assembler instead of your system
288assembler (which is most commonly "gas", the GNU assembler). The LLVM
289integrated assembler is extremely compatible with GAS, but there are a couple of
290minor places where it is more picky, particularly due to outright GAS bugs.</p>
291
292<p>One specific example is that the assembler rejects ambiguous X86 instructions
293that don't have suffixes. For example:</p>
294
295<pre>
296 asm("add %al, (%rax)");
297 asm("addw $4, (%rax)");
298 asm("add $4, (%rax)");
299</pre>
300
301<p>Both clang and GAS accept the first instruction: because the first
302instruction uses the 8-bit <tt>%al</tt> register as an operand, it is clear that
303it is an 8-bit add. The second instruction is accepted by both because the "w"
304suffix indicates that it is a 16-bit add. The last instruction is accepted by
305GAS even though there is nothing that specifies the size of the instruction (and
306the assembler randomly picks a 32-bit add). Because it is ambiguous, Clang
307rejects the instruction with this error message:
308</p>
309
310<pre>
311&lt;inline asm&gt;:3:1: error: ambiguous instructions require an explicit suffix (could be 'addb', 'addw', 'addl', or 'addq')
312add $4, (%rax)
313^
3141 error generated.
315</pre>
316
317<p>To fix this compatibility issue, add an explicit suffix to the instruction:
318this makes your code more clear and is compatible with both GCC and Clang.</p>
319
Daniel Dunbar15952c92010-11-09 22:45:16 +0000320<!-- ======================================================================= -->
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000321<h2 id="objective-c">Objective-C compatibility</h3>
322<!-- ======================================================================= -->
323
324<!-- ======================================================================= -->
325<h3 id="super-cast">Cast of super</h3>
326<!-- ======================================================================= -->
327
328<p>GCC treats the <code>super</code> identifier as an expression that
329can, among other things, be cast to a different type. Clang treats
330<code>super</code> as a context-sensitive keyword, and will reject a
331type-cast of <code>super</code>:</p>
332
333<pre>
334super.m:11:12: error: cannot cast 'super' (it isn't an expression)
335 [(Super*)super add:4];
336 ~~~~~~~~^
337</pre>
338
339<p>To fix this problem, remove the type cast, e.g.</p>
340<pre>
341 [super add:4];
342</pre>
343
344<!-- ======================================================================= -->
345<h3 id="sizeof-interface">Size of interfaces</h3>
346<!-- ======================================================================= -->
347
348<p>When using the "non-fragile" Objective-C ABI in use, the size of an
349Objective-C class may change over time as instance variables are added
350(or removed). For this reason, Clang rejects the application of the
351<code>sizeof</code> operator to an Objective-C class when using this
352ABI:</p>
353
354<pre>
355sizeof.m:4:14: error: invalid application of 'sizeof' to interface 'NSArray' in
356 non-fragile ABI
357 int size = sizeof(NSArray);
358 ^ ~~~~~~~~~
359</pre>
360
361<p>Code that relies on the size of an Objective-C class is likely to
362be broken anyway, since that size is not actually constant. To address
363this problem, use the Objective-C runtime API function
Benjamin Kramere6617502010-06-30 22:29:56 +0000364<code>class_getInstanceSize()</code>:</p>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000365
366<pre>
367 class_getInstanceSize([NSArray class])
368</pre>
369
370<!-- ======================================================================= -->
Argyrios Kyrtzidis3b5b92a2010-09-13 17:48:07 +0000371<h3 id="objc_objs-cast">Internal Objective-C types</h3>
372<!-- ======================================================================= -->
373
374<p>GCC allows using pointers to internal Objective-C objects, <tt>struct objc_object*</tt>,
375<tt>struct objc_selector*</tt>, and <tt>struct objc_class*</tt> in place of the types
376<tt>id</tt>, <tt>SEL</tt>, and <tt>Class</tt> respectively. Clang treats the
377internal Objective-C structures as implementation detail and won't do implicit conversions:
378
379<pre>
380t.mm:11:2: error: no matching function for call to 'f'
381 f((struct objc_object *)p);
382 ^
383t.mm:5:6: note: candidate function not viable: no known conversion from 'struct objc_object *' to 'id' for 1st argument
384void f(id x);
385 ^
386</pre>
387
388<p>Code should use types <tt>id</tt>, <tt>SEL</tt>, and <tt>Class</tt>
389instead of the internal types.</p>
390
391<!-- ======================================================================= -->
John McCall3cdbe442011-02-03 11:05:04 +0000392<h3 id="c_variables-class">C variables in @interface or @protocol</h3>
Fariborz Jahanianddfa6c32010-10-22 22:35:51 +0000393<!-- ======================================================================= -->
394
John McCall3cdbe442011-02-03 11:05:04 +0000395<p>GCC allows the declaration of C variables in
396an <code>@interface</code> or <code>@protocol</code>
397declaration. Clang does not allow variable declarations to appear
398within these declarations unless they are marked <code>extern</code>.</p>
399
400<p>Variables may still be declared in an @implementation.</p>
Fariborz Jahanianddfa6c32010-10-22 22:35:51 +0000401
402<pre>
403@interface XX
John McCall3cdbe442011-02-03 11:05:04 +0000404int a; // not allowed in clang
405int b = 1; // not allowed in clang
406extern int c; // allowed
Fariborz Jahanianddfa6c32010-10-22 22:35:51 +0000407@end
408
409</pre>
410
411<!-- ======================================================================= -->
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000412<h2 id="c++">C++ compatibility</h3>
413<!-- ======================================================================= -->
414
415<!-- ======================================================================= -->
416<h3 id="vla">Variable-length arrays</h3>
417<!-- ======================================================================= -->
418
419<p>GCC and C99 allow an array's size to be determined at run
420time. This extension is not permitted in standard C++. However, Clang
421supports such variable length arrays in very limited circumstances for
422compatibility with GNU C and C99 programs:</p>
423
424<ul>
425 <li>The element type of a variable length array must be a POD
426 ("plain old data") type, which means that it cannot have any
John McCall6966c672011-02-03 11:29:18 +0000427 user-declared constructors or destructors, any base classes, or any
428 members of non-POD type. All C types are POD types.</li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000429
430 <li>Variable length arrays cannot be used as the type of a non-type
431template parameter.</li> </ul>
432
433<p>If your code uses variable length arrays in a manner that Clang doesn't support, there are several ways to fix your code:
434
435<ol>
436<li>replace the variable length array with a fixed-size array if you can
John McCall6966c672011-02-03 11:29:18 +0000437 determine a reasonable upper bound at compile time; sometimes this is as
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000438 simple as changing <tt>int size = ...;</tt> to <tt>const int size
John McCall6966c672011-02-03 11:29:18 +0000439 = ...;</tt> (if the initializer is a compile-time constant);</li>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000440<li>use <tt>std::vector</tt> or some other suitable container type;
441 or</li>
442<li>allocate the array on the heap instead using <tt>new Type[]</tt> -
443 just remember to <tt>delete[]</tt> it.</li>
444</ol>
445
446<!-- ======================================================================= -->
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000447<h3 id="dep_lookup">Unqualified lookup in templates</h3>
448<!-- ======================================================================= -->
449
450<p>Some versions of GCC accept the following invalid code:
451
452<pre>
453template &lt;typename T&gt; T Squared(T x) {
454 return Multiply(x, x);
455}
456
457int Multiply(int x, int y) {
458 return x * y;
459}
460
461int main() {
462 Squared(5);
463}
464</pre>
465
466<p>Clang complains:
467
Jay Foad2a00b832011-06-14 12:59:25 +0000468<pre> <b>my_file.cpp:2:10: <span class="error">error:</span> call to function 'Multiply' that is neither visible in the template definition nor found by argument-dependent lookup</b>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000469 return Multiply(x, x);
470 <span class="caret"> ^</span>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000471 <b>my_file.cpp:10:3: <span class="note">note:</span> in instantiation of function template specialization 'Squared&lt;int&gt;' requested here</b>
472 Squared(5);
473 <span class="caret"> ^</span>
Richard Smithf50e88a2011-06-05 22:42:48 +0000474 <b>my_file.cpp:5:5: <span class="note">note:</span> 'Multiply' should be declared prior to the call site</b>
475 int Multiply(int x, int y) {
476 <span class="caret"> ^</span>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000477</pre>
478
479<p>The C++ standard says that unqualified names like <q>Multiply</q>
480are looked up in two ways.
481
482<p>First, the compiler does <i>unqualified lookup</i> in the scope
483where the name was written. For a template, this means the lookup is
484done at the point where the template is defined, not where it's
485instantiated. Since <tt>Multiply</tt> hasn't been declared yet at
486this point, unqualified lookup won't find it.
487
488<p>Second, if the name is called like a function, then the compiler
489also does <i>argument-dependent lookup</i> (ADL). (Sometimes
490unqualified lookup can suppress ADL; see [basic.lookup.argdep]p3 for
491more information.) In ADL, the compiler looks at the types of all the
492arguments to the call. When it finds a class type, it looks up the
493name in that class's namespace; the result is all the declarations it
494finds in those namespaces, plus the declarations from unqualified
495lookup. However, the compiler doesn't do ADL until it knows all the
496argument types.
497
498<p>In our example, <tt>Multiply</tt> is called with dependent
499arguments, so ADL isn't done until the template is instantiated. At
500that point, the arguments both have type <tt>int</tt>, which doesn't
501contain any class types, and so ADL doesn't look in any namespaces.
502Since neither form of lookup found the declaration
503of <tt>Multiply</tt>, the code doesn't compile.
504
505<p>Here's another example, this time using overloaded operators,
506which obey very similar rules.
507
508<pre>#include &lt;iostream&gt;
509
510template&lt;typename T&gt;
511void Dump(const T&amp; value) {
512 std::cout &lt;&lt; value &lt;&lt; "\n";
513}
514
515namespace ns {
516 struct Data {};
517}
518
519std::ostream&amp; operator&lt;&lt;(std::ostream&amp; out, ns::Data data) {
520 return out &lt;&lt; "Some data";
521}
522
523void Use() {
524 Dump(ns::Data());
525}</pre>
526
Richard Smithf50e88a2011-06-05 22:42:48 +0000527<p>Again, Clang complains:</p>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000528
Jay Foad2a00b832011-06-14 12:59:25 +0000529<pre> <b>my_file2.cpp:5:13: <span class="error">error:</span> call to function 'operator&lt;&lt;' that is neither visible in the template definition nor found by argument-dependent lookup</b>
Richard Smithf50e88a2011-06-05 22:42:48 +0000530 std::cout &lt;&lt; value &lt;&lt; "\n";
531 <span class="caret"> ^</span>
532 <b>my_file2.cpp:17:3: <span class="error">note:</span> in instantiation of function template specialization 'Dump&lt;ns::Data&gt;' requested here</b>
533 Dump(ns::Data());
534 <span class="caret"> ^</span>
535 <b>my_file2.cpp:12:15: <span class="error">note:</span> 'operator&lt;&lt;' should be declared prior to the call site or in namespace 'ns'</b>
536 std::ostream&amp; operator&lt;&lt;(std::ostream&amp; out, ns::Data data) {
537 <span class="caret"> ^</span>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000538</pre>
539
540<p>Just like before, unqualified lookup didn't find any declarations
541with the name <tt>operator&lt;&lt;</tt>. Unlike before, the argument
542types both contain class types: one of them is an instance of the
543class template type <tt>std::basic_ostream</tt>, and the other is the
544type <tt>ns::Data</tt> that we declared above. Therefore, ADL will
545look in the namespaces <tt>std</tt> and <tt>ns</tt> for
546an <tt>operator&lt;&lt;</tt>. Since one of the argument types was
547still dependent during the template definition, ADL isn't done until
548the template is instantiated during <tt>Use</tt>, which means that
549the <tt>operator&lt;&lt;</tt> we want it to find has already been
550declared. Unfortunately, it was declared in the global namespace, not
551in either of the namespaces that ADL will look in!
552
553<p>There are two ways to fix this problem:</p>
554<ol><li>Make sure the function you want to call is declared before the
555template that might call it. This is the only option if none of its
556argument types contain classes. You can do this either by moving the
557template definition, or by moving the function definition, or by
558adding a forward declaration of the function before the template.</li>
559<li>Move the function into the same namespace as one of its arguments
560so that ADL applies.</li></ol>
561
562<p>For more information about argument-dependent lookup, see
563[basic.lookup.argdep]. For more information about the ordering of
564lookup in templates, see [temp.dep.candidate].
565
566<!-- ======================================================================= -->
567<h3 id="dep_lookup_bases">Unqualified lookup into dependent bases of class templates</h3>
568<!-- ======================================================================= -->
569
570Some versions of GCC accept the following invalid code:
571
572<pre>
573template &lt;typename T&gt; struct Base {
574 void DoThis(T x) {}
575 static void DoThat(T x) {}
576};
577
578template &lt;typename T&gt; struct Derived : public Base&lt;T&gt; {
579 void Work(T x) {
580 DoThis(x); // Invalid!
581 DoThat(x); // Invalid!
582 }
583};
584</pre>
585
586Clang correctly rejects it with the following errors
587(when <tt>Derived</tt> is eventually instantiated):
588
589<pre>
590my_file.cpp:8:5: error: use of undeclared identifier 'DoThis'
591 DoThis(x);
592 ^
593 this-&gt;
594my_file.cpp:2:8: note: must qualify identifier to find this declaration in dependent base class
595 void DoThis(T x) {}
596 ^
597my_file.cpp:9:5: error: use of undeclared identifier 'DoThat'
598 DoThat(x);
599 ^
600 this-&gt;
601my_file.cpp:3:15: note: must qualify identifier to find this declaration in dependent base class
602 static void DoThat(T x) {}
603</pre>
604
605Like we said <a href="#dep_lookup">above</a>, unqualified names like
606<tt>DoThis</tt> and <tt>DoThat</tt> are looked up when the template
607<tt>Derived</tt> is defined, not when it's instantiated. When we look
608up a name used in a class, we usually look into the base classes.
609However, we can't look into the base class <tt>Base&lt;T&gt;</tt>
610because its type depends on the template argument <tt>T</tt>, so the
611standard says we should just ignore it. See [temp.dep]p3 for details.
612
613<p>The fix, as Clang tells you, is to tell the compiler that we want a
614class member by prefixing the calls with <tt>this-&gt;</tt>:
615
616<pre>
617 void Work(T x) {
618 <b>this-&gt;</b>DoThis(x);
619 <b>this-&gt;</b>DoThat(x);
620 }
621</pre>
622
623Alternatively, you can tell the compiler exactly where to look:
624
625<pre>
626 void Work(T x) {
627 <b>Base&lt;T&gt;</b>::DoThis(x);
628 <b>Base&lt;T&gt;</b>::DoThat(x);
629 }
630</pre>
631
632This works whether the methods are static or not, but be careful:
633if <tt>DoThis</tt> is virtual, calling it this way will bypass virtual
634dispatch!
635
636<!-- ======================================================================= -->
637<h3 id="undep_incomplete">Incomplete types in templates</h3>
638<!-- ======================================================================= -->
639
640The following code is invalid, but compilers are allowed to accept it:
641
642<pre>
643 class IOOptions;
644 template &lt;class T&gt; bool read(T &amp;value) {
645 IOOptions opts;
646 return read(opts, value);
647 }
648
649 class IOOptions { bool ForceReads; };
650 bool read(const IOOptions &amp;opts, int &amp;x);
651 template bool read&lt;&gt;(int &amp;);
652</pre>
653
654The standard says that types which don't depend on template parameters
655must be complete when a template is defined if they affect the
656program's behavior. However, the standard also says that compilers
657are free to not enforce this rule. Most compilers enforce it to some
658extent; for example, it would be an error in GCC to
659write <tt>opts.ForceReads</tt> in the code above. In Clang, we feel
660that enforcing the rule consistently lets us provide a better
661experience, but unfortunately it also means we reject some code that
662other compilers accept.
663
664<p>We've explained the rule here in very imprecise terms; see
665[temp.res]p8 for details.
666
667<!-- ======================================================================= -->
668<h3 id="bad_templates">Templates with no valid instantiations</h3>
669<!-- ======================================================================= -->
670
671The following code contains a typo: the programmer
672meant <tt>init()</tt> but wrote <tt>innit()</tt> instead.
673
674<pre>
675 template &lt;class T&gt; class Processor {
676 ...
677 void init();
678 ...
679 };
680 ...
681 template &lt;class T&gt; void process() {
682 Processor&lt;T&gt; processor;
683 processor.innit(); // <-- should be 'init()'
684 ...
685 }
686</pre>
687
688Unfortunately, we can't flag this mistake as soon as we see it: inside
689a template, we're not allowed to make assumptions about "dependent
690types" like <tt>Processor&lt;T&gt;</tt>. Suppose that later on in
691this file the programmer adds an explicit specialization
692of <tt>Processor</tt>, like so:
693
694<pre>
695 template &lt;&gt; class Processor&lt;char*&gt; {
696 void innit();
697 };
698</pre>
699
700Now the program will work &mdash; as long as the programmer only ever
701instantiates <tt>process()</tt> with <tt>T = char*</tt>! This is why
702it's hard, and sometimes impossible, to diagnose mistakes in a
703template definition before it's instantiated.
704
705<p>The standard says that a template with no valid instantiations is
706ill-formed. Clang tries to do as much checking as possible at
707definition-time instead of instantiation-time: not only does this
708produce clearer diagnostics, but it also substantially improves
709compile times when using pre-compiled headers. The downside to this
710philosophy is that Clang sometimes fails to process files because they
711contain broken templates that are no longer used. The solution is
712simple: since the code is unused, just remove it.
713
714<!-- ======================================================================= -->
715<h3 id="default_init_const">Default initialization of const variable of a class type requires user-defined default constructor</h3>
716<!-- ======================================================================= -->
717
718If a <tt>class</tt> or <tt>struct</tt> has no user-defined default
719constructor, C++ doesn't allow you to default construct a <tt>const</tt>
720instance of it like this ([dcl.init], p9):
721
722<pre>
723class Foo {
724 public:
725 // The compiler-supplied default constructor works fine, so we
726 // don't bother with defining one.
727 ...
728};
729
730void Bar() {
731 const Foo foo; // Error!
732 ...
733}
734</pre>
735
736To fix this, you can define a default constructor for the class:
737
738<pre>
739class Foo {
740 public:
741 Foo() {}
742 ...
743};
744
745void Bar() {
746 const Foo foo; // Now the compiler is happy.
747 ...
748}
749</pre>
750
751<!-- ======================================================================= -->
Douglas Gregora66d3bb2010-11-10 20:24:21 +0000752<h3 id="param_name_lookup">Parameter name lookup</h3>
753<!-- ======================================================================= -->
754
755<p>Due to a bug in its implementation, GCC allows the redeclaration of function parameter names within a function prototype in C++ code, e.g.</p>
756<blockquote>
757<pre>
758void f(int a, int a);
759</pre>
760</blockquote>
761<p>Clang diagnoses this error (where the parameter name has been redeclared). To fix this problem, rename one of the parameters.</p>
762
763<!-- ======================================================================= -->
Douglas Gregorfb2a0c52011-09-27 18:58:27 +0000764<h2 id="c++11">C++11 compatibility</h2>
765<!-- ======================================================================= -->
766
767<!-- ======================================================================= -->
768<h3 id="deleted-special-func">Deleted special member functions</h3>
769<!-- ======================================================================= -->
770
771<p>In C++11, the explicit declaration of a move constructor or a move
Sean Hunt04bea932011-12-29 18:40:13 +0000772assignment operator within a class deletes the implicit declaration
Douglas Gregorfb2a0c52011-09-27 18:58:27 +0000773of the copy constructor and copy assignment operator. This change came
774fairly late in the C++11 standardization process, so early
775implementations of C++11 (including Clang before 3.0, GCC before 4.7,
776and Visual Studio 2010) do not implement this rule, leading them to
777accept this ill-formed code:</p>
778
779<pre>
780struct X {
Sean Hunt04bea932011-12-29 18:40:13 +0000781 X(X&amp;&amp;); <i>// deletes implicit copy constructor:</i>
782 <i>// X(const X&amp;) = delete;</i>
Douglas Gregorfb2a0c52011-09-27 18:58:27 +0000783};
784
785void f(X x);
786void g(X x) {
Sean Hunt04bea932011-12-29 18:40:13 +0000787 f(x); <i>// error: X has a deleted copy constructor</i>
Douglas Gregorfb2a0c52011-09-27 18:58:27 +0000788}
789</pre>
790
Sean Hunt04bea932011-12-29 18:40:13 +0000791<p>This affects some early C++11 code, including Boost's popular <a
Douglas Gregorfb2a0c52011-09-27 18:58:27 +0000792href="http://www.boost.org/doc/libs/release/libs/smart_ptr/shared_ptr.htm"><tt>shared_ptr</tt></a>
793up to version 1.47.0. The fix for Boost's <tt>shared_ptr</tt> is
794<a href="https://svn.boost.org/trac/boost/changeset/73202">available here</a>.</p>
795
796<!-- ======================================================================= -->
797<h2 id="objective-c++">Objective-C++ compatibility</h2>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000798<!-- ======================================================================= -->
799
800<!-- ======================================================================= -->
801<h3 id="implicit-downcasts">Implicit downcasts</h3>
802<!-- ======================================================================= -->
803
804<p>Due to a bug in its implementation, GCC allows implicit downcasts
John McCall6966c672011-02-03 11:29:18 +0000805of Objective-C pointers (from a base class to a derived class) when
806calling functions. Such code is inherently unsafe, since the object
807might not actually be an instance of the derived class, and is
808rejected by Clang. For example, given this code:</p>
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000809
810<pre>
811@interface Base @end
812@interface Derived : Base @end
813
John McCall6966c672011-02-03 11:29:18 +0000814void f(Derived *p);
815void g(Base *p) {
816 f(p);
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000817}
818</pre>
819
820<p>Clang produces the following error:</p>
821
822<pre>
823downcast.mm:6:3: error: no matching function for call to 'f'
John McCall6966c672011-02-03 11:29:18 +0000824 f(p);
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000825 ^
Douglas Gregor92bc0272010-07-01 03:50:01 +0000826downcast.mm:4:6: note: candidate function not viable: cannot convert from
827 superclass 'Base *' to subclass 'Derived *' for 1st argument
John McCall6966c672011-02-03 11:29:18 +0000828void f(Derived *p);
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000829 ^
830</pre>
831
832<p>If the downcast is actually correct (e.g., because the code has
833already checked that the object has the appropriate type), add an
834explicit cast:</p>
835
836<pre>
837 f((Derived *)base);
838</pre>
839
Fariborz Jahanian36e738a2010-08-11 18:57:26 +0000840<!-- ======================================================================= -->
John McCall6966c672011-02-03 11:29:18 +0000841<h3 id="class-as-property-name">Using <code>class</code> as a property name</h3>
Fariborz Jahanian36e738a2010-08-11 18:57:26 +0000842<!-- ======================================================================= -->
843
John McCall6966c672011-02-03 11:29:18 +0000844<p>In C and Objective-C, <code>class</code> is a normal identifier and
845can be used to name fields, ivars, methods, and so on. In
846C++, <code>class</code> is a keyword. For compatibility with existing
847code, Clang permits <code>class</code> to be used as part of a method
848selector in Objective-C++, but this does not extend to any other part
849of the language. In particular, it is impossible to use property dot
850syntax in Objective-C++ with the property name <code>class</code>, so
851the following code will fail to parse:</p>
Fariborz Jahanian36e738a2010-08-11 18:57:26 +0000852
853<pre>
854@interface I {
855int cls;
856}
857+ (int)class;
858@end
859
860@implementation I
861- (int) Meth { return I.class; }
862@end
863<pre>
864
John McCall6966c672011-02-03 11:29:18 +0000865<p>Use explicit message-send syntax instead, i.e. <code>[I class]</code>.</p>
Fariborz Jahanian36e738a2010-08-11 18:57:26 +0000866
Douglas Gregorc41b6ff2010-06-30 22:01:08 +0000867</div>
868</body>
869</html>