1 ============================
2 Clang Compiler User's Manual
3 ============================
11 The Clang Compiler is an open-source compiler for the C family of
12 programming languages, aiming to be the best in class implementation of
13 these languages. Clang builds on the LLVM optimizer and code generator,
14 allowing it to provide high-quality optimization and code generation
15 support for many targets. For more general information, please see the
16 `Clang Web Site <http://clang.llvm.org>`_ or the `LLVM Web
17 Site <http://llvm.org>`_.
19 This document describes important notes about using Clang as a compiler
20 for an end-user, documenting the supported features, command line
21 options, etc. If you are interested in using Clang to build a tool that
22 processes code, please see :doc:`InternalsManual`. If you are interested in the
23 `Clang Static Analyzer <http://clang-analyzer.llvm.org>`_, please see its web
26 Clang is designed to support the C family of programming languages,
27 which includes :ref:`C <c>`, :ref:`Objective-C <objc>`, :ref:`C++ <cxx>`, and
28 :ref:`Objective-C++ <objcxx>` as well as many dialects of those. For
29 language-specific information, please see the corresponding language
32 - :ref:`C Language <c>`: K&R C, ANSI C89, ISO C90, ISO C94 (C89+AMD1), ISO
34 - :ref:`Objective-C Language <objc>`: ObjC 1, ObjC 2, ObjC 2.1, plus
35 variants depending on base language.
36 - :ref:`C++ Language <cxx>`
37 - :ref:`Objective C++ Language <objcxx>`
39 In addition to these base languages and their dialects, Clang supports a
40 broad variety of language extensions, which are documented in the
41 corresponding language section. These extensions are provided to be
42 compatible with the GCC, Microsoft, and other popular compilers as well
43 as to improve functionality through Clang-specific features. The Clang
44 driver and language features are intentionally designed to be as
45 compatible with the GNU GCC compiler as reasonably possible, easing
46 migration from GCC to Clang. In most cases, code "just works".
47 Clang also provides an alternative driver, :ref:`clang-cl`, that is designed
48 to be compatible with the Visual C++ compiler, cl.exe.
50 In addition to language specific features, Clang has a variety of
51 features that depend on what CPU architecture or operating system is
52 being compiled for. Please see the :ref:`Target-Specific Features and
53 Limitations <target_features>` section for more details.
55 The rest of the introduction introduces some basic :ref:`compiler
56 terminology <terminology>` that is used throughout this manual and
57 contains a basic :ref:`introduction to using Clang <basicusage>` as a
58 command line compiler.
65 Front end, parser, backend, preprocessor, undefined behavior,
73 Intro to how to use a C compiler for newbies.
75 compile + link compile then link debug info enabling optimizations
76 picking a language to use, defaults to C11 by default. Autosenses based
77 on extension. using a makefile
82 This section is generally an index into other sections. It does not go
83 into depth on the ones that are covered by other sections. However, the
84 first part introduces the language selection and other high level
85 options like :option:`-c`, :option:`-g`, etc.
87 Options to Control Error and Warning Messages
88 ---------------------------------------------
92 Turn warnings into errors.
94 .. This is in plain monospaced font because it generates the same label as
95 .. -Werror, and Sphinx complains.
99 Turn warning "foo" into an error.
101 .. option:: -Wno-error=foo
103 Turn warning "foo" into an warning even if :option:`-Werror` is specified.
107 Enable warning "foo".
111 Disable warning "foo".
115 Disable all diagnostics.
117 .. option:: -Weverything
119 :ref:`Enable all diagnostics. <diagnostics_enable_everything>`
121 .. option:: -pedantic
123 Warn on language extensions.
125 .. option:: -pedantic-errors
127 Error on language extensions.
129 .. option:: -Wsystem-headers
131 Enable warnings from system headers.
133 .. option:: -ferror-limit=123
135 Stop emitting diagnostics after 123 errors have been produced. The default is
136 20, and the error limit can be disabled with :option:`-ferror-limit=0`.
138 .. option:: -ftemplate-backtrace-limit=123
140 Only emit up to 123 template instantiation notes within the template
141 instantiation backtrace for a single warning or error. The default is 10, and
142 the limit can be disabled with :option:`-ftemplate-backtrace-limit=0`.
144 .. _cl_diag_formatting:
146 Formatting of Diagnostics
147 ^^^^^^^^^^^^^^^^^^^^^^^^^
149 Clang aims to produce beautiful diagnostics by default, particularly for
150 new users that first come to Clang. However, different people have
151 different preferences, and sometimes Clang is driven by another program
152 that wants to parse simple and consistent output, not a person. For
153 these cases, Clang provides a wide range of options to control the exact
154 output format of the diagnostics that it generates.
156 .. _opt_fshow-column:
158 **-f[no-]show-column**
159 Print column number in diagnostic.
161 This option, which defaults to on, controls whether or not Clang
162 prints the column number of a diagnostic. For example, when this is
163 enabled, Clang will print something like:
167 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
172 When this is disabled, Clang will print "test.c:28: warning..." with
175 The printed column numbers count bytes from the beginning of the
176 line; take care if your source contains multibyte characters.
178 .. _opt_fshow-source-location:
180 **-f[no-]show-source-location**
181 Print source file/line/column information in diagnostic.
183 This option, which defaults to on, controls whether or not Clang
184 prints the filename, line number and column number of a diagnostic.
185 For example, when this is enabled, Clang will print something like:
189 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
194 When this is disabled, Clang will not print the "test.c:28:8: "
197 .. _opt_fcaret-diagnostics:
199 **-f[no-]caret-diagnostics**
200 Print source line and ranges from source code in diagnostic.
201 This option, which defaults to on, controls whether or not Clang
202 prints the source line, source ranges, and caret when emitting a
203 diagnostic. For example, when this is enabled, Clang will print
208 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
213 **-f[no-]color-diagnostics**
214 This option, which defaults to on when a color-capable terminal is
215 detected, controls whether or not Clang prints diagnostics in color.
217 When this option is enabled, Clang will use colors to highlight
218 specific parts of the diagnostic, e.g.,
220 .. nasty hack to not lose our dignity
225 <b><span style="color:black">test.c:28:8: <span style="color:magenta">warning</span>: extra tokens at end of #endif directive [-Wextra-tokens]</span></b>
227 <span style="color:green">^</span>
228 <span style="color:green">//</span>
231 When this is disabled, Clang will just print:
235 test.c:2:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
240 **-fansi-escape-codes**
241 Controls whether ANSI escape codes are used instead of the Windows Console
242 API to output colored diagnostics. This option is only used on Windows and
245 .. option:: -fdiagnostics-format=clang/msvc/vi
247 Changes diagnostic output format to better match IDEs and command line tools.
249 This option controls the output format of the filename, line number,
250 and column printed in diagnostic messages. The options, and their
251 affect on formatting a simple conversion diagnostic, follow:
256 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int'
261 t.c(3,11) : warning: conversion specifies type 'char *' but the argument has type 'int'
266 t.c +3:11: warning: conversion specifies type 'char *' but the argument has type 'int'
268 .. _opt_fdiagnostics-show-option:
270 **-f[no-]diagnostics-show-option**
271 Enable ``[-Woption]`` information in diagnostic line.
273 This option, which defaults to on, controls whether or not Clang
274 prints the associated :ref:`warning group <cl_diag_warning_groups>`
275 option name when outputting a warning diagnostic. For example, in
280 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
285 Passing **-fno-diagnostics-show-option** will prevent Clang from
286 printing the [:ref:`-Wextra-tokens <opt_Wextra-tokens>`] information in
287 the diagnostic. This information tells you the flag needed to enable
288 or disable the diagnostic, either from the command line or through
289 :ref:`#pragma GCC diagnostic <pragma_GCC_diagnostic>`.
291 .. _opt_fdiagnostics-show-category:
293 .. option:: -fdiagnostics-show-category=none/id/name
295 Enable printing category information in diagnostic line.
297 This option, which defaults to "none", controls whether or not Clang
298 prints the category associated with a diagnostic when emitting it.
299 Each diagnostic may or many not have an associated category, if it
300 has one, it is listed in the diagnostic categorization field of the
301 diagnostic line (in the []'s).
303 For example, a format string warning will produce these three
304 renditions based on the setting of this option:
308 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' [-Wformat]
309 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' [-Wformat,1]
310 t.c:3:11: warning: conversion specifies type 'char *' but the argument has type 'int' [-Wformat,Format String]
312 This category can be used by clients that want to group diagnostics
313 by category, so it should be a high level category. We want dozens
314 of these, not hundreds or thousands of them.
316 .. _opt_fdiagnostics-fixit-info:
318 **-f[no-]diagnostics-fixit-info**
319 Enable "FixIt" information in the diagnostics output.
321 This option, which defaults to on, controls whether or not Clang
322 prints the information on how to fix a specific diagnostic
323 underneath it when it knows. For example, in this output:
327 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
332 Passing **-fno-diagnostics-fixit-info** will prevent Clang from
333 printing the "//" line at the end of the message. This information
334 is useful for users who may not understand what is wrong, but can be
335 confusing for machine parsing.
337 .. _opt_fdiagnostics-print-source-range-info:
339 **-fdiagnostics-print-source-range-info**
340 Print machine parsable information about source ranges.
341 This option makes Clang print information about source ranges in a machine
342 parsable format after the file/line/column number information. The
343 information is a simple sequence of brace enclosed ranges, where each range
344 lists the start and end line/column locations. For example, in this output:
348 exprs.c:47:15:{47:8-47:14}{47:17-47:24}: error: invalid operands to binary expression ('int *' and '_Complex float')
349 P = (P-42) + Gamma*4;
352 The {}'s are generated by -fdiagnostics-print-source-range-info.
354 The printed column numbers count bytes from the beginning of the
355 line; take care if your source contains multibyte characters.
357 .. option:: -fdiagnostics-parseable-fixits
359 Print Fix-Its in a machine parseable form.
361 This option makes Clang print available Fix-Its in a machine
362 parseable format at the end of diagnostics. The following example
363 illustrates the format:
367 fix-it:"t.cpp":{7:25-7:29}:"Gamma"
369 The range printed is a half-open range, so in this example the
370 characters at column 25 up to but not including column 29 on line 7
371 in t.cpp should be replaced with the string "Gamma". Either the
372 range or the replacement string may be empty (representing strict
373 insertions and strict erasures, respectively). Both the file name
374 and the insertion string escape backslash (as "\\\\"), tabs (as
375 "\\t"), newlines (as "\\n"), double quotes(as "\\"") and
376 non-printable characters (as octal "\\xxx").
378 The printed column numbers count bytes from the beginning of the
379 line; take care if your source contains multibyte characters.
381 .. option:: -fno-elide-type
383 Turns off elision in template type printing.
385 The default for template type printing is to elide as many template
386 arguments as possible, removing those which are the same in both
387 template types, leaving only the differences. Adding this flag will
388 print all the template arguments. If supported by the terminal,
389 highlighting will still appear on differing arguments.
395 t.cc:4:5: note: candidate function not viable: no known conversion from 'vector<map<[...], map<float, [...]>>>' to 'vector<map<[...], map<double, [...]>>>' for 1st argument;
401 t.cc:4:5: note: candidate function not viable: no known conversion from 'vector<map<int, map<float, int>>>' to 'vector<map<int, map<double, int>>>' for 1st argument;
403 .. option:: -fdiagnostics-show-template-tree
405 Template type diffing prints a text tree.
407 For diffing large templated types, this option will cause Clang to
408 display the templates as an indented text tree, one argument per
409 line, with differences marked inline. This is compatible with
416 t.cc:4:5: note: candidate function not viable: no known conversion from 'vector<map<[...], map<float, [...]>>>' to 'vector<map<[...], map<double, [...]>>>' for 1st argument;
418 With :option:`-fdiagnostics-show-template-tree`:
422 t.cc:4:5: note: candidate function not viable: no known conversion for 1st argument;
430 .. _cl_diag_warning_groups:
432 Individual Warning Groups
433 ^^^^^^^^^^^^^^^^^^^^^^^^^
435 TODO: Generate this from tblgen. Define one anchor per warning group.
437 .. _opt_wextra-tokens:
439 .. option:: -Wextra-tokens
441 Warn about excess tokens at the end of a preprocessor directive.
443 This option, which defaults to on, enables warnings about extra
444 tokens at the end of preprocessor directives. For example:
448 test.c:28:8: warning: extra tokens at end of #endif directive [-Wextra-tokens]
452 These extra tokens are not strictly conforming, and are usually best
453 handled by commenting them out.
455 .. option:: -Wambiguous-member-template
457 Warn about unqualified uses of a member template whose name resolves to
458 another template at the location of the use.
460 This option, which defaults to on, enables a warning in the
465 template<typename T> struct set{};
466 template<typename T> struct trait { typedef const T& type; };
468 template<typename T> void set(typename trait<T>::type value) {}
475 C++ [basic.lookup.classref] requires this to be an error, but,
476 because it's hard to work around, Clang downgrades it to a warning
479 .. option:: -Wbind-to-temporary-copy
481 Warn about an unusable copy constructor when binding a reference to a
484 This option enables warnings about binding a
485 reference to a temporary when the temporary doesn't have a usable
486 copy constructor. For example:
493 NonCopyable(const NonCopyable&);
495 void foo(const NonCopyable&);
497 foo(NonCopyable()); // Disallowed in C++98; allowed in C++11.
502 struct NonCopyable2 {
504 NonCopyable2(NonCopyable2&);
506 void foo(const NonCopyable2&);
508 foo(NonCopyable2()); // Disallowed in C++98; allowed in C++11.
511 Note that if ``NonCopyable2::NonCopyable2()`` has a default argument
512 whose instantiation produces a compile error, that error will still
513 be a hard error in C++98 mode even if this warning is turned off.
515 Options to Control Clang Crash Diagnostics
516 ------------------------------------------
518 As unbelievable as it may sound, Clang does crash from time to time.
519 Generally, this only occurs to those living on the `bleeding
520 edge <http://llvm.org/releases/download.html#svn>`_. Clang goes to great
521 lengths to assist you in filing a bug report. Specifically, Clang
522 generates preprocessed source file(s) and associated run script(s) upon
523 a crash. These files should be attached to a bug report to ease
524 reproducibility of the failure. Below are the command line options to
525 control the crash diagnostics.
527 .. option:: -fno-crash-diagnostics
529 Disable auto-generation of preprocessed source files during a clang crash.
531 The -fno-crash-diagnostics flag can be helpful for speeding the process
532 of generating a delta reduced test case.
534 Options to Emit Optimization Reports
535 ------------------------------------
537 Optimization reports trace, at a high-level, all the major decisions
538 done by compiler transformations. For instance, when the inliner
539 decides to inline function ``foo()`` into ``bar()``, or the loop unroller
540 decides to unroll a loop N times, or the vectorizer decides to
541 vectorize a loop body.
543 Clang offers a family of flags which the optimizers can use to emit
544 a diagnostic in three cases:
546 1. When the pass makes a transformation (:option:`-Rpass`).
548 2. When the pass fails to make a transformation (:option:`-Rpass-missed`).
550 3. When the pass determines whether or not to make a transformation
551 (:option:`-Rpass-analysis`).
553 NOTE: Although the discussion below focuses on :option:`-Rpass`, the exact
554 same options apply to :option:`-Rpass-missed` and :option:`-Rpass-analysis`.
556 Since there are dozens of passes inside the compiler, each of these flags
557 take a regular expression that identifies the name of the pass which should
558 emit the associated diagnostic. For example, to get a report from the inliner,
559 compile the code with:
561 .. code-block:: console
563 $ clang -O2 -Rpass=inline code.cc -o code
564 code.cc:4:25: remark: foo inlined into bar [-Rpass=inline]
565 int bar(int j) { return foo(j, j - 2); }
568 Note that remarks from the inliner are identified with `[-Rpass=inline]`.
569 To request a report from every optimization pass, you should use
570 :option:`-Rpass=.*` (in fact, you can use any valid POSIX regular
571 expression). However, do not expect a report from every transformation
572 made by the compiler. Optimization remarks do not really make sense
573 outside of the major transformations (e.g., inlining, vectorization,
574 loop optimizations) and not every optimization pass supports this
580 1. Optimization remarks that refer to function names will display the
581 mangled name of the function. Since these remarks are emitted by the
582 back end of the compiler, it does not know anything about the input
583 language, nor its mangling rules.
585 2. Some source locations are not displayed correctly. The front end has
586 a more detailed source location tracking than the locations included
587 in the debug info (e.g., the front end can locate code inside macro
588 expansions). However, the locations used by :option:`-Rpass` are
589 translated from debug annotations. That translation can be lossy,
590 which results in some remarks having no location information.
594 Clang options that that don't fit neatly into other categories.
598 When emitting a dependency file, use formatting conventions appropriate
599 for NMake or Jom. Ignored unless another option causes Clang to emit a
602 When Clang emits a dependency file (e.g., you supplied the -M option)
603 most filenames can be written to the file without any special formatting.
604 Different Make tools will treat different sets of characters as "special"
605 and use different conventions for telling the Make tool that the character
606 is actually part of the filename. Normally Clang uses backslash to "escape"
607 a special character, which is the convention used by GNU Make. The -MV
608 option tells Clang to put double-quotes around the entire filename, which
609 is the convention used by NMake and Jom.
612 Language and Target-Independent Features
613 ========================================
615 Controlling Errors and Warnings
616 -------------------------------
618 Clang provides a number of ways to control which code constructs cause
619 it to emit errors and warning messages, and how they are displayed to
622 Controlling How Clang Displays Diagnostics
623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
625 When Clang emits a diagnostic, it includes rich information in the
626 output, and gives you fine-grain control over which information is
627 printed. Clang has the ability to print this information, and these are
628 the options that control it:
630 #. A file/line/column indicator that shows exactly where the diagnostic
631 occurs in your code [:ref:`-fshow-column <opt_fshow-column>`,
632 :ref:`-fshow-source-location <opt_fshow-source-location>`].
633 #. A categorization of the diagnostic as a note, warning, error, or
635 #. A text string that describes what the problem is.
636 #. An option that indicates how to control the diagnostic (for
637 diagnostics that support it)
638 [:ref:`-fdiagnostics-show-option <opt_fdiagnostics-show-option>`].
639 #. A :ref:`high-level category <diagnostics_categories>` for the diagnostic
640 for clients that want to group diagnostics by class (for diagnostics
642 [:ref:`-fdiagnostics-show-category <opt_fdiagnostics-show-category>`].
643 #. The line of source code that the issue occurs on, along with a caret
644 and ranges that indicate the important locations
645 [:ref:`-fcaret-diagnostics <opt_fcaret-diagnostics>`].
646 #. "FixIt" information, which is a concise explanation of how to fix the
647 problem (when Clang is certain it knows)
648 [:ref:`-fdiagnostics-fixit-info <opt_fdiagnostics-fixit-info>`].
649 #. A machine-parsable representation of the ranges involved (off by
651 [:ref:`-fdiagnostics-print-source-range-info <opt_fdiagnostics-print-source-range-info>`].
653 For more information please see :ref:`Formatting of
654 Diagnostics <cl_diag_formatting>`.
659 All diagnostics are mapped into one of these 6 classes:
668 .. _diagnostics_categories:
670 Diagnostic Categories
671 ^^^^^^^^^^^^^^^^^^^^^
673 Though not shown by default, diagnostics may each be associated with a
674 high-level category. This category is intended to make it possible to
675 triage builds that produce a large number of errors or warnings in a
678 Categories are not shown by default, but they can be turned on with the
679 :ref:`-fdiagnostics-show-category <opt_fdiagnostics-show-category>` option.
680 When set to "``name``", the category is printed textually in the
681 diagnostic output. When it is set to "``id``", a category number is
682 printed. The mapping of category names to category id's can be obtained
683 by running '``clang --print-diagnostic-categories``'.
685 Controlling Diagnostics via Command Line Flags
686 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
688 TODO: -W flags, -pedantic, etc
690 .. _pragma_gcc_diagnostic:
692 Controlling Diagnostics via Pragmas
693 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
695 Clang can also control what diagnostics are enabled through the use of
696 pragmas in the source code. This is useful for turning off specific
697 warnings in a section of source code. Clang supports GCC's pragma for
698 compatibility with existing source code, as well as several extensions.
700 The pragma may control any warning that can be used from the command
701 line. Warnings may be set to ignored, warning, error, or fatal. The
702 following example code will tell Clang or GCC to ignore the -Wall
707 #pragma GCC diagnostic ignored "-Wall"
709 In addition to all of the functionality provided by GCC's pragma, Clang
710 also allows you to push and pop the current warning state. This is
711 particularly useful when writing a header file that will be compiled by
712 other people, because you don't know what warning flags they build with.
714 In the below example :option:`-Wmultichar` is ignored for only a single line of
715 code, after which the diagnostics return to whatever state had previously
720 #pragma clang diagnostic push
721 #pragma clang diagnostic ignored "-Wmultichar"
723 char b = 'df'; // no warning.
725 #pragma clang diagnostic pop
727 The push and pop pragmas will save and restore the full diagnostic state
728 of the compiler, regardless of how it was set. That means that it is
729 possible to use push and pop around GCC compatible diagnostics and Clang
730 will push and pop them appropriately, while GCC will ignore the pushes
731 and pops as unknown pragmas. It should be noted that while Clang
732 supports the GCC pragma, Clang and GCC do not support the exact same set
733 of warnings, so even when using GCC compatible #pragmas there is no
734 guarantee that they will have identical behaviour on both compilers.
736 In addition to controlling warnings and errors generated by the compiler, it is
737 possible to generate custom warning and error messages through the following
742 // The following will produce warning messages
743 #pragma message "some diagnostic message"
744 #pragma GCC warning "TODO: replace deprecated feature"
746 // The following will produce an error message
747 #pragma GCC error "Not supported"
749 These pragmas operate similarly to the ``#warning`` and ``#error`` preprocessor
750 directives, except that they may also be embedded into preprocessor macros via
751 the C99 ``_Pragma`` operator, for example:
756 #define DEFER(M,...) M(__VA_ARGS__)
757 #define CUSTOM_ERROR(X) _Pragma(STR(GCC error(X " at line " DEFER(STR,__LINE__))))
759 CUSTOM_ERROR("Feature not available");
761 Controlling Diagnostics in System Headers
762 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
764 Warnings are suppressed when they occur in system headers. By default,
765 an included file is treated as a system header if it is found in an
766 include path specified by ``-isystem``, but this can be overridden in
769 The ``system_header`` pragma can be used to mark the current file as
770 being a system header. No warnings will be produced from the location of
771 the pragma onwards within the same file.
775 char a = 'xy'; // warning
777 #pragma clang system_header
779 char b = 'ab'; // no warning
781 The :option:`--system-header-prefix=` and :option:`--no-system-header-prefix=`
782 command-line arguments can be used to override whether subsets of an include
783 path are treated as system headers. When the name in a ``#include`` directive
784 is found within a header search path and starts with a system prefix, the
785 header is treated as a system header. The last prefix on the
786 command-line which matches the specified header name takes precedence.
789 .. code-block:: console
791 $ clang -Ifoo -isystem bar --system-header-prefix=x/ \
792 --no-system-header-prefix=x/y/
794 Here, ``#include "x/a.h"`` is treated as including a system header, even
795 if the header is found in ``foo``, and ``#include "x/y/b.h"`` is treated
796 as not including a system header, even if the header is found in
799 A ``#include`` directive which finds a file relative to the current
800 directory is treated as including a system header if the including file
801 is treated as a system header.
803 .. _diagnostics_enable_everything:
805 Enabling All Diagnostics
806 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
808 In addition to the traditional ``-W`` flags, one can enable **all**
809 diagnostics by passing :option:`-Weverything`. This works as expected
811 :option:`-Werror`, and also includes the warnings from :option:`-pedantic`.
813 Note that when combined with :option:`-w` (which disables all warnings), that
816 Controlling Static Analyzer Diagnostics
817 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
819 While not strictly part of the compiler, the diagnostics from Clang's
820 `static analyzer <http://clang-analyzer.llvm.org>`_ can also be
821 influenced by the user via changes to the source code. See the available
822 `annotations <http://clang-analyzer.llvm.org/annotations.html>`_ and the
824 page <http://clang-analyzer.llvm.org/faq.html#exclude_code>`_ for more
827 .. _usersmanual-precompiled-headers:
832 `Precompiled headers <http://en.wikipedia.org/wiki/Precompiled_header>`__
833 are a general approach employed by many compilers to reduce compilation
834 time. The underlying motivation of the approach is that it is common for
835 the same (and often large) header files to be included by multiple
836 source files. Consequently, compile times can often be greatly improved
837 by caching some of the (redundant) work done by a compiler to process
838 headers. Precompiled header files, which represent one of many ways to
839 implement this optimization, are literally files that represent an
840 on-disk cache that contains the vital information necessary to reduce
841 some of the work needed to process a corresponding header file. While
842 details of precompiled headers vary between compilers, precompiled
843 headers have been shown to be highly effective at speeding up program
844 compilation on systems with very large system headers (e.g., Mac OS X).
846 Generating a PCH File
847 ^^^^^^^^^^^^^^^^^^^^^
849 To generate a PCH file using Clang, one invokes Clang with the
850 :option:`-x <language>-header` option. This mirrors the interface in GCC
851 for generating PCH files:
853 .. code-block:: console
855 $ gcc -x c-header test.h -o test.h.gch
856 $ clang -x c-header test.h -o test.h.pch
861 A PCH file can then be used as a prefix header when a :option:`-include`
862 option is passed to ``clang``:
864 .. code-block:: console
866 $ clang -include test.h test.c -o test
868 The ``clang`` driver will first check if a PCH file for ``test.h`` is
869 available; if so, the contents of ``test.h`` (and the files it includes)
870 will be processed from the PCH file. Otherwise, Clang falls back to
871 directly processing the content of ``test.h``. This mirrors the behavior
876 Clang does *not* automatically use PCH files for headers that are directly
877 included within a source file. For example:
879 .. code-block:: console
881 $ clang -x c-header test.h -o test.h.pch
884 $ clang test.c -o test
886 In this example, ``clang`` will not automatically use the PCH file for
887 ``test.h`` since ``test.h`` was included directly in the source file and not
888 specified on the command line using :option:`-include`.
890 Relocatable PCH Files
891 ^^^^^^^^^^^^^^^^^^^^^
893 It is sometimes necessary to build a precompiled header from headers
894 that are not yet in their final, installed locations. For example, one
895 might build a precompiled header within the build tree that is then
896 meant to be installed alongside the headers. Clang permits the creation
897 of "relocatable" precompiled headers, which are built with a given path
898 (into the build directory) and can later be used from an installed
901 To build a relocatable precompiled header, place your headers into a
902 subdirectory whose structure mimics the installed location. For example,
903 if you want to build a precompiled header for the header ``mylib.h``
904 that will be installed into ``/usr/include``, create a subdirectory
905 ``build/usr/include`` and place the header ``mylib.h`` into that
906 subdirectory. If ``mylib.h`` depends on other headers, then they can be
907 stored within ``build/usr/include`` in a way that mimics the installed
910 Building a relocatable precompiled header requires two additional
911 arguments. First, pass the ``--relocatable-pch`` flag to indicate that
912 the resulting PCH file should be relocatable. Second, pass
913 :option:`-isysroot /path/to/build`, which makes all includes for your library
914 relative to the build directory. For example:
916 .. code-block:: console
918 # clang -x c-header --relocatable-pch -isysroot /path/to/build /path/to/build/mylib.h mylib.h.pch
920 When loading the relocatable PCH file, the various headers used in the
921 PCH file are found from the system header root. For example, ``mylib.h``
922 can be found in ``/usr/include/mylib.h``. If the headers are installed
923 in some other system root, the :option:`-isysroot` option can be used provide
924 a different system root from which the headers will be based. For
925 example, :option:`-isysroot /Developer/SDKs/MacOSX10.4u.sdk` will look for
926 ``mylib.h`` in ``/Developer/SDKs/MacOSX10.4u.sdk/usr/include/mylib.h``.
928 Relocatable precompiled headers are intended to be used in a limited
929 number of cases where the compilation environment is tightly controlled
930 and the precompiled header cannot be generated after headers have been
933 .. _controlling-code-generation:
935 Controlling Code Generation
936 ---------------------------
938 Clang provides a number of ways to control code generation. The options
941 **-f[no-]sanitize=check1,check2,...**
942 Turn on runtime checks for various forms of undefined or suspicious
945 This option controls whether Clang adds runtime checks for various
946 forms of undefined or suspicious behavior, and is disabled by
947 default. If a check fails, a diagnostic message is produced at
948 runtime explaining the problem. The main checks are:
950 - .. _opt_fsanitize_address:
952 ``-fsanitize=address``:
953 :doc:`AddressSanitizer`, a memory error
955 - ``-fsanitize=integer``: Enables checks for undefined or
956 suspicious integer behavior.
957 - .. _opt_fsanitize_thread:
959 ``-fsanitize=thread``: :doc:`ThreadSanitizer`, a data race detector.
960 - .. _opt_fsanitize_memory:
962 ``-fsanitize=memory``: :doc:`MemorySanitizer`,
963 an *experimental* detector of uninitialized reads. Not ready for
965 - .. _opt_fsanitize_undefined:
967 ``-fsanitize=undefined``: Fast and compatible undefined behavior
968 checker. Enables the undefined behavior checks that have small
969 runtime cost and no impact on address space layout or ABI. This
970 includes all of the checks listed below other than
971 ``unsigned-integer-overflow``.
973 - ``-fsanitize=undefined-trap``: This includes all sanitizers
974 included by ``-fsanitize=undefined``, except those that require
975 runtime support. This group of sanitizers is intended to be
976 used in conjunction with the ``-fsanitize-undefined-trap-on-error``
977 flag. This includes all of the checks listed below other than
978 ``unsigned-integer-overflow`` and ``vptr``.
979 - ``-fsanitize=dataflow``: :doc:`DataFlowSanitizer`, a general data
981 - ``-fsanitize=cfi``: :doc:`control flow integrity <ControlFlowIntegrity>`
982 checks. Implies ``-flto``.
984 The following more fine-grained checks are also available:
986 - ``-fsanitize=alignment``: Use of a misaligned pointer or creation
987 of a misaligned reference.
988 - ``-fsanitize=bool``: Load of a ``bool`` value which is neither
989 ``true`` nor ``false``.
990 - ``-fsanitize=bounds``: Out of bounds array indexing, in cases
991 where the array bound can be statically determined.
992 - ``-fsanitize=cfi-cast-strict``: Enables :ref:`strict cast checks
994 - ``-fsanitize=cfi-derived-cast``: Base-to-derived cast to the wrong
995 dynamic type. Implies ``-flto``.
996 - ``-fsanitize=cfi-unrelated-cast``: Cast from ``void*`` or another
997 unrelated type to the wrong dynamic type. Implies ``-flto``.
998 - ``-fsanitize=cfi-nvcall``: Non-virtual call via an object whose vptr is of
999 the wrong dynamic type. Implies ``-flto``.
1000 - ``-fsanitize=cfi-vcall``: Virtual call via an object whose vptr is of the
1001 wrong dynamic type. Implies ``-flto``.
1002 - ``-fsanitize=enum``: Load of a value of an enumerated type which
1003 is not in the range of representable values for that enumerated
1005 - ``-fsanitize=float-cast-overflow``: Conversion to, from, or
1006 between floating-point types which would overflow the
1008 - ``-fsanitize=float-divide-by-zero``: Floating point division by
1010 - ``-fsanitize=function``: Indirect call of a function through a
1011 function pointer of the wrong type (Linux, C++ and x86/x86_64 only).
1012 - ``-fsanitize=integer-divide-by-zero``: Integer division by zero.
1013 - ``-fsanitize=nonnull-attribute``: Passing null pointer as a function
1014 parameter which is declared to never be null.
1015 - ``-fsanitize=null``: Use of a null pointer or creation of a null
1017 - ``-fsanitize=object-size``: An attempt to use bytes which the
1018 optimizer can determine are not part of the object being
1019 accessed. The sizes of objects are determined using
1020 ``__builtin_object_size``, and consequently may be able to detect
1021 more problems at higher optimization levels.
1022 - ``-fsanitize=return``: In C++, reaching the end of a
1023 value-returning function without returning a value.
1024 - ``-fsanitize=returns-nonnull-attribute``: Returning null pointer
1025 from a function which is declared to never return null.
1026 - ``-fsanitize=shift``: Shift operators where the amount shifted is
1027 greater or equal to the promoted bit-width of the left hand side
1028 or less than zero, or where the left hand side is negative. For a
1029 signed left shift, also checks for signed overflow in C, and for
1030 unsigned overflow in C++. You can use ``-fsanitize=shift-base`` or
1031 ``-fsanitize=shift-exponent`` to check only left-hand side or
1032 right-hand side of shift operation, respectively.
1033 - ``-fsanitize=signed-integer-overflow``: Signed integer overflow,
1034 including all the checks added by ``-ftrapv``, and checking for
1035 overflow in signed division (``INT_MIN / -1``).
1036 - ``-fsanitize=unreachable``: If control flow reaches
1037 ``__builtin_unreachable``.
1038 - ``-fsanitize=unsigned-integer-overflow``: Unsigned integer
1040 - ``-fsanitize=vla-bound``: A variable-length array whose bound
1041 does not evaluate to a positive value.
1042 - ``-fsanitize=vptr``: Use of an object whose vptr indicates that
1043 it is of the wrong dynamic type, or that its lifetime has not
1044 begun or has ended. Incompatible with ``-fno-rtti``.
1046 You can turn off or modify checks for certain source files, functions
1047 or even variables by providing a special file:
1049 - ``-fsanitize-blacklist=/path/to/blacklist/file``: disable or modify
1050 sanitizer checks for objects listed in the file. See
1051 :doc:`SanitizerSpecialCaseList` for file format description.
1052 - ``-fno-sanitize-blacklist``: don't use blacklist file, if it was
1053 specified earlier in the command line.
1055 Extra features of MemorySanitizer (require explicit
1056 ``-fsanitize=memory``):
1058 - ``-fsanitize-memory-track-origins[=level]``: Enables origin tracking in
1059 MemorySanitizer. Adds a second section to MemorySanitizer
1060 reports pointing to the heap or stack allocation the
1061 uninitialized bits came from. Slows down execution by additional
1064 Possible values for level are 0 (off), 1, 2 (default). Level 2
1065 adds more sections to MemorySanitizer reports describing the
1066 order of memory stores the uninitialized value went
1067 through. This mode may use extra memory in programs that copy
1068 uninitialized memory a lot.
1070 Extra features of UndefinedBehaviorSanitizer:
1072 - ``-fsanitize-undefined-trap-on-error``: Causes traps to be emitted
1073 rather than calls to runtime libraries when a problem is detected.
1074 This option is intended for use in cases where the sanitizer runtime
1075 cannot be used (for instance, when building libc or a kernel module).
1076 This is only compatible with the sanitizers in the ``undefined-trap``
1079 The ``-fsanitize=`` argument must also be provided when linking, in
1080 order to link to the appropriate runtime library. When using
1081 ``-fsanitize=vptr`` (or a group that includes it, such as
1082 ``-fsanitize=undefined``) with a C++ program, the link must be
1083 performed by ``clang++``, not ``clang``, in order to link against the
1084 C++-specific parts of the runtime library.
1086 It is not possible to combine more than one of the ``-fsanitize=address``,
1087 ``-fsanitize=thread``, and ``-fsanitize=memory`` checkers in the same
1088 program. The ``-fsanitize=undefined`` checks can only be combined with
1089 ``-fsanitize=address``.
1091 **-f[no-]sanitize-recover=check1,check2,...**
1093 Controls which checks enabled by ``-fsanitize=`` flag are non-fatal.
1094 If the check is fatal, program will halt after the first error
1095 of this kind is detected and error report is printed.
1097 By default, non-fatal checks are those enabled by UndefinedBehaviorSanitizer,
1098 except for ``-fsanitize=return`` and ``-fsanitize=unreachable``. Some
1099 sanitizers (e.g. :doc:`AddressSanitizer`) may not support recovery,
1100 and always crash the program after the issue is detected.
1102 **-f[no-]sanitize-coverage=[type,features,...]**
1104 Enable simple code coverage in addition to certain sanitizers.
1105 See :doc:`SanitizerCoverage` for more details.
1107 .. option:: -fno-assume-sane-operator-new
1109 Don't assume that the C++'s new operator is sane.
1111 This option tells the compiler to do not assume that C++'s global
1112 new operator will always return a pointer that does not alias any
1113 other pointer when the function returns.
1115 .. option:: -ftrap-function=[name]
1117 Instruct code generator to emit a function call to the specified
1118 function name for ``__builtin_trap()``.
1120 LLVM code generator translates ``__builtin_trap()`` to a trap
1121 instruction if it is supported by the target ISA. Otherwise, the
1122 builtin is translated into a call to ``abort``. If this option is
1123 set, then the code generator will always lower the builtin to a call
1124 to the specified function regardless of whether the target ISA has a
1125 trap instruction. This option is useful for environments (e.g.
1126 deeply embedded) where a trap cannot be properly handled, or when
1127 some custom behavior is desired.
1129 .. option:: -ftls-model=[model]
1131 Select which TLS model to use.
1133 Valid values are: ``global-dynamic``, ``local-dynamic``,
1134 ``initial-exec`` and ``local-exec``. The default value is
1135 ``global-dynamic``. The compiler may use a different model if the
1136 selected model is not supported by the target, or if a more
1137 efficient model can be used. The TLS model can be overridden per
1138 variable using the ``tls_model`` attribute.
1140 .. option:: -mhwdiv=[values]
1142 Select the ARM modes (arm or thumb) that support hardware division
1145 Valid values are: ``arm``, ``thumb`` and ``arm,thumb``.
1146 This option is used to indicate which mode (arm or thumb) supports
1147 hardware division instructions. This only applies to the ARM
1150 .. option:: -m[no-]crc
1152 Enable or disable CRC instructions.
1154 This option is used to indicate whether CRC instructions are to
1155 be generated. This only applies to the ARM architecture.
1157 CRC instructions are enabled by default on ARMv8.
1159 .. option:: -mgeneral-regs-only
1161 Generate code which only uses the general purpose registers.
1163 This option restricts the generated code to use general registers
1164 only. This only applies to the AArch64 architecture.
1166 **-f[no-]max-unknown-pointer-align=[number]**
1167 Instruct the code generator to not enforce a higher alignment than the given
1168 number (of bytes) when accessing memory via an opaque pointer or reference.
1169 This cap is ignored when directly accessing a variable or when the pointee
1170 type has an explicit “aligned” attribute.
1172 The value should usually be determined by the properties of the system allocator.
1173 Some builtin types, especially vector types, have very high natural alignments;
1174 when working with values of those types, Clang usually wants to use instructions
1175 that take advantage of that alignment. However, many system allocators do
1176 not promise to return memory that is more than 8-byte or 16-byte-aligned. Use
1177 this option to limit the alignment that the compiler can assume for an arbitrary
1178 pointer, which may point onto the heap.
1180 This option does not affect the ABI alignment of types; the layout of structs and
1181 unions and the value returned by the alignof operator remain the same.
1183 This option can be overridden on a case-by-case basis by putting an explicit
1184 “aligned” alignment on a struct, union, or typedef. For example:
1186 .. code-block:: console
1188 #include <immintrin.h>
1189 // Make an aligned typedef of the AVX-512 16-int vector type.
1190 typedef __v16si __aligned_v16si __attribute__((aligned(64)));
1192 void initialize_vector(__aligned_v16si *v) {
1193 // The compiler may assume that ‘v’ is 64-byte aligned, regardless of the
1194 // value of -fmax-unknown-pointer-align.
1198 Profile Guided Optimization
1199 ---------------------------
1201 Profile information enables better optimization. For example, knowing that a
1202 branch is taken very frequently helps the compiler make better decisions when
1203 ordering basic blocks. Knowing that a function ``foo`` is called more
1204 frequently than another function ``bar`` helps the inliner.
1206 Clang supports profile guided optimization with two different kinds of
1207 profiling. A sampling profiler can generate a profile with very low runtime
1208 overhead, or you can build an instrumented version of the code that collects
1209 more detailed profile information. Both kinds of profiles can provide execution
1210 counts for instructions in the code and information on branches taken and
1211 function invocation.
1213 Regardless of which kind of profiling you use, be careful to collect profiles
1214 by running your code with inputs that are representative of the typical
1215 behavior. Code that is not exercised in the profile will be optimized as if it
1216 is unimportant, and the compiler may make poor optimization choices for code
1217 that is disproportionately used while profiling.
1219 Using Sampling Profilers
1220 ^^^^^^^^^^^^^^^^^^^^^^^^
1222 Sampling profilers are used to collect runtime information, such as
1223 hardware counters, while your application executes. They are typically
1224 very efficient and do not incur a large runtime overhead. The
1225 sample data collected by the profiler can be used during compilation
1226 to determine what the most executed areas of the code are.
1228 Using the data from a sample profiler requires some changes in the way
1229 a program is built. Before the compiler can use profiling information,
1230 the code needs to execute under the profiler. The following is the
1231 usual build cycle when using sample profilers for optimization:
1233 1. Build the code with source line table information. You can use all the
1234 usual build flags that you always build your application with. The only
1235 requirement is that you add ``-gline-tables-only`` or ``-g`` to the
1236 command line. This is important for the profiler to be able to map
1237 instructions back to source line locations.
1239 .. code-block:: console
1241 $ clang++ -O2 -gline-tables-only code.cc -o code
1243 2. Run the executable under a sampling profiler. The specific profiler
1244 you use does not really matter, as long as its output can be converted
1245 into the format that the LLVM optimizer understands. Currently, there
1246 exists a conversion tool for the Linux Perf profiler
1247 (https://perf.wiki.kernel.org/), so these examples assume that you
1248 are using Linux Perf to profile your code.
1250 .. code-block:: console
1252 $ perf record -b ./code
1254 Note the use of the ``-b`` flag. This tells Perf to use the Last Branch
1255 Record (LBR) to record call chains. While this is not strictly required,
1256 it provides better call information, which improves the accuracy of
1259 3. Convert the collected profile data to LLVM's sample profile format.
1260 This is currently supported via the AutoFDO converter ``create_llvm_prof``.
1261 It is available at http://github.com/google/autofdo. Once built and
1262 installed, you can convert the ``perf.data`` file to LLVM using
1265 .. code-block:: console
1267 $ create_llvm_prof --binary=./code --out=code.prof
1269 This will read ``perf.data`` and the binary file ``./code`` and emit
1270 the profile data in ``code.prof``. Note that if you ran ``perf``
1271 without the ``-b`` flag, you need to use ``--use_lbr=false`` when
1272 calling ``create_llvm_prof``.
1274 4. Build the code again using the collected profile. This step feeds
1275 the profile back to the optimizers. This should result in a binary
1276 that executes faster than the original one. Note that you are not
1277 required to build the code with the exact same arguments that you
1278 used in the first step. The only requirement is that you build the code
1279 with ``-gline-tables-only`` and ``-fprofile-sample-use``.
1281 .. code-block:: console
1283 $ clang++ -O2 -gline-tables-only -fprofile-sample-use=code.prof code.cc -o code
1286 Sample Profile Format
1287 """""""""""""""""""""
1289 If you are not using Linux Perf to collect profiles, you will need to
1290 write a conversion tool from your profiler to LLVM's format. This section
1291 explains the file format expected by the backend.
1293 Sample profiles are written as ASCII text. The file is divided into sections,
1294 which correspond to each of the functions executed at runtime. Each
1295 section has the following format (taken from
1296 https://github.com/google/autofdo/blob/master/profile_writer.h):
1298 .. code-block:: console
1300 function1:total_samples:total_head_samples
1301 offset1[.discriminator]: number_of_samples [fn1:num fn2:num ... ]
1302 offset2[.discriminator]: number_of_samples [fn3:num fn4:num ... ]
1304 offsetN[.discriminator]: number_of_samples [fn5:num fn6:num ... ]
1306 The file may contain blank lines between sections and within a
1307 section. However, the spacing within a single line is fixed. Additional
1308 spaces will result in an error while reading the file.
1310 Function names must be mangled in order for the profile loader to
1311 match them in the current translation unit. The two numbers in the
1312 function header specify how many total samples were accumulated in the
1313 function (first number), and the total number of samples accumulated
1314 in the prologue of the function (second number). This head sample
1315 count provides an indicator of how frequently the function is invoked.
1317 Each sampled line may contain several items. Some are optional (marked
1320 a. Source line offset. This number represents the line number
1321 in the function where the sample was collected. The line number is
1322 always relative to the line where symbol of the function is
1323 defined. So, if the function has its header at line 280, the offset
1324 13 is at line 293 in the file.
1326 Note that this offset should never be a negative number. This could
1327 happen in cases like macros. The debug machinery will register the
1328 line number at the point of macro expansion. So, if the macro was
1329 expanded in a line before the start of the function, the profile
1330 converter should emit a 0 as the offset (this means that the optimizers
1331 will not be able to associate a meaningful weight to the instructions
1334 b. [OPTIONAL] Discriminator. This is used if the sampled program
1335 was compiled with DWARF discriminator support
1336 (http://wiki.dwarfstd.org/index.php?title=Path_Discriminators).
1337 DWARF discriminators are unsigned integer values that allow the
1338 compiler to distinguish between multiple execution paths on the
1339 same source line location.
1341 For example, consider the line of code ``if (cond) foo(); else bar();``.
1342 If the predicate ``cond`` is true 80% of the time, then the edge
1343 into function ``foo`` should be considered to be taken most of the
1344 time. But both calls to ``foo`` and ``bar`` are at the same source
1345 line, so a sample count at that line is not sufficient. The
1346 compiler needs to know which part of that line is taken more
1349 This is what discriminators provide. In this case, the calls to
1350 ``foo`` and ``bar`` will be at the same line, but will have
1351 different discriminator values. This allows the compiler to correctly
1352 set edge weights into ``foo`` and ``bar``.
1354 c. Number of samples. This is an integer quantity representing the
1355 number of samples collected by the profiler at this source
1358 d. [OPTIONAL] Potential call targets and samples. If present, this
1359 line contains a call instruction. This models both direct and
1360 number of samples. For example,
1362 .. code-block:: console
1364 130: 7 foo:3 bar:2 baz:7
1366 The above means that at relative line offset 130 there is a call
1367 instruction that calls one of ``foo()``, ``bar()`` and ``baz()``,
1368 with ``baz()`` being the relatively more frequently called target.
1371 Profiling with Instrumentation
1372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1374 Clang also supports profiling via instrumentation. This requires building a
1375 special instrumented version of the code and has some runtime
1376 overhead during the profiling, but it provides more detailed results than a
1377 sampling profiler. It also provides reproducible results, at least to the
1378 extent that the code behaves consistently across runs.
1380 Here are the steps for using profile guided optimization with
1383 1. Build an instrumented version of the code by compiling and linking with the
1384 ``-fprofile-instr-generate`` option.
1386 .. code-block:: console
1388 $ clang++ -O2 -fprofile-instr-generate code.cc -o code
1390 2. Run the instrumented executable with inputs that reflect the typical usage.
1391 By default, the profile data will be written to a ``default.profraw`` file
1392 in the current directory. You can override that default by setting the
1393 ``LLVM_PROFILE_FILE`` environment variable to specify an alternate file.
1394 Any instance of ``%p`` in that file name will be replaced by the process
1395 ID, so that you can easily distinguish the profile output from multiple
1398 .. code-block:: console
1400 $ LLVM_PROFILE_FILE="code-%p.profraw" ./code
1402 3. Combine profiles from multiple runs and convert the "raw" profile format to
1403 the input expected by clang. Use the ``merge`` command of the llvm-profdata
1406 .. code-block:: console
1408 $ llvm-profdata merge -output=code.profdata code-*.profraw
1410 Note that this step is necessary even when there is only one "raw" profile,
1411 since the merge operation also changes the file format.
1413 4. Build the code again using the ``-fprofile-instr-use`` option to specify the
1414 collected profile data.
1416 .. code-block:: console
1418 $ clang++ -O2 -fprofile-instr-use=code.profdata code.cc -o code
1420 You can repeat step 4 as often as you like without regenerating the
1421 profile. As you make changes to your code, clang may no longer be able to
1422 use the profile data. It will warn you when this happens.
1425 Controlling Size of Debug Information
1426 -------------------------------------
1428 Debug info kind generated by Clang can be set by one of the flags listed
1429 below. If multiple flags are present, the last one is used.
1433 Don't generate any debug info (default).
1435 .. option:: -gline-tables-only
1437 Generate line number tables only.
1439 This kind of debug info allows to obtain stack traces with function names,
1440 file names and line numbers (by such tools as ``gdb`` or ``addr2line``). It
1441 doesn't contain any other data (e.g. description of local variables or
1442 function parameters).
1444 .. option:: -fstandalone-debug
1446 Clang supports a number of optimizations to reduce the size of debug
1447 information in the binary. They work based on the assumption that
1448 the debug type information can be spread out over multiple
1449 compilation units. For instance, Clang will not emit type
1450 definitions for types that are not needed by a module and could be
1451 replaced with a forward declaration. Further, Clang will only emit
1452 type info for a dynamic C++ class in the module that contains the
1453 vtable for the class.
1455 The **-fstandalone-debug** option turns off these optimizations.
1456 This is useful when working with 3rd-party libraries that don't come
1457 with debug information. Note that Clang will never emit type
1458 information for types that are not referenced at all by the program.
1460 .. option:: -fno-standalone-debug
1462 On Darwin **-fstandalone-debug** is enabled by default. The
1463 **-fno-standalone-debug** option can be used to get to turn on the
1464 vtable-based optimization described above.
1468 Generate complete debug info.
1470 Comment Parsing Options
1471 -----------------------
1473 Clang parses Doxygen and non-Doxygen style documentation comments and attaches
1474 them to the appropriate declaration nodes. By default, it only parses
1475 Doxygen-style comments and ignores ordinary comments starting with ``//`` and
1478 .. option:: -Wdocumentation
1480 Emit warnings about use of documentation comments. This warning group is off
1483 This includes checking that ``\param`` commands name parameters that actually
1484 present in the function signature, checking that ``\returns`` is used only on
1485 functions that actually return a value etc.
1487 .. option:: -Wno-documentation-unknown-command
1489 Don't warn when encountering an unknown Doxygen command.
1491 .. option:: -fparse-all-comments
1493 Parse all comments as documentation comments (including ordinary comments
1494 starting with ``//`` and ``/*``).
1496 .. option:: -fcomment-block-commands=[commands]
1498 Define custom documentation commands as block commands. This allows Clang to
1499 construct the correct AST for these custom commands, and silences warnings
1500 about unknown commands. Several commands must be separated by a comma
1501 *without trailing space*; e.g. ``-fcomment-block-commands=foo,bar`` defines
1502 custom commands ``\foo`` and ``\bar``.
1504 It is also possible to use ``-fcomment-block-commands`` several times; e.g.
1505 ``-fcomment-block-commands=foo -fcomment-block-commands=bar`` does the same
1513 The support for standard C in clang is feature-complete except for the
1514 C99 floating-point pragmas.
1516 Extensions supported by clang
1517 -----------------------------
1519 See :doc:`LanguageExtensions`.
1521 Differences between various standard modes
1522 ------------------------------------------
1524 clang supports the -std option, which changes what language mode clang
1525 uses. The supported modes for C are c89, gnu89, c94, c99, gnu99, c11,
1526 gnu11, and various aliases for those modes. If no -std option is
1527 specified, clang defaults to gnu11 mode. Many C99 and C11 features are
1528 supported in earlier modes as a conforming extension, with a warning. Use
1529 ``-pedantic-errors`` to request an error if a feature from a later standard
1530 revision is used in an earlier mode.
1532 Differences between all ``c*`` and ``gnu*`` modes:
1534 - ``c*`` modes define "``__STRICT_ANSI__``".
1535 - Target-specific defines not prefixed by underscores, like "linux",
1536 are defined in ``gnu*`` modes.
1537 - Trigraphs default to being off in ``gnu*`` modes; they can be enabled by
1538 the -trigraphs option.
1539 - The parser recognizes "asm" and "typeof" as keywords in ``gnu*`` modes;
1540 the variants "``__asm__``" and "``__typeof__``" are recognized in all
1542 - The Apple "blocks" extension is recognized by default in ``gnu*`` modes
1543 on some platforms; it can be enabled in any mode with the "-fblocks"
1545 - Arrays that are VLA's according to the standard, but which can be
1546 constant folded by the frontend are treated as fixed size arrays.
1547 This occurs for things like "int X[(1, 2)];", which is technically a
1548 VLA. ``c*`` modes are strictly compliant and treat these as VLAs.
1550 Differences between ``*89`` and ``*99`` modes:
1552 - The ``*99`` modes default to implementing "inline" as specified in C99,
1553 while the ``*89`` modes implement the GNU version. This can be
1554 overridden for individual functions with the ``__gnu_inline__``
1556 - Digraphs are not recognized in c89 mode.
1557 - The scope of names defined inside a "for", "if", "switch", "while",
1558 or "do" statement is different. (example: "``if ((struct x {int
1560 - ``__STDC_VERSION__`` is not defined in ``*89`` modes.
1561 - "inline" is not recognized as a keyword in c89 mode.
1562 - "restrict" is not recognized as a keyword in ``*89`` modes.
1563 - Commas are allowed in integer constant expressions in ``*99`` modes.
1564 - Arrays which are not lvalues are not implicitly promoted to pointers
1566 - Some warnings are different.
1568 Differences between ``*99`` and ``*11`` modes:
1570 - Warnings for use of C11 features are disabled.
1571 - ``__STDC_VERSION__`` is defined to ``201112L`` rather than ``199901L``.
1573 c94 mode is identical to c89 mode except that digraphs are enabled in
1574 c94 mode (FIXME: And ``__STDC_VERSION__`` should be defined!).
1576 GCC extensions not implemented yet
1577 ----------------------------------
1579 clang tries to be compatible with gcc as much as possible, but some gcc
1580 extensions are not implemented yet:
1582 - clang does not support #pragma weak (`bug
1583 3679 <http://llvm.org/bugs/show_bug.cgi?id=3679>`_). Due to the uses
1584 described in the bug, this is likely to be implemented at some point,
1586 - clang does not support decimal floating point types (``_Decimal32`` and
1587 friends) or fixed-point types (``_Fract`` and friends); nobody has
1588 expressed interest in these features yet, so it's hard to say when
1589 they will be implemented.
1590 - clang does not support nested functions; this is a complex feature
1591 which is infrequently used, so it is unlikely to be implemented
1592 anytime soon. In C++11 it can be emulated by assigning lambda
1593 functions to local variables, e.g:
1597 auto const local_function = [&](int parameter) {
1603 - clang does not support global register variables; this is unlikely to
1604 be implemented soon because it requires additional LLVM backend
1606 - clang does not support static initialization of flexible array
1607 members. This appears to be a rarely used extension, but could be
1608 implemented pending user demand.
1609 - clang does not support
1610 ``__builtin_va_arg_pack``/``__builtin_va_arg_pack_len``. This is
1611 used rarely, but in some potentially interesting places, like the
1612 glibc headers, so it may be implemented pending user demand. Note
1613 that because clang pretends to be like GCC 4.2, and this extension
1614 was introduced in 4.3, the glibc headers will not try to use this
1615 extension with clang at the moment.
1616 - clang does not support the gcc extension for forward-declaring
1617 function parameters; this has not shown up in any real-world code
1618 yet, though, so it might never be implemented.
1620 This is not a complete list; if you find an unsupported extension
1621 missing from this list, please send an e-mail to cfe-dev. This list
1622 currently excludes C++; see :ref:`C++ Language Features <cxx>`. Also, this
1623 list does not include bugs in mostly-implemented features; please see
1625 tracker <http://llvm.org/bugs/buglist.cgi?quicksearch=product%3Aclang+component%3A-New%2BBugs%2CAST%2CBasic%2CDriver%2CHeaders%2CLLVM%2BCodeGen%2Cparser%2Cpreprocessor%2CSemantic%2BAnalyzer>`_
1626 for known existing bugs (FIXME: Is there a section for bug-reporting
1627 guidelines somewhere?).
1629 Intentionally unsupported GCC extensions
1630 ----------------------------------------
1632 - clang does not support the gcc extension that allows variable-length
1633 arrays in structures. This is for a few reasons: one, it is tricky to
1634 implement, two, the extension is completely undocumented, and three,
1635 the extension appears to be rarely used. Note that clang *does*
1636 support flexible array members (arrays with a zero or unspecified
1637 size at the end of a structure).
1638 - clang does not have an equivalent to gcc's "fold"; this means that
1639 clang doesn't accept some constructs gcc might accept in contexts
1640 where a constant expression is required, like "x-x" where x is a
1642 - clang does not support ``__builtin_apply`` and friends; this extension
1643 is extremely obscure and difficult to implement reliably.
1647 Microsoft extensions
1648 --------------------
1650 clang has some experimental support for extensions from Microsoft Visual
1651 C++; to enable it, use the ``-fms-extensions`` command-line option. This is
1652 the default for Windows targets. Note that the support is incomplete.
1653 Some constructs such as ``dllexport`` on classes are ignored with a warning,
1654 and others such as `Microsoft IDL annotations
1655 <http://msdn.microsoft.com/en-us/library/8tesw2eh.aspx>`_ are silently
1658 clang has a ``-fms-compatibility`` flag that makes clang accept enough
1659 invalid C++ to be able to parse most Microsoft headers. For example, it
1660 allows `unqualified lookup of dependent base class members
1661 <http://clang.llvm.org/compatibility.html#dep_lookup_bases>`_, which is
1662 a common compatibility issue with clang. This flag is enabled by default
1663 for Windows targets.
1665 ``-fdelayed-template-parsing`` lets clang delay parsing of function template
1666 definitions until the end of a translation unit. This flag is enabled by
1667 default for Windows targets.
1669 - clang allows setting ``_MSC_VER`` with ``-fmsc-version=``. It defaults to
1670 1700 which is the same as Visual C/C++ 2012. Any number is supported
1671 and can greatly affect what Windows SDK and c++stdlib headers clang
1673 - clang does not support the Microsoft extension where anonymous record
1674 members can be declared using user defined typedefs.
1675 - clang supports the Microsoft ``#pragma pack`` feature for controlling
1676 record layout. GCC also contains support for this feature, however
1677 where MSVC and GCC are incompatible clang follows the MSVC
1679 - clang supports the Microsoft ``#pragma comment(lib, "foo.lib")`` feature for
1680 automatically linking against the specified library. Currently this feature
1681 only works with the Visual C++ linker.
1682 - clang supports the Microsoft ``#pragma comment(linker, "/flag:foo")`` feature
1683 for adding linker flags to COFF object files. The user is responsible for
1684 ensuring that the linker understands the flags.
1685 - clang defaults to C++11 for Windows targets.
1689 C++ Language Features
1690 =====================
1692 clang fully implements all of standard C++98 except for exported
1693 templates (which were removed in C++11), and all of standard C++11
1694 and the current draft standard for C++1y.
1696 Controlling implementation limits
1697 ---------------------------------
1699 .. option:: -fbracket-depth=N
1701 Sets the limit for nested parentheses, brackets, and braces to N. The
1704 .. option:: -fconstexpr-depth=N
1706 Sets the limit for recursive constexpr function invocations to N. The
1709 .. option:: -ftemplate-depth=N
1711 Sets the limit for recursively nested template instantiations to N. The
1714 .. option:: -foperator-arrow-depth=N
1716 Sets the limit for iterative calls to 'operator->' functions to N. The
1721 Objective-C Language Features
1722 =============================
1726 Objective-C++ Language Features
1727 ===============================
1730 .. _target_features:
1732 Target-Specific Features and Limitations
1733 ========================================
1735 CPU Architectures Features and Limitations
1736 ------------------------------------------
1741 The support for X86 (both 32-bit and 64-bit) is considered stable on
1742 Darwin (Mac OS X), Linux, FreeBSD, and Dragonfly BSD: it has been tested
1743 to correctly compile many large C, C++, Objective-C, and Objective-C++
1746 On ``x86_64-mingw32``, passing i128(by value) is incompatible with the
1747 Microsoft x64 calling convention. You might need to tweak
1748 ``WinX86_64ABIInfo::classify()`` in lib/CodeGen/TargetInfo.cpp.
1750 For the X86 target, clang supports the :option:`-m16` command line
1751 argument which enables 16-bit code output. This is broadly similar to
1752 using ``asm(".code16gcc")`` with the GNU toolchain. The generated code
1753 and the ABI remains 32-bit but the assembler emits instructions
1754 appropriate for a CPU running in 16-bit mode, with address-size and
1755 operand-size prefixes to enable 32-bit addressing and operations.
1760 The support for ARM (specifically ARMv6 and ARMv7) is considered stable
1761 on Darwin (iOS): it has been tested to correctly compile many large C,
1762 C++, Objective-C, and Objective-C++ codebases. Clang only supports a
1763 limited number of ARM architectures. It does not yet fully support
1769 The support for PowerPC (especially PowerPC64) is considered stable
1770 on Linux and FreeBSD: it has been tested to correctly compile many
1771 large C and C++ codebases. PowerPC (32bit) is still missing certain
1772 features (e.g. PIC code on ELF platforms).
1777 clang currently contains some support for other architectures (e.g. Sparc);
1778 however, significant pieces of code generation are still missing, and they
1779 haven't undergone significant testing.
1781 clang contains limited support for the MSP430 embedded processor, but
1782 both the clang support and the LLVM backend support are highly
1785 Other platforms are completely unsupported at the moment. Adding the
1786 minimal support needed for parsing and semantic analysis on a new
1787 platform is quite easy; see ``lib/Basic/Targets.cpp`` in the clang source
1788 tree. This level of support is also sufficient for conversion to LLVM IR
1789 for simple programs. Proper support for conversion to LLVM IR requires
1790 adding code to ``lib/CodeGen/CGCall.cpp`` at the moment; this is likely to
1791 change soon, though. Generating assembly requires a suitable LLVM
1794 Operating System Features and Limitations
1795 -----------------------------------------
1800 Thread Sanitizer is not supported.
1805 Clang has experimental support for targeting "Cygming" (Cygwin / MinGW)
1808 See also :ref:`Microsoft Extensions <c_ms>`.
1813 Clang works on Cygwin-1.7.
1818 Clang works on some mingw32 distributions. Clang assumes directories as
1821 - ``C:/mingw/include``
1823 - ``C:/mingw/lib/gcc/mingw32/4.[3-5].0/include/c++``
1825 On MSYS, a few tests might fail.
1830 For 32-bit (i686-w64-mingw32), and 64-bit (x86\_64-w64-mingw32), Clang
1833 - ``GCC versions 4.5.0 to 4.5.3, 4.6.0 to 4.6.2, or 4.7.0 (for the C++ header search path)``
1834 - ``some_directory/bin/gcc.exe``
1835 - ``some_directory/bin/clang.exe``
1836 - ``some_directory/bin/clang++.exe``
1837 - ``some_directory/bin/../include/c++/GCC_version``
1838 - ``some_directory/bin/../include/c++/GCC_version/x86_64-w64-mingw32``
1839 - ``some_directory/bin/../include/c++/GCC_version/i686-w64-mingw32``
1840 - ``some_directory/bin/../include/c++/GCC_version/backward``
1841 - ``some_directory/bin/../x86_64-w64-mingw32/include``
1842 - ``some_directory/bin/../i686-w64-mingw32/include``
1843 - ``some_directory/bin/../include``
1845 This directory layout is standard for any toolchain you will find on the
1846 official `MinGW-w64 website <http://mingw-w64.sourceforge.net>`_.
1848 Clang expects the GCC executable "gcc.exe" compiled for
1849 ``i686-w64-mingw32`` (or ``x86_64-w64-mingw32``) to be present on PATH.
1851 `Some tests might fail <http://llvm.org/bugs/show_bug.cgi?id=9072>`_ on
1852 ``x86_64-w64-mingw32``.
1859 clang-cl is an alternative command-line interface to Clang driver, designed for
1860 compatibility with the Visual C++ compiler, cl.exe.
1862 To enable clang-cl to find system headers, libraries, and the linker when run
1863 from the command-line, it should be executed inside a Visual Studio Native Tools
1864 Command Prompt or a regular Command Prompt where the environment has been set
1865 up using e.g. `vcvars32.bat <http://msdn.microsoft.com/en-us/library/f2ccy3wt.aspx>`_.
1867 clang-cl can also be used from inside Visual Studio by using an LLVM Platform
1870 Command-Line Options
1871 --------------------
1873 To be compatible with cl.exe, clang-cl supports most of the same command-line
1874 options. Those options can start with either ``/`` or ``-``. It also supports
1875 some of Clang's core options, such as the ``-W`` options.
1877 Options that are known to clang-cl, but not currently supported, are ignored
1878 with a warning. For example:
1882 clang-cl.exe: warning: argument unused during compilation: '/Zi'
1884 To suppress warnings about unused arguments, use the ``-Qunused-arguments`` option.
1886 Options that are not known to clang-cl will cause errors. If they are spelled with a
1887 leading ``/``, they will be mistaken for a filename:
1891 clang-cl.exe: error: no such file or directory: '/foobar'
1893 Please `file a bug <http://llvm.org/bugs/enter_bug.cgi?product=clang&component=Driver>`_
1894 for any valid cl.exe flags that clang-cl does not understand.
1896 Execute ``clang-cl /?`` to see a list of supported options:
1900 CL.EXE COMPATIBILITY OPTIONS:
1901 /? Display available options
1902 /arch:<value> Set architecture for code generation
1903 /C Don't discard comments when preprocessing
1905 /D <macro[=value]> Define macro
1906 /EH<value> Exception handling model
1907 /EP Disable linemarker output and preprocess to stdout
1908 /E Preprocess to stdout
1909 /fallback Fall back to cl.exe if clang-cl fails to compile
1910 /FA Output assembly code file during compilation
1911 /Fa<file or directory> Output assembly code to this file during compilation
1912 /Fe<file or directory> Set output executable file or directory (ends in / or \)
1913 /FI <value> Include file before parsing
1914 /Fi<file> Set preprocess output file name
1915 /Fo<file or directory> Set output object file, or directory (ends in / or \)
1916 /GF- Disable string pooling
1917 /GR- Disable emission of RTTI data
1918 /GR Enable emission of RTTI data
1919 /Gw- Don't put each data item in its own section
1920 /Gw Put each data item in its own section
1921 /Gy- Don't put each function in its own section
1922 /Gy Put each function in its own section
1923 /help Display available options
1924 /I <dir> Add directory to include search path
1925 /J Make char type unsigned
1926 /LDd Create debug DLL
1928 /link <options> Forward options to the linker
1929 /MDd Use DLL debug run-time
1930 /MD Use DLL run-time
1931 /MTd Use static debug run-time
1932 /MT Use static run-time
1933 /Ob0 Disable inlining
1934 /Od Disable optimization
1935 /Oi- Disable use of builtin functions
1936 /Oi Enable use of builtin functions
1937 /Os Optimize for size
1938 /Ot Optimize for speed
1939 /Ox Maximum optimization
1940 /Oy- Disable frame pointer omission
1941 /Oy Enable frame pointer omission
1942 /O<n> Optimization level
1943 /P Preprocess to file
1944 /showIncludes Print info about included files to stderr
1945 /TC Treat all source files as C
1946 /Tc <filename> Specify a C source file
1947 /TP Treat all source files as C++
1948 /Tp <filename> Specify a C++ source file
1949 /U <macro> Undefine macro
1950 /vd<value> Control vtordisp placement
1951 /vmb Use a best-case representation method for member pointers
1952 /vmg Use a most-general representation for member pointers
1953 /vmm Set the default most-general representation to multiple inheritance
1954 /vms Set the default most-general representation to single inheritance
1955 /vmv Set the default most-general representation to virtual inheritance
1956 /W0 Disable all warnings
1962 /WX- Do not treat warnings as errors
1963 /WX Treat warnings as errors
1964 /w Disable all warnings
1965 /Zi Enable debug information
1966 /Zp Set the default maximum struct packing alignment to 1
1967 /Zp<value> Specify the default maximum struct packing alignment
1968 /Zs Syntax-check only
1971 -### Print (but do not run) the commands to run for this compilation
1972 -fms-compatibility-version=<value>
1973 Dot-separated value representing the Microsoft compiler version
1974 number to report in _MSC_VER (0 = don't define it (default))
1975 -fmsc-version=<value> Microsoft compiler version number to report in _MSC_VER (0 = don't
1976 define it (default))
1977 -fsanitize-blacklist=<value>
1978 Path to blacklist file for sanitizers
1979 -fsanitize=<check> Enable runtime instrumentation for bug detection: address (memory
1980 errors) | thread (race detection) | undefined (miscellaneous
1982 -mllvm <value> Additional arguments to forward to LLVM's option processing
1983 -Qunused-arguments Don't emit warning for unused driver arguments
1984 --target=<value> Generate code for the given target
1985 -v Show commands to run and use verbose output
1986 -W<warning> Enable the specified warning
1987 -Xclang <arg> Pass <arg> to the clang compiler
1989 The /fallback Option
1990 ^^^^^^^^^^^^^^^^^^^^
1992 When clang-cl is run with the ``/fallback`` option, it will first try to
1993 compile files itself. For any file that it fails to compile, it will fall back
1994 and try to compile the file by invoking cl.exe.
1996 This option is intended to be used as a temporary means to build projects where
1997 clang-cl cannot successfully compile all the files. clang-cl may fail to compile
1998 a file either because it cannot generate code for some C++ feature, or because
1999 it cannot parse some Microsoft language extension.