Fold patches into contrib.
[dragonfly.git] / contrib / gcc-4.1 / BUGS
2 GCC Bugs
4 The latest version of this document is always available at
5 [1]
6 _________________________________________________________________
8Table of Contents
10 * [2]Reporting Bugs
11 + [3]What we need
12 + [4]What we DON'T want
13 + [5]Where to post it
14 + [6]Detailed bug reporting instructions
15 + [7]Detailed bug reporting instructions for GNAT
16 + [8]Detailed bug reporting instructions when using a
17 precompiled header
18 * [9]Frequently Reported Bugs in GCC
19 + [10]C++
20 o [11]Missing features
21 o [12]Bugs fixed in the 3.4 series
22 + [13]Fortran
23 * [14]Non-bugs
24 + [15]General
25 + [16]C
26 + [17]C++
27 o [18]Common problems when upgrading the compiler
28 _________________________________________________________________
30 Reporting Bugs
32 The main purpose of a bug report is to enable us to fix the bug. The
33 most important prerequisite for this is that the report must be
34 complete and self-contained.
36 Before you report a bug, please check the [19]list of well-known bugs
37 and, if possible, try a current development snapshot. If you want to
38 report a bug with versions of GCC before 3.4 we strongly recommend
39 upgrading to the current release first.
41 Before reporting that GCC compiles your code incorrectly, please
42 compile it with gcc -Wall and see whether this shows anything wrong
43 with your code that could be the cause instead of a bug in GCC.
45Summarized bug reporting instructions
47 After this summary, you'll find detailed bug reporting instructions,
48 that explain how to obtain some of the information requested in this
49 summary.
51 What we need
53 Please include in your bug report all of the following items, the
54 first three of which can be obtained from the output of gcc -v:
55 * the exact version of GCC;
56 * the system type;
57 * the options given when GCC was configured/built;
58 * the complete command line that triggers the bug;
59 * the compiler output (error messages, warnings, etc.); and
60 * the preprocessed file (*.i*) that triggers the bug, generated by
61 adding -save-temps to the complete compilation command, or, in the
62 case of a bug report for the GNAT front end, a complete set of
63 source files (see below).
65 What we do not want
67 * A source file that #includes header files that are left out of the
68 bug report (see above)
69 * That source file and a collection of header files.
70 * An attached archive (tar, zip, shar, whatever) containing all (or
71 some :-) of the above.
72 * A code snippet that won't cause the compiler to produce the exact
73 output mentioned in the bug report (e.g., a snippet with just a
74 few lines around the one that apparently triggers the bug, with
75 some pieces replaced with ellipses or comments for extra
76 obfuscation :-)
77 * The location (URL) of the package that failed to build (we won't
78 download it, anyway, since you've already given us what we need to
79 duplicate the bug, haven't you? :-)
80 * An error that occurs only some of the times a certain file is
81 compiled, such that retrying a sufficient number of times results
82 in a successful compilation; this is a symptom of a hardware
83 problem, not of a compiler bug (sorry)
84 * Assembly files (*.s) produced by the compiler, or any binary
85 files, such as object files, executables, core files, or
86 precompiled header files
87 * Duplicate bug reports, or reports of bugs already fixed in the
88 development tree, especially those that have already been reported
89 as fixed last week :-)
90 * Bugs in the assembler, the linker or the C library. These are
91 separate projects, with separate mailing lists and different bug
92 reporting procedures
93 * Bugs in releases or snapshots of GCC not issued by the GNU
94 Project. Report them to whoever provided you with the release
95 * Questions about the correctness or the expected behavior of
96 certain constructs that are not GCC extensions. Ask them in forums
97 dedicated to the discussion of the programming language
99 Where to post it
101 Please submit your bug report directly to the [20]GCC bug database.
102 Alternatively, you can use the gccbug script that mails your bug
103 report to the bug database.
104 Only if all this is absolutely impossible, mail all information to
105 [21]
107Detailed bug reporting instructions
109 Please refer to the [22]next section when reporting bugs in GNAT, the
110 Ada compiler, or to the [23]one after that when reporting bugs that
111 appear when using a precompiled header.
113 In general, all the information we need can be obtained by collecting
114 the command line below, as well as its output and the preprocessed
115 file it generates.
117 gcc -v -save-temps all-your-options source-file
119 The only excuses to not send us the preprocessed sources are (i) if
120 you've found a bug in the preprocessor, (ii) if you've reduced the
121 testcase to a small file that doesn't include any other file or (iii)
122 if the bug appears only when using precompiled headers. If you can't
123 post the preprocessed sources because they're proprietary code, then
124 try to create a small file that triggers the same problem.
126 Since we're supposed to be able to re-create the assembly output
127 (extension .s), you usually should not include it in the bug report,
128 although you may want to post parts of it to point out assembly code
129 you consider to be wrong.
131 Please avoid posting an archive (.tar, .shar or .zip); we generally
132 need just a single file to reproduce the bug (the .i/.ii/.f
133 preprocessed file), and, by storing it in an archive, you're just
134 making our volunteers' jobs harder. Only when your bug report requires
135 multiple source files to be reproduced should you use an archive. This
136 is, for example, the case if you are using INCLUDE directives in
137 Fortran code, which are not processed by the preprocessor, but the
138 compiler. In that case, we need the main file and all INCLUDEd files.
139 In any case, make sure the compiler version, error message, etc, are
140 included in the body of your bug report as plain text, even if
141 needlessly duplicated as part of an archive.
143Detailed bug reporting instructions for GNAT
145 See the [24]previous section for bug reporting instructions for GCC
146 language implementations other than Ada.
148 Bug reports have to contain at least the following information in
149 order to be useful:
150 * the exact version of GCC, as shown by "gcc -v";
151 * the system type;
152 * the options when GCC was configured/built;
153 * the exact command line passed to the gcc program triggering the
154 bug (not just the flags passed to gnatmake, but gnatmake prints
155 the parameters it passed to gcc)
156 * a collection of source files for reproducing the bug, preferably a
157 minimal set (see below);
158 * a description of the expected behavior;
159 * a description of actual behavior.
161 If your code depends on additional source files (usually package
162 specifications), submit the source code for these compilation units in
163 a single file that is acceptable input to gnatchop, i.e. contains no
164 non-Ada text. If the compilation terminated normally, you can usually
165 obtain a list of dependencies using the "gnatls -d main_unit" command,
166 where main_unit is the file name of the main compilation unit (which
167 is also passed to gcc).
169 If you report a bug which causes the compiler to print a bug box,
170 include that bug box in your report, and do not forget to send all the
171 source files listed after the bug box along with your report.
173 If you use gnatprep, be sure to send in preprocessed sources (unless
174 you have to report a bug in gnatprep).
176 When you have checked that your report meets these criteria, please
177 submit it according to our [25]generic instructions. (If you use a
178 mailing list for reporting, please include an "[Ada]" tag in the
179 subject.)
181Detailed bug reporting instructions when using a precompiled header
183 If you're encountering a bug when using a precompiled header, the
184 first thing to do is to delete the precompiled header, and try running
185 the same GCC command again. If the bug happens again, the bug doesn't
186 really involve precompiled headers, please report it without using
187 them by following the instructions [26]above.
189 If you've found a bug while building a precompiled header (for
190 instance, the compiler crashes), follow the usual instructions
191 [27]above.
193 If you've found a real precompiled header bug, what we'll need to
194 reproduce it is the sources to build the precompiled header (as a
195 single .i file), the source file that uses the precompiled header, any
196 other headers that source file includes, and the command lines that
197 you used to build the precompiled header and to use it.
199 Please don't send us the actual precompiled header. It is likely to be
200 very large and we can't use it to reproduce the problem.
201 _________________________________________________________________
203 Frequently Reported Bugs in GCC
205 This is a list of bugs in GCC that are reported very often, but not
206 yet fixed. While it is certainly better to fix bugs instead of
207 documenting them, this document might save people the effort of
208 writing a bug report when the bug is already well-known.
210 There are many reasons why a reported bug doesn't get fixed. It might
211 be difficult to fix, or fixing it might break compatibility. Often,
212 reports get a low priority when there is a simple work-around. In
213 particular, bugs caused by invalid code have a simple work-around: fix
214 the code.
215 _________________________________________________________________
219 Missing features
221 The export keyword is not implemented.
222 Most C++ compilers (G++ included) do not yet implement export,
223 which is necessary for separate compilation of template
224 declarations and definitions. Without export, a template
225 definition must be in scope to be used. The obvious workaround
226 is simply to place all definitions in the header itself.
227 Alternatively, the compilation unit containing template
228 definitions may be included from the header.
230 Bugs fixed in the 3.4 series
232 The following bugs are present up to (and including) GCC 3.3.x. They
233 have been fixed in 3.4.0.
235 Two-stage name-lookup.
236 GCC did not implement two-stage name-lookup (also see
237 [28]below).
239 Covariant return types.
240 GCC did not implement non-trivial covariant returns.
242 Parse errors for "simple" code.
243 GCC gave parse errors for seemingly simple code, such as
245struct A
247 A();
248 A(int);
251struct B
253 B(A);
254 B(A,A);
255 void foo();
258A bar()
260 B b(A(),A(1)); // Variable b, initialized with two temporaries
261 B(A(2)).foo(); // B temporary, initialized with A temporary
262 return (A()); // return A temporary
265 Although being valid code, each of the three lines with a
266 comment was rejected by GCC. The work-arounds for older
267 compiler versions proposed below do not change the semantics of
268 the programs at all.
270 The problem in the first case was that GCC started to parse the
271 declaration of b as a function called b returning B, taking a
272 function returning A as an argument. When it encountered the 1,
273 it was too late. To show the compiler that this should be
274 really an expression, a comma operator with a dummy argument
275 could be used:
277B b((0,A()),A(1));
279 The work-around for simpler cases like the second one was to
280 add additional parentheses around the expressions that were
281 mistaken as declarations:
285 In the third case, however, additional parentheses were causing
286 the problems: The compiler interpreted A() as a function
287 (taking no arguments, returning A), and (A()) as a cast lacking
288 an expression to be casted, hence the parse error. The
289 work-around was to omit the parentheses:
291return A();
293 This problem occurred in a number of variants; in throw
294 statements, people also frequently put the object in
295 parentheses.
296 _________________________________________________________________
6610fcc6 300 G77 bugs are documented in the G77 manual rather than explicitly
301 listed here. Please see [29]Known Causes of Trouble with GNU Fortran
302 in the G77 manual.
303 _________________________________________________________________
305 Non-bugs
307 The following are not actually bugs, but are reported often enough to
308 warrant a mention here.
310 It is not always a bug in the compiler, if code which "worked" in a
311 previous version, is now rejected. Earlier versions of GCC sometimes
312 were less picky about standard conformance and accepted invalid source
313 code. In addition, programming languages themselves change, rendering
314 code invalid that used to be conforming (this holds especially for
315 C++). In either case, you should update your code to match recent
316 language standards.
317 _________________________________________________________________
321 Problems with floating point numbers - the [30]most often reported
322 non-bug.
323 In a number of cases, GCC appears to perform floating point
324 computations incorrectly. For example, the C++ program
326#include <iostream>
328int main()
330 double a = 0.5;
331 double b = 0.01;
332 std::cout << (int)(a / b) << std::endl;
333 return 0;
336 might print 50 on some systems and optimization levels, and 49
337 on others.
339 This is the result of rounding: The computer cannot represent
340 all real numbers exactly, so it has to use approximations. When
341 computing with approximation, the computer needs to round to
342 the nearest representable number.
344 This is not a bug in the compiler, but an inherent limitation
345 of the floating point types. Please study [31]this paper for
346 more information.
347 _________________________________________________________________
351 Increment/decrement operator (++/--) not working as expected - a
352 [32]problem with many variations.
353 The following expressions have unpredictable results:
357i*(++i) /* special case with foo=="operator*" */
358std::cout << i << ++i /* foo(foo(std::cout,i),++i) */
360 since the i without increment can be evaluated before or after
361 ++i.
363 The C and C++ standards have the notion of "sequence points".
364 Everything that happens between two sequence points happens in
365 an unspecified order, but it has to happen after the first and
366 before the second sequence point. The end of a statement and a
367 function call are examples for sequence points, whereas
368 assignments and the comma between function arguments are not.
370 Modifying a value twice between two sequence points as shown in
371 the following examples is even worse:
375(++i)*(++i) /* special case with foo=="operator*" */
376std::cout << ++i << ++i /* foo(foo(std::cout,++i),++i) */
378 This leads to undefined behavior (i.e. the compiler can do
379 anything).
381 Casting does not work as expected when optimization is turned on.
382 This is often caused by a violation of aliasing rules, which
383 are part of the ISO C standard. These rules say that a program
384 is invalid if you try to access a variable through a pointer of
385 an incompatible type. This is happening in the following
386 example where a short is accessed through a pointer to integer
387 (the code assumes 16-bit shorts and 32-bit ints):
389#include <stdio.h>
391int main()
393 short a[2];
395 a[0]=0x1111;
396 a[1]=0x1111;
398 *(int *)a = 0x22222222; /* violation of aliasing rules */
400 printf("%x %x\n", a[0], a[1]);
401 return 0;
404 The aliasing rules were designed to allow compilers more
405 aggressive optimization. Basically, a compiler can assume that
406 all changes to variables happen through pointers or references
407 to variables of a type compatible to the accessed variable.
408 Dereferencing a pointer that violates the aliasing rules
409 results in undefined behavior.
411 In the case above, the compiler may assume that no access
412 through an integer pointer can change the array a, consisting
413 of shorts. Thus, printf may be called with the original values
414 of a[0] and a[1]. What really happens is up to the compiler and
415 may change with architecture and optimization level.
417 Recent versions of GCC turn on the option -fstrict-aliasing
418 (which allows alias-based optimizations) by default with -O2.
419 And some architectures then really print "1111 1111" as result.
420 Without optimization the executable will generate the
421 "expected" output "2222 2222".
423 To disable optimizations based on alias-analysis for faulty
424 legacy code, the option -fno-strict-aliasing can be used as a
425 work-around.
427 The option -Wstrict-aliasing (which is included in -Wall) warns
428 about some - but not all - cases of violation of aliasing rules
429 when -fstrict-aliasing is active.
431 To fix the code above, you can use a union instead of a cast
432 (note that this is a GCC extension which might not work with
433 other compilers):
435#include <stdio.h>
437int main()
439 union
440 {
441 short a[2];
442 int i;
443 } u;
445 u.a[0]=0x1111;
446 u.a[1]=0x1111;
448 u.i = 0x22222222;
450 printf("%x %x\n", u.a[0], u.a[1]);
451 return 0;
454 Now the result will always be "2222 2222".
456 For some more insight into the subject, please have a look at
457 [33]this article.
459 Cannot use preprocessor directive in macro arguments.
460 Let me guess... you used an older version of GCC to compile
461 code that looks something like this:
463 memcpy(dest, src,
464#ifdef PLATFORM1
465 12
467 24
469 );
471 and you got a whole pile of error messages:
473test.c:11: warning: preprocessing directive not recognized within macro arg
474test.c:11: warning: preprocessing directive not recognized within macro arg
475test.c:11: warning: preprocessing directive not recognized within macro arg
476test.c: In function `foo':
477test.c:6: undefined or invalid # directive
478test.c:8: undefined or invalid # directive
479test.c:9: parse error before `24'
480test.c:10: undefined or invalid # directive
482 This is because your C library's <string.h> happens to define
483 memcpy as a macro - which is perfectly legitimate. In recent
484 versions of glibc, for example, printf is among those functions
485 which are implemented as macros.
487 Versions of GCC prior to 3.3 did not allow you to put #ifdef
488 (or any other preprocessor directive) inside the arguments of a
489 macro. The code therefore would not compile.
491 As of GCC 3.3 this kind of construct is always accepted and the
492 preprocessor will probably do what you expect, but see the
493 manual for detailed semantics.
495 However, this kind of code is not portable. It is "undefined
496 behavior" according to the C standard; that means different
497 compilers may do different things with it. It is always
498 possible to rewrite code which uses conditionals inside macros
499 so that it doesn't. You could write the above example
501#ifdef PLATFORM1
502 memcpy(dest, src, 12);
504 memcpy(dest, src, 24);
507 This is a bit more typing, but I personally think it's better
508 style in addition to being more portable.
510 Cannot initialize a static variable with stdin.
511 This has nothing to do with GCC, but people ask us about it a
512 lot. Code like this:
514#include <stdio.h>
516FILE *yyin = stdin;
518 will not compile with GNU libc, because stdin is not a
519 constant. This was done deliberately, to make it easier to
520 maintain binary compatibility when the type FILE needs to be
521 changed. It is surprising for people used to traditional Unix C
522 libraries, but it is permitted by the C standard.
524 This construct commonly occurs in code generated by old
525 versions of lex or yacc. We suggest you try regenerating the
526 parser with a current version of flex or bison, respectively.
527 In your own code, the appropriate fix is to move the
528 initialization to the beginning of main.
530 There is a common misconception that the GCC developers are
531 responsible for GNU libc. These are in fact two entirely
532 separate projects; please check the [34]GNU libc web pages for
533 details.
534 _________________________________________________________________
538 Nested classes can access private members and types of the containing
539 class.
540 Defect report 45 clarifies that nested classes are members of
541 the class they are nested in, and so are granted access to
542 private members of that class.
544 G++ emits two copies of constructors and destructors.
545 In general there are three types of constructors (and
546 destructors).
548 1. The complete object constructor/destructor.
549 2. The base object constructor/destructor.
550 3. The allocating constructor/deallocating destructor.
552 The first two are different, when virtual base classes are
553 involved.
555 Global destructors are not run in the correct order.
556 Global destructors should be run in the reverse order of their
557 constructors completing. In most cases this is the same as the
558 reverse order of constructors starting, but sometimes it is
559 different, and that is important. You need to compile and link
560 your programs with --use-cxa-atexit. We have not turned this
561 switch on by default, as it requires a cxa aware runtime
562 library (libc, glibc, or equivalent).
564 Classes in exception specifiers must be complete types.
565 [15.4]/1 tells you that you cannot have an incomplete type, or
566 pointer to incomplete (other than cv void *) in an exception
567 specification.
569 Exceptions don't work in multithreaded applications.
570 You need to rebuild g++ and libstdc++ with --enable-threads.
571 Remember, C++ exceptions are not like hardware interrupts. You
572 cannot throw an exception in one thread and catch it in
573 another. You cannot throw an exception from a signal handler
574 and catch it in the main thread.
576 Templates, scoping, and digraphs.
577 If you have a class in the global namespace, say named X, and
578 want to give it as a template argument to some other class, say
579 std::vector, then std::vector<::X> fails with a parser error.
581 The reason is that the standard mandates that the sequence <:
582 is treated as if it were the token [. (There are several such
583 combinations of characters - they are called digraphs.)
584 Depending on the version, the compiler then reports a parse
585 error before the character : (the colon before X) or a missing
586 closing bracket ].
588 The simplest way to avoid this is to write std::vector< ::X>,
589 i.e. place a space between the opening angle bracket and the
590 scope operator.
592 Copy constructor access check while initializing a reference.
593 Consider this code:
595class A
598 A();
601 A(const A&); // private copy ctor
604A makeA(void);
605void foo(const A&);
607void bar(void)
609 foo(A()); // error, copy ctor is not accessible
610 foo(makeA()); // error, copy ctor is not accessible
612 A a1;
613 foo(a1); // OK, a1 is a lvalue
616 Starting with GCC 3.4.0, binding an rvalue to a const reference
617 requires an accessible copy constructor. This might be
618 surprising at first sight, especially since most popular
619 compilers do not correctly implement this rule.
621 The C++ Standard says that a temporary object should be created
622 in this context and its contents filled with a copy of the
623 object we are trying to bind to the reference; it also says
624 that the temporary copy can be elided, but the semantic
625 constraints (eg. accessibility) of the copy constructor still
626 have to be checked.
628 For further information, you can consult the following
629 paragraphs of the C++ standard: [dcl.init.ref]/5, bullet 2,
630 sub-bullet 1, and [class.temporary]/2.
632 Common problems when upgrading the compiler
634 ABI changes
636 The C++ application binary interface (ABI) consists of two components:
637 the first defines how the elements of classes are laid out, how
638 functions are called, how function names are mangled, etc; the second
639 part deals with the internals of the objects in libstdc++. Although we
640 strive for a non-changing ABI, so far we have had to modify it with
641 each major release. If you change your compiler to a different major
642 release you must recompile all libraries that contain C++ code. If you
643 fail to do so you risk getting linker errors or malfunctioning
644 programs. Some of our Java support libraries also contain C++ code, so
645 you might want to recompile all libraries to be safe. It should not be
646 necessary to recompile if you have changed to a bug-fix release of the
647 same version of the compiler; bug-fix releases are careful to avoid
648 ABI changes. See also the [35]compatibility section of the GCC manual.
650 Remark: A major release is designated by a change to the first or
651 second component of the two- or three-part version number. A minor
652 (bug-fix) release is designated by a change to the third component
653 only. Thus GCC 3.2 and 3.3 are major releases, while 3.3.1 and 3.3.2
654 are bug-fix releases for GCC 3.3. With the 3.4 series we are
655 introducing a new naming scheme; the first release of this series is
656 3.4.0 instead of just 3.4.
658 Standard conformance
660 With each release, we try to make G++ conform closer to the ISO C++
661 standard (available at [36] We
662 have also implemented some of the core and library defect reports
663 (available at
664 [37] &
665 [38]
666 respectively).
668 Non-conforming legacy code that worked with older versions of GCC may
669 be rejected by more recent compilers. There is no command-line switch
670 to ensure compatibility in general, because trying to parse
671 standard-conforming and old-style code at the same time would render
672 the C++ frontend unmaintainable. However, some non-conforming
673 constructs are allowed when the command-line option -fpermissive is
674 used.
676 Two milestones in standard conformance are GCC 3.0 (including a major
677 overhaul of the standard library) and the 3.4.0 version (with its new
678 C++ parser).
680 New in GCC 3.0
682 * The standard library is much more conformant, and uses the std::
683 namespace (which is now a real namespace, not an alias for ::).
684 * The standard header files for the c library don't end with .h, but
685 begin with c (i.e. <cstdlib> rather than <stdlib.h>). The .h names
686 are still available, but are deprecated.
687 * <strstream> is deprecated, use <sstream> instead.
688 * streambuf::seekoff & streambuf::seekpos are private, instead use
689 streambuf::pubseekoff & streambuf::pubseekpos respectively.
690 * If std::operator << (std::ostream &, long long) doesn't exist, you
691 need to recompile libstdc++ with --enable-long-long.
693 If you get lots of errors about things like cout not being found,
694 you've most likely forgotten to tell the compiler to look in the std::
695 namespace. There are several ways to do this:
696 * Say std::cout at the call. This is the most explicit way of saying
697 what you mean.
698 * Say using std::cout; somewhere before the call. You will need to
699 do this for each function or type you wish to use from the
700 standard library.
701 * Say using namespace std; somewhere before the call. This is the
702 quick-but-dirty fix. This brings the whole of the std:: namespace
703 into scope. Never do this in a header file, as every user of your
704 header file will be affected by this decision.
706 New in GCC 3.4.0
708 The new parser brings a lot of improvements, especially concerning
709 name-lookup.
710 * The "implicit typename" extension got removed (it was already
711 deprecated since GCC 3.1), so that the following code is now
712 rejected, see [14.6]:
714template <typename> struct A
716 typedef int X;
719template <typename T> struct B
721 A<T>::X x; // error
722 typename A<T>::X y; // OK
725B<void> b;
727 * For similar reasons, the following code now requires the template
728 keyword, see [14.2]:
730template <typename> struct A
732 template <int> struct X {};
735template <typename T> struct B
737 typename A<T>::X<0> x; // error
738 typename A<T>::template X<0> y; // OK
741B<void> b;
743 * We now have two-stage name-lookup, so that the following code is
744 rejected, see [14.6]/9:
746template <typename T> int foo()
748 return i; // error
751 * This also affects members of base classes, see [14.6.2]:
753template <typename> struct A
755 int i, j;
758template <typename T> struct B : A<T>
760 int foo1() { return i; } // error
761 int foo2() { return this->i; } // OK
762 int foo3() { return B<T>::i; } // OK
763 int foo4() { return A<T>::i; } // OK
765 using A<T>::j;
766 int foo5() { return j; } // OK
769 In addition to the problems listed above, the manual contains a
770 section on [39]Common Misunderstandings with GNU C++.
774 1.
775 2.
776 3.
777 4.
778 5.
779 6.
780 7.
781 8.
782 9.
783 10.
784 11.
785 12.
786 13.
787 14.
788 15.
789 16.
790 17.
791 18.
792 19.
793 20.
794 21.
795 22.
796 23.
797 24.
798 25.
799 26.
800 27.
801 28.
6610fcc6 802 29.
803 30.
804 31.
805 32.
806 33.
807 34.
808 35.
809 36.
810 37.
811 38.
812 39.