Howard Hinnant
|
abc770690a
|
By changing all of the throw() specs to noexcept I've been able to compile and link all of the source files into a dylib. Prior to this substitution the changed functions were calling __cxa_call_unexpected which isn't implemented yet. However in none of these cases do we actaully want __cxa_call_unexpected to be called. Primative buildit script added.
llvm-svn: 148880
|
2012-01-24 23:42:30 +00:00 |
|
Howard Hinnant
|
7fdfd5d269
|
Add some needed symbols in exception.cpp and eliminate dependence upon uncaught_exception() from cxa_vector.cpp. libc++abi is very nearly a self-contained (though not complete) library now.
llvm-svn: 148866
|
2012-01-24 21:48:10 +00:00 |
|
Howard Hinnant
|
73ab1864ef
|
Marshall Clow reminded me that fallback_malloc.cpp was broken out for testing purposes. We've agreed to simply rename fallback_malloc.cpp to fallback_malloc.ipp. Thanks Marshall.
llvm-svn: 148863
|
2012-01-24 21:41:27 +00:00 |
|
Howard Hinnant
|
421bbe5cc5
|
Add some needed symbols to cxa_new_delete.cpp and put the display() stuff under #DEBUG (again).
llvm-svn: 148860
|
2012-01-24 21:35:18 +00:00 |
|
Howard Hinnant
|
378a59f15c
|
Got fooled by fallback_malloc.cpp source file: it is really a private header. I've manually included it into the only place it is used: cxa_exception.cpp.
llvm-svn: 148854
|
2012-01-24 21:02:21 +00:00 |
|
Howard Hinnant
|
8b26c3713b
|
Add new file for public type_info-related signatures
llvm-svn: 148853
|
2012-01-24 21:01:01 +00:00 |
|
Howard Hinnant
|
0d517a4278
|
CrashReporterClient.h is back, but this time protected with __has_include. Thanks for the suggestion Doug. The use is consistent with how the same header is used in llvm/lib/Support/PrettyStackTrace.cpp (though there autoconfig is used instead of __has_include).
llvm-svn: 148851
|
2012-01-24 19:58:25 +00:00 |
|
Howard Hinnant
|
a9d8ec45f6
|
Remove dependence upon std::exception_ptr from the default_terminate_handler. Recovered the equivalent functionality at a lower level.
llvm-svn: 148830
|
2012-01-24 18:26:29 +00:00 |
|
Howard Hinnant
|
9e5d74b239
|
Forgot to include this file in the last commit: Move kOurExceptionClass and kOurDependentExceptionClass from source to header so that they can be used in multiple sources. This is a private header, these constants are not publicly exposed.
llvm-svn: 148829
|
2012-01-24 18:17:40 +00:00 |
|
Howard Hinnant
|
6830b2a493
|
Move kOurExceptionClass and kOurDependentExceptionClass from source to header so that they can be used in multiple sources. This is a private header, these constants are not publicly exposed.
llvm-svn: 148827
|
2012-01-24 18:15:20 +00:00 |
|
Howard Hinnant
|
82a39907f1
|
Removed reference to internal header
llvm-svn: 148763
|
2012-01-24 00:52:33 +00:00 |
|
Howard Hinnant
|
e33b2f5c97
|
Changed my mind about __cxa_uncaught_exception and added it.
llvm-svn: 148754
|
2012-01-24 00:01:31 +00:00 |
|
Howard Hinnant
|
bfbf700151
|
I renamed abort_message to be a C++ file to simplify my simplistic build script which I'm still working on. I also added a struct for the crash reporter on __APPLE__.
llvm-svn: 148752
|
2012-01-23 23:58:26 +00:00 |
|
Howard Hinnant
|
fad744dd50
|
A lot of the code in cxa_exception.cpp depends on __cxa_get_globals_fast() returning null if __cxa_get_globals() hasn't been called yet. However it doesn't reliably do that, at least on OS X if __cxa_get_globals_fast() is called prior to pthread_key_create() running. Our choice is to either limit our use of __cxa_get_globals_fast() more than we have, or to have __cxa_get_globals_fast() initialize with pthread_key_create() if necessary. I chose the latter, and replaced pthread_once with a C++11 local static (which should do the same thing).
llvm-svn: 148750
|
2012-01-23 23:55:58 +00:00 |
|
Howard Hinnant
|
671caae892
|
Added missing #include
llvm-svn: 148749
|
2012-01-23 23:51:56 +00:00 |
|
Howard Hinnant
|
53418916fc
|
I'm beginning to be able to throw/catch a wide variety of objects.
llvm-svn: 148713
|
2012-01-23 18:39:47 +00:00 |
|
Howard Hinnant
|
699692a04c
|
Getting started on matching a thrown exception to a catch clause, and setting the adjusted pointer to the caught object appearing in the catch clause.
llvm-svn: 148666
|
2012-01-22 21:47:40 +00:00 |
|
Howard Hinnant
|
afcf7ac5f6
|
I kept getting confused among the __cxa_exception*, the _Unwind_Exception* and the void* to the thrown object. So I've gone through these two files and attempted to institute a consistent variable naming scheme, and in a few instances, turned void* into a concrete* to have the type system help me out. No change in functionality for this commit is intended.
llvm-svn: 148663
|
2012-01-22 19:14:27 +00:00 |
|
Howard Hinnant
|
1a56ee3798
|
Add __shim_type_info which fits below std::type_info and above all of the other type_info-derived classes. This is where all of the virtual functions that serve as details of the inner-workings of type_info will live (safely hidden from public view). All type_info objects will be safely down-castable to __shim_type_info, so as to access implementation detail virtual functions. Also temporarily add some print/display statements to each type_info-derived class. This is in support of the continuing development on the personality function.
llvm-svn: 148647
|
2012-01-22 01:23:02 +00:00 |
|
Howard Hinnant
|
242566e552
|
Change header to generic, instead of implementation specific
llvm-svn: 148646
|
2012-01-22 01:19:20 +00:00 |
|
Howard Hinnant
|
b24c9441d1
|
One more small optimization: Where possible, for loops that do a search and then try to break out of the loop early, eliminate the attempt to break out of the loop after the last search. And with that, I'm declaring __dynamic_cast done. Though if anyone sees any problems, has suggestions for improvements, or wants to contribute some test cases, that is certainly welcome feedback.
llvm-svn: 148246
|
2012-01-16 18:21:05 +00:00 |
|
Howard Hinnant
|
44a2895a03
|
Comment smithing. Changed some casts from C-style to C++. And added timings to all of the tests.
llvm-svn: 148241
|
2012-01-16 17:06:51 +00:00 |
|
Howard Hinnant
|
77fb8ab9f2
|
I think this is getting close on __dynamic_cast. There's been quite a bit of code rearrangement, renaming, and better commenting. This exercise has exposed and fixed a few more bugs. I've also added several more tests (there's definitely a need for more tests here).
llvm-svn: 148227
|
2012-01-16 00:13:46 +00:00 |
|
Howard Hinnant
|
f55694af01
|
Two bug fixes, several clarifications, and a few comment updates.
llvm-svn: 148180
|
2012-01-14 14:15:43 +00:00 |
|
Howard Hinnant
|
fcf732c7a6
|
A bug fix involving the updating of path_dst_ptr_to_static_ptr. Some minor code rearrangement optimizations (putting most likely 'if' first in an if-else series. And some major optimizations which involve stopping the search prior to an exhaustive walk over the entire tree. Some of these stops are because an ambiguity is detected earlier. And some of the short circuiting is due to the information from the bits __diamond_shaped_mask and __non_diamond_repeat_mask. The stress test checked in last night is now about 28% faster for the B<Width/2, Depth> -O3 case. I'm still playing with some more optimization possibilities but I'm not sure they will play out.
llvm-svn: 148153
|
2012-01-13 23:06:03 +00:00 |
|
Howard Hinnant
|
50f7eeec9d
|
Enclosed is a stress test for dynamic_cast. It stresses both libc++abi, and clang itself. It creates a ridiculously large class hierarchy using variadic templates. You can specify both the width and depth of the class hierarchy. And you can specify whether the cast is to the actual run time type, or to an intermediate layer in the class. About 1/3 of the time I compile this, it crashes the compiler. There seems to be an uninitialized area of memory, and I'm probably blowing past an assumption on class hierarchy size within clang (and understandably so). I can get it work most of the time with a class hierarchy width of 20 and a depth of 7. I'm making timings with both -O3 and -Os, using both cast to root and cast to intermediate, on both libc++abi, and gcc's dynamic_cast. I've put the results in a comment/table at the bottom of the test.
llvm-svn: 148083
|
2012-01-13 01:22:31 +00:00 |
|
Howard Hinnant
|
b1ca881191
|
Oops, forgot to svn add the new test.
llvm-svn: 148047
|
2012-01-12 21:08:25 +00:00 |
|
Howard Hinnant
|
b7d59ec6e1
|
Put debug print statments under a flag so that I can get a first glimpse at performance. So far I haven't noticed any performance difference between this new __dynamic_cast and gcc's implementation. But I've barely started looking. Also adding a couple of tests which come straight out of the standard.
llvm-svn: 148046
|
2012-01-12 21:06:12 +00:00 |
|
Howard Hinnant
|
af07bfdb67
|
Fixed a couple of bugs, updated many comments, and am including a comprehensive test for when there are only 3 types in place. I need to do something similar for 4 and maybe more types, but I'm not sure how comprehensive I can make the test at 4 and above types.
llvm-svn: 148038
|
2012-01-12 19:48:14 +00:00 |
|
Howard Hinnant
|
f88480363e
|
Initial implementaiton of __dynamic_cast. There is still lots of debugging code in here that needs to be stripped out. And many, many unit tests need to be written. And comments and probably code cleanliness needs to be improved. But I *think* the basic algorithm is sound. There also may still be some oportunities for algorithm optimization, I'm not positive.
llvm-svn: 147981
|
2012-01-11 23:49:18 +00:00 |
|
Howard Hinnant
|
d9e7393741
|
This is a transitory commit for __dynamic_cast. It contains debugging statements that are not intended to be in the finished product. However some of the dubbing statements themselves contain important documentation such as how to navigate a __class_type_info hierarchy, documenting object offset and inheritance access. The intention is that this debugging code will migrate into both actual code and comments. And capturing it here so that there is no chance this stuff will be lost.
llvm-svn: 147898
|
2012-01-11 00:11:17 +00:00 |
|
Howard Hinnant
|
d88ba82a47
|
Getting a start on the typeinfo infrastructure.
llvm-svn: 147776
|
2012-01-09 17:10:00 +00:00 |
|
Howard Hinnant
|
86b4dfac01
|
First brush with testing __gxx_personality_v0 reveals there is still a long way to go. But my understanding of what it is supposed to do continues to improve. I am currently contemplating whether I need to implement typeinfo before completing __gxx_personality_v0 in order to get matching catch handlers correct.
llvm-svn: 147761
|
2012-01-08 23:50:46 +00:00 |
|
Howard Hinnant
|
7530f19d03
|
Add a few TODO's and FIXME's. Making notes as I go along, but not slowing down to fix these yet. Just don't want anything to fall through the cracks.
llvm-svn: 147760
|
2012-01-08 23:40:41 +00:00 |
|
Howard Hinnant
|
5dc3179e5d
|
Correct comment regarding members of __cxa_dependent_exception. Actually it is useful to disguise a __cxa_dependent_exception as a __cxa_exception by filling in most (not all) of these members at __cxa_dependent_exception construction time. That way most routines don't have to care which kind of exception they have.
llvm-svn: 147759
|
2012-01-08 23:39:14 +00:00 |
|
Howard Hinnant
|
d3dba31762
|
__gxx_personality_v0 update. This is completely untested code. But my brain is scrambled and I wanted to get it checked in. Code review from anyone who knows anything at all about personality functions would be much appreciated. What is in here is borrowed heavily from llvm/examples/ExceptionDemo/ExceptionDemo.cpp. There are some calls to abort() that should probably be changed to return an error code instead. There may be encodings under readEncodedPointer that need to be implemented. And my handling of type_info is almost a complete guess.
llvm-svn: 147680
|
2012-01-06 20:39:47 +00:00 |
|
Marshall Clow
|
d2bab35080
|
Improved comment to explain why we can use __cxa_get_globals_fast here
llvm-svn: 147554
|
2012-01-04 22:18:10 +00:00 |
|
Howard Hinnant
|
66d9327b80
|
Just getting started on the personality routine. This is just a skeleton. Still learning how to fill it in...
llvm-svn: 147547
|
2012-01-04 20:49:43 +00:00 |
|
Marshall Clow
|
f368486f4c
|
Better comment
llvm-svn: 147532
|
2012-01-04 14:56:09 +00:00 |
|
Marshall Clow
|
f83663a9cd
|
Don't allocate TLS storage when checking to see if an exception has been thrown - really
llvm-svn: 147497
|
2012-01-03 23:26:09 +00:00 |
|
Marshall Clow
|
3e417e7a5c
|
Don't allocate TLS storage when checking to see if an exception has been thrown
llvm-svn: 147492
|
2012-01-03 23:10:20 +00:00 |
|
Howard Hinnant
|
f5bde09cc1
|
credits adjustment
llvm-svn: 147148
|
2011-12-22 16:00:06 +00:00 |
|
Marshall Clow
|
1de4fc0dfa
|
Updated __cxa_current_exception_type to call __cxa_get_globals_fast - will not allocate globals
llvm-svn: 147147
|
2011-12-22 15:45:05 +00:00 |
|
Howard Hinnant
|
2aa433398a
|
Updated spec.html regarding __cxa_uncaught_exception.
llvm-svn: 147110
|
2011-12-22 00:03:36 +00:00 |
|
Howard Hinnant
|
22f28b2d52
|
Forgot to increment uncaughtExceptions in __cxa_rethrow_primary_exception
llvm-svn: 147108
|
2011-12-21 23:48:05 +00:00 |
|
Howard Hinnant
|
e04f51662c
|
Added __cxa_increment_exception_refcount, __cxa_decrement_exception_refcount, __cxa_current_primary_exception, __cxa_rethrow_primary_exception
llvm-svn: 147106
|
2011-12-21 23:32:11 +00:00 |
|
Howard Hinnant
|
72a05a9198
|
Add new and delete operators
llvm-svn: 146989
|
2011-12-20 20:38:05 +00:00 |
|
Howard Hinnant
|
bf8a39bb94
|
Fixed several bugs, implemented support for vector types, and cleaned out dead code.
llvm-svn: 146677
|
2011-12-15 20:02:15 +00:00 |
|
Howard Hinnant
|
7a084fdd39
|
Fix size calculation for pointer to member function
llvm-svn: 146463
|
2011-12-13 01:23:16 +00:00 |
|
Howard Hinnant
|
58926c9a9b
|
Made some minor tweaks to __cxa_rethrow
llvm-svn: 146396
|
2011-12-12 19:11:42 +00:00 |