Commit Graph

215 Commits

Author SHA1 Message Date
nobody
fbe0c3eeca Add overloads of hash_value for more built in types. They're not strictly
needed and aren't in the original specifiction but they avoid a warning. See
ticket #1095 for details.


[SVN r38563]
2007-08-10 10:32:21 +00:00
Daniel James
d25ca5a376 Put the borland workarounds back - it still seems to need them. Will probably
have to do something else on top of this for any problems this causes.


[SVN r37803]
2007-05-27 09:28:50 +00:00
Daniel James
27d2c57873 Update copyright on hash headers.
[SVN r37801]
2007-05-27 09:07:09 +00:00
Daniel James
4ef5d0d98c Full namespace for std::size_t.
[SVN r37779]
2007-05-25 23:58:33 +00:00
Daniel James
555fefae20 Try turning off a borland workaround in the new version of borland, as it
seems to be causing problems in the tests.


[SVN r37426]
2007-04-12 20:33:48 +00:00
Daniel James
3ebd297644 Rename hash_detail::float_limits => hash_detail::limits, since it's also used
for integers in the tests.


[SVN r35559]
2006-10-11 18:58:40 +00:00
Daniel James
8aa2b19669 Only use the float.h macros on OpenBSD, it seems that on FreeBSD numeric_limits is actually a better option.
[SVN r35501]
2006-10-05 21:22:56 +00:00
Daniel James
64da418fb2 Fix a typo.
[SVN r35500]
2006-10-05 21:21:43 +00:00
Daniel James
a60b7d12f3 Remove checks for float macros, I was being over-paranoid.
[SVN r35349]
2006-09-26 23:28:34 +00:00
Daniel James
a08d42c10f Attempting to work round the dodgy std::numeric_limits<long double> on
some BSDs.


[SVN r35345]
2006-09-26 20:35:58 +00:00
Daniel James
9430503530 Move the hash extensions into their own header, which hopefully makes the
extension implementation a little clearer. It also fixes a problem where the
visual C++ include once pragma was preventing the extensions from being loaded
on the second include.


[SVN r35314]
2006-09-24 20:00:47 +00:00
Daniel James
311bf0d6c5 The Digital Mars workarounds aren't required for the latest version.
[SVN r34510]
2006-07-12 10:13:45 +00:00
Daniel James
46654045d7 Comment out an assert which is giving bogus failures. On platforms where it isn't possible to detect NaN or infinity the full hash computation is performed on them. For NaN the result doesn't matter (since NaN != NaN), for infinity it should always give the same result), so the calculation works but unfortuanately, I don't know how to correctly check that v is what I expect in the assertion.
I could use a different hash function when I don't have fpclassify/fpclass but that shouldn't be necessary.


[SVN r34490]
2006-07-09 22:50:43 +00:00
Daniel James
572d82713b Update copyright details.
[SVN r34390]
2006-06-25 09:59:05 +00:00
Daniel James
4283c17309 Avoid a warning on Visual C++
[SVN r34329]
2006-06-17 21:22:51 +00:00
Daniel James
d8c636ef9f Tweak the float hashing algorithm to perform fewer calls to ldexp for some
types and possibly generate a better hash value.


[SVN r34273]
2006-06-11 15:29:09 +00:00
Daniel James
82cca89876 Improve support for floating point types with radix != 2.
[SVN r34272]
2006-06-11 14:52:55 +00:00
Daniel James
a4b51721db Oops again, I checked in the wrong version of hash.hpp, which had a broken
version of boost::hash_value<complex> in it. So fix that.


[SVN r34207]
2006-06-06 20:51:24 +00:00
Daniel James
3b29b95b8e Add support for long long to Boost.Hash.
[SVN r34202]
2006-06-06 20:13:33 +00:00
Daniel James
e4b7765da1 Workaround for Visual C++ 7.
[SVN r34191]
2006-06-06 08:14:23 +00:00
Daniel James
f3e6c4e96a Hash workarounds for Digital Mars.
[SVN r33958]
2006-05-07 17:52:38 +00:00
Daniel James
58aec4566a Don't use fpclassify for hashing on OS X.
[SVN r33875]
2006-04-30 19:16:08 +00:00
Daniel James
e592b7f057 Update the copyright comments in the hash headers.
[SVN r33370]
2006-03-18 15:54:53 +00:00
Daniel James
3ce1a740b7 Don't forward declare STL containers when using STLport - this is slightly over the top as the forward declarations seem to work on mingw-3_4_4-stlport-5_0, so maybe it's just a Visual C++ problem, but at this stage I'd rather keep things simple.
[SVN r33178]
2006-02-28 20:47:59 +00:00
Daniel James
d11d2dd17a Reintroduce workaround for hashing const built-in array on Visual C++.
[SVN r33164]
2006-02-28 00:49:14 +00:00
Daniel James
17d5f71c35 Remove TODO note to self.
For the record: I removed the overload because it was causing an ambiguity that I was too lazy to work out at the time. It turned out this was because wchar_t was defined as unsigned short - using BOOST_NO_INTRINSIC_WCHAR_T was the right way to deal with this so the overload no longer needs to be disabled.


[SVN r33114]
2006-02-26 15:58:37 +00:00
Daniel James
a332300682 Fix a typo in the workaround code for compilers without partial specialization.
[SVN r33104]
2006-02-25 15:17:01 +00:00
Daniel James
fa8ce5ac40 Use _fpclass and fpclassify on less platforms. Hopefully will use them more widely after the release.
[SVN r33047]
2006-02-21 00:12:36 +00:00
Daniel James
5fdfe5d328 Add support for the BOOST_HASH_NO_EXTENSIONS macro.
[SVN r32900]
2006-02-13 18:21:20 +00:00
Daniel James
b36cac82c3 In hash_float, only use _fpclass on Visual C++, I'll find out about other compilers later.
[SVN r32833]
2006-02-11 21:15:15 +00:00
Daniel James
0a1800f9f8 A couple of workarounds for Digital Mars.
[SVN r32786]
2006-02-09 19:22:54 +00:00
Daniel James
514757c312 Hold your nose, it's a Visual C++ 6.5 workaround. For some reason it requires
boost::hash to define operator() taking const and non-const arguments.


[SVN r32783]
2006-02-09 19:11:54 +00:00
Daniel James
1cf74208aa Use fpclassify/_fpclass for the hash function on compilers where I
know it's available.


[SVN r32782]
2006-02-09 19:11:09 +00:00
Daniel James
1d5bfbcc9c Since all the hash functions are now declared in a single header, there's no
need for the call_hash hack on compliant compilers. So only use it on
compilers where it's needed.


[SVN r32781]
2006-02-09 19:10:03 +00:00
Daniel James
5a5bbbf403 Make hash_float.hpp use its own combine function. This serves two purposes:
1. hash_float is no longer dependent on the main hash header.
2. hash_float(0) == 0 which makes it easier to implement hash<complex>(),
although this might lead to users assuming that it can be expected even though
it isn't part of the documented interface.


[SVN r32780]
2006-02-09 19:07:33 +00:00
Daniel James
19564252b4 Move all the boost/functional/hash/*.hpp files into a single file
(hash/hash.hpp for now, but will eventually move into hash.hpp), add warnings
that the old headers are deprecated.


[SVN r32778]
2006-02-09 18:57:07 +00:00
Daniel James
b2be632311 Move include errorno & limits from hash.hpp into hash_float where they should be.
[SVN r32777]
2006-02-09 18:55:35 +00:00
Daniel James
d8328a403c Attempt to get container_fwd.hpp working on more compilers.
[SVN r32684]
2006-02-07 00:43:06 +00:00
Daniel James
81483cc533 Move the float hash function into its own header.
[SVN r32594]
2006-02-05 19:51:29 +00:00
Daniel James
b922722090 Use BOOST_WORKAROUND in boost/functional/hash/hash.hpp
[SVN r32589]
2006-02-05 16:31:45 +00:00
Daniel James
da55a4fcca Use hash_fwd.hpp in the main hash header.
[SVN r32588]
2006-02-05 16:27:19 +00:00
Daniel James
1f2c6bfaba Add a forward header for the hash functions.
[SVN r32586]
2006-02-05 16:03:43 +00:00
Daniel James
ed548575d4 Use container_fwd.hpp for the hash functions.
[SVN r32583]
2006-02-05 14:44:24 +00:00
Daniel James
1ff1366c5e Add std::pair, and a GCC 2.95 workaround to container_fwd.hpp
[SVN r32581]
2006-02-05 14:04:56 +00:00
Daniel James
332201e5c8 Change container_fwd.hpp to cooperate with gcc stdlib++'s debug mode.
[SVN r32081]
2005-12-16 19:14:16 +00:00
Daniel James
016766fdec Add char_traits support to the container_fwd header.
[SVN r31133]
2005-09-27 19:45:04 +00:00
Daniel James
bb16ae385a Add STL forward declarations for the hash library (just for testing, not in use yet)
[SVN r31043]
2005-09-19 20:29:51 +00:00
John Maddock
9158f88562 Large patch from Ulrich Eckhardt to fix support for EVC++ 4.
[SVN r30670]
2005-08-25 16:27:28 +00:00
Daniel James
74ec71fe8d Check errno after calling frexp - this is pretty slow and unecessary on most platforms, but this close to release it seems better to be over cautious.
[SVN r29255]
2005-05-28 12:10:33 +00:00
Daniel James
8090093f06 Initialise the exponent before calling frexp in the hash float function.
[SVN r29200]
2005-05-26 07:36:26 +00:00
Daniel James
7092068156 Another attempt at getting the hash array tests to pass on Borland.
[SVN r28390]
2005-04-21 22:20:43 +00:00
Daniel James
f2afc2a08d Another fix for the Borland hash array code, and get rid of my attempt at working around Borland problems with infinity and NaN in the hash float functions. It seems impossible to do anything with them.
[SVN r28286]
2005-04-17 09:09:42 +00:00
Daniel James
7223c8da60 Remove workaround for bug in gcc 4.0 which appears to have been fixed.
[SVN r28282]
2005-04-16 16:59:30 +00:00
Daniel James
72cd8cdaa1 Overload hash_value for non-constant arrays.
[SVN r28276]
2005-04-16 10:32:29 +00:00
Daniel James
dbed5ebafa Use non-const references on Visual C++ 6.5 to work around a bug with array references.
On Borland add a pointer overload of hash_range to use instead of the array workaround that I checked in earlier today.

Fix the array size calculation for compilers with function template ordering.


[SVN r28228]
2005-04-13 22:56:43 +00:00
Daniel James
976683836e Borland workaround for hashing arrays.
[SVN r28226]
2005-04-13 20:24:50 +00:00
Daniel James
be1be0d1fe Deal with NaN & infinity in the hash float function on Broland.
[SVN r28186]
2005-04-12 20:23:09 +00:00
Daniel James
0a43718366 Setup hash float function config for Tru64 on alpha and add some comments on
the different libraries & compilers.


[SVN r28138]
2005-04-11 22:20:33 +00:00
Daniel James
8b08528611 Better version of the built in array hash function.
[SVN r28089]
2005-04-10 20:07:50 +00:00
Daniel James
15d97d95e7 Implement hash functions for built-in arrays. Uses partial specialization
instead of function overloading - which disobeys the specification of
the extension (hash_combine is defined in terms of hash_value, but hash_value
doesn't deal with every case now).


[SVN r28078]
2005-04-10 12:36:24 +00:00
Daniel James
5e9f0e4c8e Use the named variants of the float functions for stlport on gcc-2.95.
[SVN r28067]
2005-04-09 09:27:20 +00:00
Daniel James
50e6bad2fd Remove the Boost.Range overloads of hash_range.
[SVN r28051]
2005-04-07 21:08:58 +00:00
Daniel James
a65dc47fcb Change the hash float config for Rogue Wave so that it only applies to Borland,
as it seems to be wrong on SunPro.


[SVN r28029]
2005-04-06 22:52:25 +00:00
Daniel James
27d6247dac Include config files before any other header to prevent Visual C++ 6 warnings.
Remove Boost.Range support when there are no const volatile specializations, as it currently requires them to compile. Hopefully this will be changed back in the future.


[SVN r27914]
2005-04-01 22:38:39 +00:00
Daniel James
2b375867f5 Initial import of hash function headers - so that they can be used by
Boost.MultiIndex.


[SVN r27819]
2005-03-27 22:58:52 +00:00