4f2a715c9f
Fully implement float_value_test and double_value_test
2015-03-09 17:11:54 -04:00
09ae6c961f
Rename BOOST_TEST_MEMCMP_EQ to BOOST_TEST_MEM_EQ. Add float_test to Jamfile. Add float_value_test() and double_value_test().
2015-03-08 10:50:39 -04:00
072b495240
Add Endian specific lightweight_test.hpp with new macro BOOST_TEST_MEMCMP_EQ that uses std::memcmp(...)==0 to determine equality. Bitwise equality is the standard we want floating point round trips to be held to, since NaN's are then equal.
2015-03-05 10:52:28 -05:00
9c0cbae556
Revert the default (*_t) types to unaligned, and used *_at to identify the aligned types. See mini-review thread for rationale.
2015-02-23 07:24:17 -05:00
b6b03ea900
Fix incorrect header guard
2015-02-21 09:02:41 -05:00
b0ef03b9d6
Rearranges rows and columns in tables. Add big_int8_t, big_uint8_t, little_int8_t, little_uint8_t arithmetic and buffer types, in response to mini-review query.
2015-02-16 19:31:48 -05:00
adee2fe513
Add support for deprecated names from pre-review releases.
2015-01-18 19:48:27 -05:00
35f4c47a22
Add endian_buffer stream inserter and extractor, beef up buffer_test a bit.
2015-01-17 10:25:29 -05:00
321d5a087a
For X86 and X64 architectures, which permit unaligned loads and stores, unaligned little endian buffer and arithmetic types use regular loads and stores when the size is exact. This makes unaligned little endian buffer and arithmetic types significantly more efficient on these architectures.(Pull request from Jeremy Maitin-Shepard) Docs updated, but timing results have not yet been updated.
2015-01-06 16:57:53 -05:00
d15b2fdb1c
Refined rationale for choice of exact-length types for conversion support.
2014-12-16 08:49:01 -05:00
05ea08995c
Pass through index.html making corrections and editorial changes.
2014-12-15 08:40:35 -05:00
971ba888d6
Decompose Reversible into EndianReversible and EndianReversibleInplace.
2014-12-11 11:41:44 -05:00
6c173766e3
Eliminate gcc -Wconversion warnings.
2014-12-10 15:03:06 -05:00
dcde9463fc
Do not define _SCL_SECURE_NO_WARNINGS or _CRT_SECURE_NO_WARNINGS if already defined (such as on the command line), plus related cleanups.
2014-12-10 10:26:49 -05:00
c201455c84
Change "in_place" to "inplace". Rationale: shortens names without impacting expressiveness.
2014-12-10 09:39:03 -05:00
2cbcd6f7d3
Rename "runtime_conditional_reverse" to "conditional_reverse". The differing number of arguments is sufficient to disambiguate the compile-time and run-time overloads.
2014-12-10 09:26:47 -05:00
dcca2835a9
Rename "reverse_endianness" to "endian_reverse". Shorter but still plenty explicit. See bikeshed.txt for alternatives considered.
2014-12-10 08:30:30 -05:00
e4466edf4e
Drop the _unless_big and _unless_little functions. KISS. They were confusing, and less confusing names were never found.
2014-12-09 20:43:47 -05:00
55a65c3806
Revert to again supply the reverse_endianness_in_place template. This has the effect of providing a default reverse_endianness_in_place for user-defined types. Add error checking for UDT's to verify this works as desired. Provide a note in the conversion docs to that effect.
2014-12-09 11:36:42 -05:00
a0b5805b73
Minor namespace detail cleanup.
2014-12-09 07:38:04 -05:00
f0532a3854
Improve comments.
2014-12-09 07:08:41 -05:00
db16b51b41
Tweak comments.
2014-12-05 09:05:05 -05:00
e2045b7ffa
Finalize buffer and arithmetic decomposition. Finalize name changes. Finalize test cases. Remove cruft. Docs still to do.
2014-11-26 10:04:33 -05:00
56103a7bc6
Finish conversion.hpp cleanup after name changes. conversion_test.cpp return-by-value tests reorganized and passing. UTD and in-place tests still to come. Add doc/bikeshed.txt to keep record of naming rationale.
2014-11-24 10:04:23 -05:00
4197f996ac
More conversion.hpp fixes. Some, but not all, conversion_test tests passing.
2014-11-23 10:38:21 -05:00
ce73f59741
Add "_endianness" to the names of the two customization points. These names must be defined in user namespaces (for ADL to work) and so must be much more explicit than names that are defined in namespace endian.
2014-11-23 08:40:58 -05:00
f9c562b4b4
Rework conversion.hpp: Make the UDT customization point for the value returning conversion functions much more explicit. Provide a separate UDT customization point for the in-place conversion functions, and reverse_in_place() overloads for the built-in arithmetic types. Eliminate the unsafe general-use reverse functions. With explicit customization points these are no longer needed. Complete the renaming of conversion functions started in the prior commit. The reverse and reverse_in_place names are much more meaningful and less likely to be misunderstood. The longer "reverse_in_place" name will hopefully at least slow users down before inappropriate (and possibly unsafe use, such as when trying to recover data after an exception).
2014-11-22 16:27:11 -05:00
49a91c293a
Rename endian_buffer typedefs. New names a bit longer, but also a bit more explicit and more obviously related to endian_arithmetic typedefs.
2014-11-21 15:49:28 -05:00
84faaa7bbf
Rename the conversion functions to discourage unthinking use of the in-place versions as these are unsafe in that they effectively break the type system by treating the buffer as an undescriminated union. Remove the synonym functions based on names popularized by BSD, OS X, and Linux since these are permitted to be implemented as macros, which would cause endless difficulties if encountered.
2014-11-20 15:38:25 -05:00
fc7733ade1
Major renaming. Class "endian" becomes class "endian_arithmetic" to mimic the C++ standard's description of such types as "arithmetic types". Rename header accordingly. The convenience typedefs have been changed shorten aligned names, as they are often being, but keeping the unaligned type names relatively short via a simple "_ut" suffix.
2014-11-19 10:53:11 -05:00
61d9046348
Remove "operator value_type() const noexcept" from buffers. Make the buffer ctor explicit. Buffers were requested by people who do not want automatic conversions because they want full control over conversions, and are very concerned about "efficiency".
...
Add "operator value_type() const noexcept" to the endian arithmetic types. These types are used by people who do want fully automatic conversions. The arithmetic types need to be drop-in replacements for the built-in arithmetic types for these users.
2014-11-18 14:33:31 -05:00
e9f6d2c37c
Re-implement the endian types by inheriting from endian_buffer.
2014-11-15 21:04:19 -05:00
561d8f7070
Initial commit buffers.hpp and related infrastructure. A separate buffer class without arithmetic operators was requested during the formal review.
2014-11-15 08:02:08 -05:00
c755a6b6af
Add overall css style sheet for more uniform presentation and ease of maintenance.
2014-11-12 08:26:01 -05:00
c41521cead
For the endian types, add stream inserters and extractors as requested during the formal review.
2014-11-11 21:08:08 -05:00
4c58fc2efa
Clear [-Wunused-parameter] warnings (Roger Leigh)
2014-08-29 08:12:16 -04:00
f1dc05db63
Replaced deprecated macro with BOOST_NO_CXX11_DEFAULTED_FUNCTIONS
2014-08-15 13:10:58 -04:00
17ca240b10
Use boost/predef/detail/endian_compat.h to determine endianness.
2014-08-15 10:59:49 -04:00
f489fad5e0
Also add aligned float types. Add docs. No tests yet.
2014-08-13 13:44:28 -04:00
8ea7d201f9
native_float32_t and native_float64_t, fixing issue #2 .
2014-08-13 13:23:11 -04:00
071099b656
Remove commented out code residue from treating native as a synonym for big or little.
2014-08-12 08:29:24 -04:00
d31caa413c
Add types native synonyms. All tests now passing, at least on a little endian machine.
2014-08-11 17:17:50 -04:00
8e5cd7dfc4
Treat "native" as synonym for "big" or "little" according to platform endianness, as suggested by Vicente J Botet. Converter_test working.
2014-08-11 16:21:04 -04:00
8c36890e77
Test and doc updates for Pierre Talbot's int8_t/uint8_t reverse_value() and templated reverse() patch.
2014-04-16 13:52:24 -04:00
11fc1ea374
Add overloads for int8_t and uint8_t (rational: useful with generic programming). Template the reverse function (same body in each overloads).
2014-01-22 18:09:50 +01:00
d339470e6e
Remove "explicit" from Endian type constructors. Rationale: Having to write "cntr.insert(big_int32_t(value))" instead of "cntr.insert(value)" gets old fast when you are writing application code and the use case arises more or less continuously. Additionally, the use of an endian type is an implementation detail that should not leak into the application code that needs to create values of the type. Furthermore, the ambiguities that caused the constructors to be made explicit no longer occur; they were eliminated when BOOST_MINIMAL_INTEGER_COVER_OPERATORS was introduced to removed a dependency on boost/operators.hpp. If an ambiguity does arise, it is easy to clear via a static_cast or a function-style cast.
2013-09-06 10:36:54 -04:00
016d3b2e7c
Rename the endian typedefs. Rationale: Subjective - during three months working on the B-tree library, the *un_t names were a constant source of irritation. Objective - the new scheme, which reverts the *un_t names to having no alignment decoration, provides consistent *_align* decoration for both FP and integer types. Also, the library now provides conversion functions which may reduce use of aligned endian types.
2013-09-01 08:40:32 -04:00
d1d62f3adc
Use detail::reverse_value() to implement reverse_value() for float and double, thus eliminating type-punning warning from gcc.
2013-06-07 15:46:14 -04:00
db3864b3d4
Fix boo boos detected by inspect program.
2013-05-29 09:40:35 -04:00
fbedc9bb9f
Remove another bin_manip file. Note that all the bin-manip files have been copied to git@github.com:Beman/bin-manip.git
2013-05-29 08:36:40 -04:00
Beman
Pierre Talbot