* rewrite span subspan checks to help optimizations
* Removed checking pointer for null for subspans. We would never
check pointer for null in between ptr and ptr+size in the
original span, so there seems to be no reason to do so for
subspans, provided that the subspan's boundaries are ensured
to be within the range of the original span's boundaries.
This change allows to simplify generated code, for example, to
remove 5 out of 9 branches in code generated from the following
by MSVC, and 4 out 8 branches in clang and gcc-generated code:
span<int> mysubspan(int* p, std::ptrdiff_t size, std::ptrdiff_t i)
{
if (p != nullptr)
{
span<int> s = { p, size };
return s.subspan(i);
}
return { nullptr };
}
Similar effects are achieved for dynamic subspans of static spans,
where in the following code we remove 2 out of 4 branchs in MSVC
and GCC-generated code:
int test_dynamic_subspan_of_static_span(std::ptrdiff_t i)
{
int x[] = { 0,1,2,3,4,5 };
span<int, 6> s = { x };
auto subspan = s.subspan(i);
return subspan.size();
}
* Made string_span details::string_length() generic
removed overloads & specialized classes
Creating string_spans using `char16_t` and `char32_t` was not possible
without creating new specializations and function overloads.
This patch makes details::string_length() generic removing the need to
extend the overloads and specializations.
* added type aliases for string_span types char16_t and char32_t
* Added char16_t & char32_t overloads for ensure_z
* added string_span tests for char16_T & char32_t
* added zstring type aliases for char16_t & char32_t
* Added tests for char16_t & char31_t zstring and string_span types
* applies clang format to <gsl/string_span>
* Clang format tests/string_span_tests.cpp
* Removed ensure_z() overloads as they don't add functionality.
using the GSL with MSVC 2017 causes the following warning"
`'std::copy_n::_Unchecked_iterators::_Deprecate': Call to 'std::copy_n'`
This patch removes this since the point of this code is to deal with this issue.
not called even by the unit test (it was calling the "container" ctor
instead). I mimicked the constructor taking a non-const std::array to fix
the issue.
* Fixes issue with dereferencing operator issue #491 introduced in PR #513
dereferencing operator added in PR#513 returned a copy of the object
instead of reference to it.
Adding decltype(auto) as return type of operator* fixes this issue.
* added more tests for not_null::operator*
* Restricting usage of owner<T> to pointer types
* Removing an additional type that was created for testing
* Added comment about the new constraint on owner
* Adding dereference operator to not_null
* Removing dereference operator changes for not-null
* Removing dereference operator changes for not-null
* Review comments
Fixes#510.
* Constrain the converting constructor to not participate in overload resolution when IsConst is true, so that it is never a copy constructor.
* Use Default Member Initializers for span_iterator's data members so that the default constructor can be explicitly defaulted.
* Declare all members of span_iterator GSL_NOEXCEPT: they only throw when contract violations throw.
* Don't use & in operator-> since evil types may overload it.
Project files were not following the clang-format style. For people
using IDEs were clang-format is always run after a save this would
cause unwanted changes.
This commit only applies "clang-format -i" to files.
* Add value_type to span
Currently I'm working on project which involves a lot of `span`s and mocking via Google Mock. Unfortunately a lot of standard matchers requires `value_type` type definition inside container which `gsl::span` lacks.
This pull request add `value_type` type definition inside `gsl::span`
* Strip cv from value_type of span and span_iterator
* Added support of not_null<smart_ptr> comparison
* The return type of not_null comparison operators is determined using SFINAE
#474
* tests for gsl::not_null comparison were added
* not_null comparison tests were rewritten to compare pointers to objects located in the same array
* not_null<shared_ptr> comparison was simplified
* initializer_list overload returns by value to avoid lifetime issues
* generic overload uses expression SFINAE to work with any type that has member size() and operator[], which notably includes const/non-const vector and array.
* Add test coverage for const objects, rvalue initializer_lists, and constexpr usage.
Fixes#357.
* Improve const correctness in string_span
* Improve const correctness in bounds_tests.cpp and byte_tests.cpp
* Improve const correctness in span_tests.cpp
* Improve const correctness in utils_tests.cpp
* Use gsl::owner for dynamically allocated memory in string_span_tests.cpp
* Improve const correctness in string_span_tests.cpp
* Improve const correctness for strided_span_tests.cpp
* GSL_LIKELY/_UNLIKELY always contextually convert their argument to bool.
* Add macro GSL_ASSUME(cond) to inform the optimizer that the predicate cond must hold.
* Reimplement Expects & Ensures as expressions for flexibility and correctness. (Think about "if (cond1) Expects(cond2); else goto fail;")
* Expects and Ensures GSL_ASSUME their predicate when GSL_UNENFORCED_ON_CONTRACT_VIOLATION is defined.
* add gsl::not_null operator<<
Hippomocks, which is a popular mocking engine for C++ uses operator<< on pointers and gets confused about gsl::not_null not having this operator.
* Update gsl
* constexpr all the things.
* remove operator=(const T&)
* it leaves *this in an invalid state if ensure_invariant fails
* implicitly converting the T to not_null and then assigning is in every way superior.
* simplify conversion from not_null<U> with constructor delegation.
* remove the converting assignment operator; again let the conversion constructor and self-assignment operator do the work.
* Cover the remaining pointer arithmetic operations as Wakely suggests in issue #447.
* Cleanup not_null conversions from null pointer constants:
* replace constructor that accepts T with constructor template that accepts U convertible to T
* remove deleted constructor that accepts int
* Attempts to initialize with nullptr, 0, 0L, 0LL, etc. all unambiguously select the deleted nullptr_t constructor.
* Nest "gsl" directory inside a new "include" directory.
* Cleanup the _MSC_VER conditionals a bit; use constexpr on VS2017+.
* Don't #define noexcept on non-Microsoft implementations.
* Workaround VS2017 bug in multi_span. (Also implement == and != for static_bounds_dynamic_range_t because I'm an EoP semantic soundness snob.)
Fixes#441.
whether using it from the development folder, from the installation
folder or from being copied into a project.
#include <gsl/gsl.h>
Updated headers/tests/instructions/cmake build accordingly
This PR should address https://github.com/Microsoft/GSL/issues/277 (less
the renaming of gsl itself)
initializer_list do not have subscript operator, so the generic container overload of gsl::at fails to compile.
This commits adds an overload of gsl::at for initializer_lists, using *(initializer_list::begin()+index) instead of subscript operator
Also removed unused constant member variable that seemed to be there
to prevent maybe_null_* being used with anything other than a pointer,
which is being taken care of with a static_assert now.
for eg consider this case
[code]
string value = "someVeryLongErrorMessageIAm";
finally([value] { PrintErrorMessage(value); }
[/code]
With the current changes before the call to PrintErrorMessage there will be 3 calls to copy constructor for string(1 when it's captured in closure, 2nd when finally is called and 3rd when it's passed to Final_act . With my patch there will be 1 call to the copy constructor and 2 to the move constructor for the scenario in example, so 2 potential deep copies will be saved for some objects.
Validated that code builds from root, and all tests pass after my change. Also validated that indeed copy constructor calls are saved for objects that support move semantics.
