GitLab

* Change node_map type from map<ptr,ptr> to vector<pair<ptr,ptr>>

Map nodes are now iterated over in document order.

* Change insert_map_pair to always append

Always append in insert_map_pair even if the key is already present.
This breaks the behavior of force_insert which now always inserts KVs
even if the key is already present. The first insert for duplicated keys
now takes precedence for lookups.

* Remove extra semicolon

* Fix automatic type conversion

* Replace dynamic exception specifications by C++11 noexcept

* Fix deprecated definition of implicit copy constructor for 'Exception'

(Not stack<unique_ptr<T>> because it wasn't quite a stack; we needed to get the second-to-last element sometimes.)

Add the .codedocs config file and badge to the README.md file.

Fix for broken build on cygwin using gcc - "error '[fileno, strdup, fdopen]' are not in scope" (#374)

Apparently the default is

on_success: change
on_failure: always

which seems reasonable.

Reason: broken build.

This reverts commit 0d810ad6.

Hooray for testing travis configs via pushing to master.

gcc 4.6 (the default) does have some C++11 support, but its flag is -std=c++0x. Since 4.7 was released in 2012, I think it's reasonable to require it.

This adds yaml-cpp-config.cmake, yaml-cpp-config-version.cmake, and
yaml-cpp-targets.cmake to the cmake install. As a result, cmake's
find_package can easily find yaml-cpp for software that depends on
yaml-cpp.

Add code to install cmake files to $CMAKE_INSTALL_PREFIX/CMake on
Windows, which is the de-facto standard.

Closes jbeder/yaml-cpp#336 jbeder/yaml-cpp#127

This improves performance on the test.yaml attached to #158 by about
25% on my machine as compared to the previous commit (0.25s -> 0.20s),
as measured by `time build/util/parse < test.yaml > /dev/null`.

This improves performance on the test.yaml attached to #158 by about
35% on my machine (0.39s -> 0.25s), as measured by
`time build/util/parse < test.yaml > /dev/null`.

Flag -n N repeats parsing N times.
Flag -c (--cache) caches the input in a string and uses that to parse.

On my Macbook Pro, reading from standard input incurs a bunch of locking
overhead, which complicates profiling and (IMO) adds noise to
results. This adds the option to read from a file, which doesn't incur
this overhead.

 - Adds 'std=c++11' compiler flags
 - Replaces boost::type_traits with std::type_traits
 - Replaces boost::shared_ptr with std::shared_ptr
 - Replaces std::auto_ptr with std::unique_ptr
 - Replaces raw pointers with std::unique_ptr in ptr_vector, ptr_stack, and SettingChanges
 - Replaces boost::noncopyable with deleted copy and assignment operators
 - Replaces boost::next with std::next
 - Replaces boost::enable_if with std::enable_if
 - Replaces boost::is_convertible with std::is_convertible
 - Replaces ptrdiff_t with std::ptrdiff_t
 - Replaces boost::iterator_facade and boost::iterator_adaptor with std::iterator, borrowing the 'proxy reference' technique from boost
 - Removes Boost dependency from CMakeLists
 - Formats changed files using clang-format

- ensures shared library code is position-independent.