This patch changes implementation to be the same as before
708bd019 (implementation of
runtime_initialize(void) is left unchanged).

The change is needed for backward compatibility with pmemkv-1.0. There, we
called runtime_initialize(true) in a transaction after tx commit (we assumed
there is nothing transactional in runtime_initialize). Introducing a tx
inside new runtime_initialize() results in crash on 1.0

Fix delete_persistent to be called properly, on object, in ptr test

Add defragmentation feature

Can be used to iterate over all internal vector's pointers.
It's used i.a. in defrag class for storing all ptrs
in internal list for defragmentation process.

Add a defrag class, which holds pointers which will be a subject
of a defragmentation process, using libpmemobj's pmemobj_defrag()

ref. pmem/pmdk#4066

Hash map consistent size feature

The bool parameter is not needed anymore, size is always restored from
'consistent_size'.

using enumerable_thread_local.

they will be set inside tx so we need p<>.

Implementation of insert_or_assign in concurrent_hash_map

Fix regression in add_test_generic()

Remove duplicated cmake scripts

Add compatibility tests (with different libpmemobj-cpp versions)

Fix tests_clang_release_cpp11_no_valgrind in run-build.sh

Fix types in enumerable_thread_specific_access test

On travis, without this flag rm asks for input if it should
"remove write-protected regular file".

Set CMAKE_BUILD_TYPE to Release instead of Debug.

typo fixes

It fixes the following compilation error on openSUSE Leap:
In file included from /libpmemobj-cpp/tests/enumerable_thread_specific_access/enumerable_thread_specific_access.cpp:33:0:
/libpmemobj-cpp/tests/enumerable_thread_specific_access/enumerable_thread_specific_access.cpp: In function 'void test(pmem::obj::pool<root>&)':
/libpmemobj-cpp/tests/enumerable_thread_specific_access/enumerable_thread_specific_access.cpp:100:19: error: conversion to 'size_t {aka long unsigned int}' from 'int' may change the sign of the result [-Werror=sign-conversion]
   UT_ASSERT(n_100 + n_zeros == concurrency);
             ~~~~~~^~~
/libpmemobj-cpp/tests/common/unittest.hpp:130:11: note: in definition of macro 'UT_ASSERT'
  ((void)((cnd) ||                                                       \
           ^~~

The add_test_common() function requires the TEST_CASE variable
to be defined.
It fixes regression introduced by b969cb5f.

Insert_or_assign from std::unordered_map (since C++17) without hints, additional method with heterogeneous key-type and tests

travis: update valgrind to 3.15

Different spell checkers find different typos.

Refactor enumerable_thread_specific

The add_test_common() function requires the TEST_CASE variable
to be defined.
It fixes regression introduced by b969cb5f.

change building behavior for gcc 4.8.5

Now, thread id of each thread is also an index to persistent storage
for enumerbale_thread_specific (this index is the same for all
instances of enumerable_thread_specifc which certain thread uses).

This change means that persistent storage can be bigger than neccesary
in some cases (it must be as big as largest 'thread id + 1'). This
implementation also allowed for removal of read lock in local() method.
Now, it's enough to check atomic size variable and if size < thread id
just return element with this index.

This commit also removes local(bool) method beacuse it's not possible to
tell whether certain element existed before or not.

* Replace v<> with volatile_state

* Use std::shared_lock and std::unique_lock instead of
  hand made scoped_lock

* Implement custom thread_id assigning mechanism so that slots in
  persistent storage can be reused after a thread exited

This is needed for enumerable_thread_specific tests
in which we need guarantee on number of threads executing
simultaneously.

Add SCRIPT parameter to add_test_generic()