[ci skip]

This also fixes an issue in the "writing a sorter" tutorial that used
the pre-1.0.0 arguments order for cppsort::sort.

Wherever the library implements special handling of utility::identity,
provide the same level of support for std::identity when it is
available.

…ssue #130)

Add new C++20 category to describe the new support for std::identity,
std::ranges::less and std::ranges::greater.

Remove the section about how bitmap_allocator might be used and the text
about the relevance of the page across old versions. The embedding of
the documentation into the main repository made that text useless
(issue #157).

[ci skip]

Mostly small things that I noticed one working on the 2.0.0 branch,
which will undoubtly result in merge issues later, but these ought to be
fixed in 1.x.y anyway.

The bidirectional version of vergesort was modified as follows:
- It now uses the same k-way merge as the random-access version.
- It creates fewer partitions with one element, generally being smarter
  about which run an element should belong to.
- It spends less time recomputing runs sizes, leading to a lower impact
  on the running time overall.

The k-way merge algorithm is now shared between both versions and has
been tweaked to handle 1-element runs better. The in-code documentation
was also reworked to be more comprehensible and comprehensive. Some
variable names were changed to make the logic more obvious.

- The forward iterator version used utility::as_function without
properly qualifying it, which was a bug.
- The bidirectional version was updated with the "shrink on the left"
trick, which can both decrease the cost of the size computation and
reduce the number of elements to allocate.

[ci skip]

First step to move the build to GitHub Actions: move the coverage
generation from Travis to Actions. It doesn't generate additional
information with lcov anymore, and now always runs on the latest
available Ubuntu.

[ci skip]

[ci skip]

It is totally possible for nested wrapping of stable_adapter to happen,
this commit makes stable_adapter<stable_adapter<Sorter>> equivalent to
stable_adapter<Sorter>.

[ci skip]

Declare that the result of std::mem_fn is branchless when it wraps a
pointer to data member. This is part of a wide library audit to make
pdqsort use its branchless partitioning algorithm more often. This
specific change uses standard library implementation-specific class
names, but it is quite important since as_function currently uses
std::mem_fn where it makes sense.

The following modifications were needed:
- Use the GitHub Action codecov/codecov-action
- Rename the workflow from codecov.yml to code-coverage.yml, otherwise
  the Codecov Action things it is the Codecov configuration file
- Change codecov.yml to ensure that the test files are not covered

[ci skip]

[ci skip]

Cleanup the links and add information:
- Mention the 128-bit integers support for counting_sorter
- Mention the conditional std::mem_fn extended support
- Mention the condition improvements brought by assumptions

[ci skip]

The type alias stable_t is kind of a better equivalent to
stable_adapter: it aliases either a sorter directly if it is always
stable, otherwise stable_adapter<Sorter>::type if it exists, otherwise
stable_adapter<Sorter> itself. The goal is to alias the "least nested"
type that is always stable to decrease template instantiation, template
depth, and somehow bonary size. It also improves the error messages is
some cases.

Bundled with this feature: stable_adapter specializations now expose a
member type named 'type' to implement the mechanism described above.

The documentation page about sorters adapters was slightly cleaned up
and reworked as part of this change.

This change allows to use them with distributions. This commit also
takes advantage of this to change several parts of the test suite to use
distribution, and applies some drive-by cleanups.

Link to cppreference to make the doc more amenable. Also use the
lowercase concept name, and remove the duplicate "concept" word.

[ci skip]

[ci skip]

[ci skip]

Mostly quality-of-life changes that allow me to change them faster when
I need to prototype, debug and test components more easily.

[ci skip]

An likewise for stable_adapter<verge_adapter>. Instead of simply relying
on make_stable, the new specializations use a variant of vergesort that
detects strictly descending runs instead of non-ascending ones, and
wraps the fallback sorter in stable_t.

The technique of detecting non-descending and strictly descending runs
is a trick borrowed from timsort in order to preserve stability in
natural mergesort algorithms: descending runs are reversed in-place, so
equivalent elements wouldn't retain their original order when reversed
in such runs.

While verge_adapter does not handle bidirectional iterators, the
underlying code was changed so that it will be easier to do in the
future, but currently this involves some dirty tricks with a stripped
down implementation of a sized iterator. This will make the transition
to C++20 ranges and standard sized iterators and sentinels easier in the
long term. Meanwhile it is just an implementation detail.

This commit addresses external issue Morwenn/vergesort#11 and somehow
Morwenn/vergesort#7 too. cpp-sort is a better recipient for those
variations on vergesort than the standalone project, but some
cross-project documentation will be needed anyway.

The previous named caused the file to be found by the Codecov bash
uploader, which isn't exactly what we want...

[ci skip]

[ci skip]

[ci skip]

Also add a test, which doesn't actually test that the failing
constructor works, but will do in the 2.0.0-develop branch.

This specific distribution is implement after the paper A Killer
Adversary for Quicksort, by M. D. McIlroy. The goal is to force some
pivot-based algorithms to use their fallback algorithm that ensure that
they won't turn quadratic.

[ci skip]

This commit also comments the Travis CI MacOS configurations, because
they there too slow to run to completion anyway.

[ci skip]

[ci skip]

[ci skip]

[ci skip]

[ci skip]

[ci skip]

[ci skip]

[ci skip]

This commit is hopefully the last to get things to work with MacOS
builds. Compared to the old Travis CI builds, we don't have Valgrind
builds, but we have g++, ubsan and asan ones, which is more coverage
than what we used to have.

[ci skip]

[ci skip]

The actions tests both g++ 5.5 and clang++ 6, with ubsan, asan and
Valgrind memcheck.

[ci skip]

[ci skip]

[ci skip]

[ci skip]

- Remove the license badge, redudant with GitHub UI
- Better text for release and Conan badges
- Remove the Travis CI badge
- Add build badges to the compiler support section
- Update the information about supported compilers

Drive-by fixes:
- Remove useless default template parameters
- std::move a projection in counting_adapter

indirect_adaper is implemented with mountain sort for random-access
iterator. This commits turns internal lambdas used for projections into
a proper named function object, and marks it as likely branchless when
the projection it wraps is likely branchless itself.

The lambda was a leftover from when ska_sort still relied on std::sort
as a fallback. Replacing it with std::less<> + the passed projection
means that pdqsort might use its branchless partition algorithm when it
makes sense.

Due to a copy-paste error, ska_sort did not work properly with signed
128-bit integers: it probably sorted signed and unsigned values
correctly, but not a mix of both.

This commit also uses std::uintptr_t when available instead of
std::size_t when sorting pointers, making it generally more correct.

…177)"

This reverts commit 78a93e9.

Change originally proposed by @vedgy in timsort/cpp-TimSort#35

This makes quick_merge_sort O(n log n) instead of O(n²) like it used to
be (the libc++ implementation of std::nth_element used until then was
quadratic).

The new implementation was taken from Danila Kutenin's miniselect
library.

Split swap_ranges calls into swap_ranges_inner and swap_ranges_overlap,
where the first has the "no overlap" guarantee and checks to ensure that
said guarantee holds.

This commit also adds an audit mechanism to cpp-sort for checks that
are too expensive for CPPSORT_ASSERT, and starts using it in
swap_ranges_inner to ensure that the "no overlap" guarantee holds.

This includes the following changes:
- Move it out of detail and in cpp-sort/comparators
- Get rid of its unused getters
- Make it conditionally branchless (#177)
- Make it a transparent comparator
- Document it

This commit includes a little cleanup of the corresponding documentation
page.

This reverts commit 0e8c906.