Take maintainership
Update to v1.10
== 21 January 2011 ==
I've just released sparsehash 1.10. This fixes a performance
regression in sparsehash 1.8, where sparse_hash_map would copy
hashtable keys by value even when the key was explicitly a reference.
It also fixes compiler warnings from MSVC 10, which uses some c++0x
features that did not interact well with sparsehash.
There is no reason to upgrade unless you use references for your
hashtable keys, or compile with MSVC 10. A full list of changes is
described in
[http://google-sparsehash.googlecode.com/svn/tags/sparsehash-1.10/ChangeLog
ChangeLog].
=== 24 September 2010 ===
I've just released sparsehash 1.9. This fixes a size regression in
sparsehash 1.8, where the new allocator would take up space in
`sparse_hash_map`, doubling the sparse_hash_map overhead (from 1-2
bits per bucket to 3 or so). All users are encouraged to upgrade.
This change also marks enums as being Plain Old Data, which can speed
up hashtables with enum keys and/or values. A full list of changes is
described in
[http://google-sparsehash.googlecode.com/svn/tags/sparsehash-1.9/ChangeLog
ChangeLog].
=== 29 July 2010 ===
I've just released sparsehash 1.8. This includes improved support for
`Allocator`, including supporting the allocator constructor arg and
`get_allocator()` access method.
To work around a bug in gcc 4.0.x, I've renamed the static variables
`HT_OCCUPANCY_FLT` and `HT_SHRINK_FLT` to `HT_OCCUPANCY_PCT` and
`HT_SHRINK_PCT`, and changed their type from float to int. This
should not be a user-visible change, since these variables are only
used in the internal hashtable classes (sparsehash clients should use
`max_load_factor()` and `min_load_factor()` instead of modifying these
static variables), but if you do access these constants, you will need
to change your code.
Internally, the biggest change is a revamp of the test suite. It now
has more complete coverage, and a more capable timing tester. There
are other, more minor changes as well. A full list of changes is
described in the
[http://google-sparsehash.googlecode.com/svn/tags/sparsehash-1.8/ChangeLog
ChangeLog].
=== 31 March 2010 ===
I've just released sparsehash 1.7. The major news here is the
addition of `Allocator` support. Previously, these hashtable classes
would just ignore the `Allocator` template parameter. They now
respect it, and even inherit `size_type`, `pointer`, etc. from the
allocator class. By default, they use a special allocator we provide
that uses libc `malloc` and `free` to allocate. The hash classes
notice when this special allocator is being used, and use `realloc`
when it can. This means that the default allocator is significantly
faster than custom allocators are likely to be (since realloc-like
functionality is not supported by STL allocators).
There are a few more minor changes as well. A full list of changes is
described in the
[http://google-sparsehash.googlecode.com/svn/tags/sparsehash-1.7/ChangeLog
ChangeLog].
=== 11 January 2010 ===
I've just released sparsehash 1.6. The API has widened a bit with the
addition of `deleted_key()` and `empty_key()`, which let you query
what values these keys have. A few rather obscure bugs have been
fixed (such as an error when copying one hashtable into another when
the empty_keys differ). A full list of changes is described in the
[http://google-sparsehash.googlecode.com/svn/tags/sparsehash-1.6/ChangeLog
ChangeLog].
=== 9 May 2009 ===
I've just released sparsehash 1.5.1. Hot on the heels of sparsehash
1.5, this release fixes a longstanding bug in the sparsehash code,
where `equal_range` would always return an empty range. It now works
as documented. All sparsehash users are encouraged to upgrade.
=== 7 May 2009 ===
I've just released sparsehash 1.5. This release introduces tr1
compatibility: I've added `rehash`, `begin(i)`, and other methods that
are expected to be part of the `unordered_map` API once `tr1` in
introduced. This allows `sparse_hash_map`, `dense_hash_map`,
`sparse_hash_set`, and `dense_hash_set` to be (almost) drop-in
replacements for `unordered_map` and `unordered_set`.
There is no need to upgrade unless you need this functionality, or
need one of the other, more minor, changes described in the
[http://google-sparsehash.googlecode.com/svn/tags/sparsehash-1.5/ChangeLog
ChangeLog].
Mon May 2 07:04:46 2005 Google Inc.
* sparsehash: version 0.2 release
* Preliminary support for msvc++ compilation. (csilvers)
* Documentation fixes -- some example code was incomplete! (csilvers)
* Minimize size of config.h to avoid other-package conflicts (csilvers)
* Contribute a C-based version of sparsehash that served as the
inspiration for this code. One day, I hope to clean it up and
support it, but for now it's just in experimental/, for playing
around with. (csilvers)
* Change default namespace from std to google. (csilvers)
Thu Nov 3 20:12:31 2005 Google Inc.
* sparsehash: version 0.3 release
* Quiet compiler warnings on some compilers. (csilvers)
* Some documentation fixes: example code for dense_hash_map. (csilvers)
* Fix a bug where swap() wasn't swapping delete_key(). (csilvers)
* set_deleted_key() and set_empty_key() now take a key only,
allowing hash-map values to be forward-declared. (csilvers)
* support for std::insert_iterator (and std::inserter). (csilvers)
Sun Apr 23 22:42:35 2006 Google Inc.
* sparsehash: version 0.4 release
* Remove POD requirement for keys and values! (austern)
* Add tr1-compatible type-traits system to speed up POD ops. (austern)
* Fixed const-iterator bug where postfix ++ didn't compile. (csilvers)
* Fixed iterator comparison bugs where <= was incorrect. (csilvers)
* Clean up config.h to keep its #defines from conflicting. (csilvers)
* Big documentation sweep and cleanup. (csilvers)
* Update documentation to talk more about good hash fns. (csilvers)
* Fixes to compile on MSVC (working around some MSVC bugs). (rennie)
* Avoid resizing hashtable on operator[] lookups (austern)
implementation. with 2 bits/entry overhead!
The Google SparseHash project contains several hash-map implementations in use
at Google, with different performance characteristics, including an
implementation that optimizes for space and one that optimizes for speed.
SparseHash is a template library; there are no binaries to install.
SparseHash is distributed under the terms of the BSD License.
