TPIE 1.1 release

New features in TPIE 1.1:

The pipelining framework makes it easier to implement efficient
streaming algorithms. For more information about the pipelining
framework, see doc/pipelining.dox, or read the documentation on the TPIE
homepage.

The serialization framework provides streaming primitives for working
with item streams where items vary in size. A particular practical use
is for streams of polygons or polygonal line segments. We use the term
`serialization stream', because it requires that items are
(un)serializable (rather than trivially copyable, which is the
requirement in file_stream).

If you have previously run off the TPIE master branch, notice that we
have changed the pipelining API recently to be more internally
consistent.

Other changes since TPIE 1.0:

To get the number of workers used by the TPIE job framework, use
tpie::default_worker_count(). (65c1de4)

We now support Boost filesystem version 3 (34ac4be).

The array implementation has changed, removing segmented_array which
nobody used anyway. This makes the array implementation much simpler and
less error-prone. The TPIE array now supports custom allocators, and it
has specialized std::copy so that an efficient memcpy implementation (or
similar) may be used automatically. (fad27fe and 2f79196)

The memory manager now has switched from a protected integer to an
atomic primitive to keep track of used memory, speeding up non-debug
builds. (7a4189a)

The progress_indicator_arrow class no longer sends unnecessary updates
to the standard output stream. (cc15583)

The new memory manager enforcement level ENFORCE_DEBUG functions like
ENFORCE_WARN, except that memory overusage warnings are sent to
log_warning() rather than log_debug(). (8394479)

TPIE 1.1 Release Candidate 1

New features in TPIE 1.1:

The pipelining framework makes it easier to implement efficient streaming
algorithms. For more information about the pipelining framework, see
doc/pipelining.dox, or read the documentation on the TPIE homepage.

The serialization framework provides streaming primitives for working with item
streams where items vary in size. A particular practical use is for streams
of polygons or polygonal line segments. We use the term `serialization stream',
because it requires that items are (un)serializable (rather than trivially
copyable, which is the requirement in file_stream).

If you have previously run off the TPIE master branch, notice that we have
changed the pipelining API recently to be more internally consistent.

Other changes since TPIE 1.0:

To get the number of workers used by the TPIE job framework,
use tpie::default_worker_count(). (65c1de4)

We now support Boost filesystem version 3 (34ac4be).

The array implementation has changed, removing segmented_array which nobody
used anyway. This makes the array implementation much simpler and less
error-prone. The TPIE array now supports custom allocators, and it has
specialized std::copy so that an efficient memcpy implementation (or similar)
may be used automatically. (fad27fe and 2f79196)

The memory manager now has switched from a protected integer to an atomic
primitive to keep track of used memory, speeding up non-debug builds. (7a4189a)

The progress_indicator_arrow class no longer sends unnecessary updates to the
standard output stream. (cc15583)

The new memory manager enforcement level ENFORCE_DEBUG functions like
ENFORCE_WARN, except that memory overusage warnings are sent to log_warning()
rather than log_debug(). (8394479)

This marks the release of TPIE version 1.0.

The following improvements have been merged in since 1.0rc2:

* Time estimation warning has been moved to debug log. (8a9138e)
* Priority queue memory usage slightly improved. (fa1e12e)
* Move backtrace printed by fractional_subindicator to debug log. (f016d26)
* Fix memory management crashes in debug mode. (91b940b)
* Get rid of CMake option TPIE_THREADSAFE_MEMORY_MANAGEMNT. (21eb5b8)
* Various documentation cleanup; code comments have been added/clarified.
* Some compiler warnings have been addressed.

The following bugs have been fixed since v1.0rc1.

* Priority queue failing with a tpie::invalid_file_exception on large datasets. (0d22e10)
* Priority queue correctness error on large datasets. (f209097)
* tpie_name and tpie_dir_name failing rarely when the attempted temporary name was already in use. (fcc2863)
* Memory manager warning when array::swap was used. (9496caa)
* External queue not using the correct file::stream protocol. (0a5b9ab)
* Several unhelpful compiler warnings inside TPIE have been muted. (f51b069)
* Memory manager now only warns about memory overusage in exponential steps. (d98736e)

Improvements:

* If boost::thread reports 4 or more CPUs, use one less worker thread than the number of CPUs. (8cee7dc)
* progress_indicator_arrow now displays breadcrumbs from fractional progress indicators. (fe47191)

Release candidate 1 of TPIE 1.0

Hopefully fix default temp path under windows

Account for the extra buffer used in remove_group_buffer. This method…

… is rarely called and the extra memory it used were not present in the memory calculations as far as I can see. This hopefully fixes a bunch of crashers that I've had with the pq, I'm currently running extensive tests.

Rolled back accidental commit in a branch.

Moved to tpie::ami namespace.

Include config.h correctly