Fitsio

---------------

Bug Fixes

- Fix variable length string column copying in python 3
- Fix bug checking for max size in a variable length table column.
- Raise an exception when writing to a table with data
that has shape ()
- exit test suite with non-zero exit code if a test fails

Continuous integration

- the travis ci now runs unit tests, ignoring those that may fail
when certain libraries/headers are not installed on the users system (for
now this is only bzip2 support)
- only particular pairs of python version/numpy version are tested

python3 compatibility

- the compatibility is now built into the code rather than
using 2to3 to modify code at install time.

Workarounds

- It turns out that when python, numpy etc. are compiled with gcc 4*
and fitsio is compiled with gcc 5* there is a problem, in some cases,
reading from an array with not aligned memory. This has to do with using
the -O3 optimization flag when compiling cfitsio. For replacing -O3 with
-O2 fixes the issue. This was an issue on linux in both anaconda python2
and python3.

----------------------------------

New tag so that pypi will accept the updated version

----------------------------------

New Features

- header_start, data_start, data_end now available in the
info dictionary, as well as the new get_offsets() method
to access these in a new dict.
(thanks Dimitri Muna for the initial version of this)

Bug Fixes

- Fix bug when writing new COMMENT fields (thanks Alex Drlica-Wagner for
initial fix)
- deal correctly with aligned data in some scenarios
(thanks Ole Streicher)
- use correct data type long for tile_dims_fits in
the set_compression C code. This avoids a crash
on 32 but systems. (thanks Ole Streicher)
- use correct data type npy_int64 for pointer in
get_long_slices (this function is not not correctly
named). Avoids crash on some 32 bit systems.
(thanks Ole Streicher)
- use correct data type npy_int64 for pointer in
PyFITSObject_create_image_hdu, rather than npy_intp.
(thanks Ole Streicher)

----------------------------------

New Features

- added read_scamp_head function to read the .head files output
by SCAMP and return a FITSHDR object
- reserved header space when creating image and table extensions
and a header is being written. This can improve performance
substantially, especially on distributed file systems.
- When possible write image data at HDU creation. This can
be a big performance improvement, especially on distributed file
systems.
- Support for reading bzipped FITS files. (Dustin Lang)

- Added option to use the system CFITSIO instead of the bundled one,
by sending --use-system-fitsio. Strongly recommend only use cfitsio
that are as new as the bundled one. Also note the bundled cfitsio
sometimes contains patches that are not yet upstream in an
official cfitsio release
- proper support for reading unsigned images compressed with PLIO.
This is a patch directly on the cfitsio code base. The same
code is in the upstream, but not yet released.
- New method reshape(dims) for images
- When writing into an existing image HDU, and larger dimensions
are required, the image is automatically expanded.

Bug Fixes

- Fixed broken boolean fields in new versions of numpy (rainwoodman) Fixed
- bug when image was None (for creating empty first HDU) removed -iarch in
- setup.py for mac OS X. This should
work for versions Mavericks and Snow Leapard (Christopher Bonnett)
- Reading a single string column was failing in some cases, this
has been fixed
- When creating a TableColumnSubset using [cols], the existence
of the columns is checked immediately, rather than waiting for the
check in the read()
- make sure to convert correct endianness when writing during image HDU
creation
- Corrected the repr for single column subsets
- only clean bzero,bscale,bunit from headers for TableHDU

Dev features

- added travis ci

----------------------------------

New Features

- python 3 compatibility
- Adding a new HDU is now near constant time
- Can now create an empty image extension using create_image_hdu
and sending the dims= and dtype= keywords
- Can now write into a sub-section of an existing image using the
start= keyword.
- Can now use a scalar slice for reading images, e.g.
hdu[row1:row2, col]
although this still currently retains the extra dimension
- Use warnings instead of printing to stdout
- IOError is now used to indicate a number of errors that
were previously ValueError

--------------

New Features

- use cfitsio 3370 to support new tile compression features
- FITSRecord class to encapsulate all the ways one can represent header
records. This is now used internally in the FITSHDR class instead of raw
dicts, but as FITSRecord inherits from dict this should be transparent.
- FITSCard class; inherits from FITSRecord and is a special case for header
card strings
- One can directly add a fits header card string to the FITSHDR object
using add_record

Bug Fixes

- use literal_eval instead of eval for evaluating header values (D. Lang)
- If input to write_keys is a FITSHDR, just use it instead of creating a
new FITSHDR object. (D. Lang)
- update existing keys when adding records to FITSHDR, except for
comment and history fields.
- fixed bug with empty string in header card
- deal with cfitsio treating first 4 comments specially