Fitsio

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

Bug Fixes
- Bug in repr of FITS where it only worked the first time it was generated

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

Bug Fixes

- Bug converting extension names, should just ensure ascii rather than
subset of ascii
- Bugs in the repr of headers where history keywords had extra spaces.
Also a bug with not upper casing things like 1.0E20 which is the
fits standard

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

Bug Fixes

- fixed bug where Updating an existing record in an image header raises an
exception (user ussegliog)
- fix bug append not forwarding arguments to write (Nicolas Tessore)

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

Bug Fixes

- Deal with case that a header keyword has non-ascii
characters and the value is numerical. In this case
we cannot convert the value because fits_read_key
can segfault in some scenarios. We instead return
a string.
- Fixed residual segfaults for header cards with non-ascii
characters.
- HIERARCH keywords are now properly parsed.
- Header keywords with `*`, `` or `?` in them now properly raise an error
before writing since the generated FITS files cannot be read.

Changes

- Non-allowed characters in header keywords are now converted to `_` instead
of `JUNK___...`.

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

New Features

- Moved most testing to GitHub actions (linux, osx).
- Added testing on ppc64le w/ TravisCI (thanks asellappen)

Bug Fixes

- Don't remove BLANK keywords in header clean
- Preserve order of comments in header

Compatibility changes

- moved to sing `bool` rather than `np.bool` to be compatible
with numpy 1.2

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

This release moves to cfitsio 3.49, which has bug fixes and now properly
supports reading certain classes of lossless compressed files

New Features

- Added keywords to control compression
- qlevel control the quantization level
- qmethod set the quantization method
- hcomp_scale, hcomp_smooth HCOMPRESS specific settings

A nice result of this is that one can do lossless gzip compression
(setting qlevel=0) and
- Work around some types of garbage characters that might appear
in headers

BACKWARDS INCOMPATIBLE CHANGES

- non-ascii junk in headers is replaced by ascii characters to
avoid segmentation faults in the python standard library
when non-unicode characters are detected. This will cause
codes that check for consistency between copied headers
to fail, since the header data is modified.

Bug Fixes

- Write integer keywords using the long long support rather than long
- Fix bug where a new file is started and the user can access a
fictional HDU, causing book keeping problems
- Return zero length result when requested rows have
zero length (rainwoodman)