PyFITS release notes

PyFITS is being used within the STScI operations pipelines and is a supported product. Currently PyFITS takes a fairly strict interpretation of FITS files. There are likely to be problems with FITS data that do not strictly conform to the standard. Undoubtedly, use with a wider variety of files and uses will uncover problems; please contact help@stsci.edu when such problems are encountered.

We intend to accomodate such variances, particularly if they involve widely used or available data (so please let us know when such problems occur).

PyFITS requires numpy (or numarray) to be installed.

Version 2.2.2; October 12 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following bugs were fixed:
- Corrected a bug that caused an exception to be raised when creating a CompImageHDU using an initial header that does not match the image data in terms of the number of axis.

Version 2.2.1; October 6 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following bugs were fixed:
- Corrected a bug that prevented the opening of a fits file where a header contained a CHECKSUM card but no DATASUM card.
- Corrected a bug that caused NULLs to be written instead of blanks when an ASCII table was created using a numpy chararray in which the original data contained trailing blanks. (CNSHD695419)

Version 2.2; September 23 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following enhancements were made:
- Provide support for the FITS Checksum Keyword Convention. (CNSHD754301)
  - Adding the checksum=True keyword argument to the open convenience function will cause checksums to be verified on file open.
```
hdul=pyfits.open('in.fits',checksum=True)
```
  - On output, CHECKSUM and DATASUM cards may be output to all HDU's in a fits file by using the keyword argument checksum=True in calls to the writeto convenience function, the HDUList.writeto method, the writeto methods of all of the HDU classes, and the append convenience function.
```
hdul.writeto('out.fits',checksum=True)
```
- Implemented a new insert method to the HDUList class that allows for the insertion of a HDU into a HDUList at a given index.
```
   hdul.insert(2,hdu)
```
- Provided the capability to handle unicode input for file names.
- Provided support for integer division required by Python 3.0.
The following bugs were fixed:
- Corrected a bug that caused an index out of bounds exception to be raised when iterating over the rows of a binary table HDU using the syntax " for row in tbhdu.data: ". (CNSHD748609)
- Corrected a bug that prevented the use of the writeto convenience function for writing table data to a file. (CNSHD749024)
- Modified the code to raise an IOError exception with the comment "Header missing END card." when pyfits can't find a valid END card for a header when opening a file.
  - This change addressed a problem with a non-standard fits file that contained several new-line characters at the end of each header and at the end of the file. However, since some people want to be able to open these non-standard files anyway, an option was added to the open convenience function to allow these files to be opened without exception.
```
pyfits.open('infile.fits',ignore_missing_end=True)
```
- Corrected a bug that prevented the use of StringIO objects as fits files when reading and writing table data. Previously, only image data was supported. (CNSHD753698)
- Corrected a bug that caused a bus error to be generated when compressing image data using GZIP_1 under the Solaris operating system.
- Corrected bugs that prevented pyfits from properly reading Random Groups HDU's using numpy. (CNSHD756570)
- Corrected a bug that can occur when writing a fits file. (CNSHD757508)
  - If no default SIGINT signal handler has not been assigned, before the write, a TypeError exception is raised in the _File.flush() method when attempting to return the signal handler to its previous state. Notably this occurred when using mod_python. The code was changed to use SIG_DFL when no old handler was defined.
- Corrected a bug in CompImageHDU that prevented rescaling the image data using hdu.scale(option='old').

Version 2.1.1; April 22 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following bugs were fixed:
- Corrected a bug that caused an exception to be raised when closing a file opened for append, where an HDU was appended to the file, after data was accessed from the file. This exception was only raised when running on a Windows platform.
- Updated the installation scripts, compression source code, and benchmark test scripts to properly install, build, and execute on a Windows platform.

Version 2.1; April 14 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following enhancements were made:
- Added new tdump and tcreate capabilities to pyfits.
  - The new tdump convenience function allows the contents of a binary table HDU to be dumped to a set of three files in ASCII format. One file will contain column definitions, the second will contain header parameters, and the third will contain header data.
  - The new tcreate convenience function allows the creation of a binary table HDU from the three files dumped by the tdump convenience function.
  - The primary use for the tdump/tcreate methods are to allow editing in a standard text editor of the binary table data and parameters.
- Added support for case sensitive values of the EXTNAME card in an extension header. (CNSHD745784)
  - By default, pyfits converts the value of EXTNAME cards to upper case when reading from a file. A new convenience function (setExtensionNameCaseSensitive) was implemented to allow a user to circumvent this behavior so that the EXTNAME value remains in the same case as it is in the file.
  - With the following function call, pyfits will maintain the case of all characters in the EXTNAME card values of all extension HDU's during the entire python session, or until another call to the function is made:
```
>>> import pyfits
>>> pyfits.setExtensionNameCaseSensitive()
```
  - The following function call will return pyfits to its default (all upper case) behavior:
```
>>> pyfits.setExtensionNameCaseSensitive(False)
```
- Added support for reading and writing FITS files in which the value of the first card in the header is 'SIMPLE=F'. In this case, the pyfits open function returns an HDUList object that contains a single HDU of the new type _NonstandardHDU. The header for this HDU is like a normal header (with the exception that the first card contains SIMPLE=F instead of SIMPLE=T). Like normal HDU's the reading of the data is delayed until actually requested. The data is read from the file into a string starting from the first byte after the header END card and continuing till the end of the file. When written, the header is written, followed by the data string. No attempt is made to pad the data string so that it fills into a standard 2880 byte FITS block. (CNSHD744730)
- Added support for FITS files containing extensions with unknown XTENSION card values. (CNSHD744730) Standard FITS files support extension HDU's of types TABLE, IMAGE, BINTABLE, and A3DTABLE. Accessing a nonstandard extension from a FITS file will now create a _NonstandardExtHDU object. Accessing the data of this object will cause the data to be read from the file into a string. If the HDU is written back to a file the string data is written after the Header and padded to fill a standard 2880 byte FITS block.
The following bugs were fixed:
- Extensive changes were made to the tiled image compression code to support the latest enhancements made in CFITSIO version 3.13 to support this convention.
- Eliminated a memory leak in the tiled image compression code.
- Corrected a bug in the FITS_record.__setitem__ method which raised a NameError exception when attempting to set a value in a FITS_record object. (CNSHD745844)
- Corrected a bug that caused a TypeError exception to be raised when reading fits files containing large table HDU's (>2Gig). (CNSHD745522)
- Corrected a bug that caused a TypeError exception to be raised for all calls to the warnings module when running under Python 2.6. The formatwarning method in the warnings module was changed in Python 2.6 to include a new argument. (CNSHD746592)
- Corrected the behavior of the membership (in) operator in the Header class to check against header card keywords instead of card values. (CNSHD744730)
- Corrected the behavior of iteration on a Header object. The new behavior iterates over the unique card keywords instead of the card values.

Version 2.0.1; February 3 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following bugs were fixed:
- Eliminated a memory leak when reading Table HDU's from a fits file. (CNSHD741877)

Version 2.0; January 30 2009

Updates described in this release are only supported in the NUMPY version of pyfits.
The following enhancements were made:
- Provide initial support for an image compression convention known as the "Tiled Image Compression Convention".
  - The principle used in this convention is to first divide the n-dimensional image into a rectangular grid of subimages or "tiles". Each tile is then compressed as a continuous block of data, and the resulting compressed byte stream is stored in a row of a variable length column in a FITS binary table. Several commonly used algorithms for compressing image tiles are supported. These include, GZIP, RICE, H-Compress and IRAF pixel list (PLIO).
  - Support for compressed image data is provided using the optional "pyfitsComp" module contained in a C shared library (pyfitsCompmodule.so).
  - The header of a compressed image HDU appears to the user like any image header. The actual header stored in the FITS file is that of a binary table HDU with a set of special keywords, defined by the convention, to describe the structure of the compressed image. The conversion between binary table HDU header and image HDU header is all performed behind the scenes. Since the HDU is actually a binary table, it may not appear as a primary HDU in a FITS file.
  - The data of a compressed image HDU appears to the user as standard uncompressed image data. The actual data is stored in the fits file as Binary Table data containing at least one column (COMPRESSED_DATA). Each row of this variable-length column contains the byte stream that was generated as a result of compressing the corresponding image tile. Several optional columns may also appear. These include, UNCOMPRESSED_DATA to hold the uncompressed pixel values for tiles that cannot be compressed, ZSCALE and ZZERO to hold the linear scale factor and zero point offset which may be needed to transform the raw uncompressed values back to the original image pixel values, and ZBLANK to hold the integer value used to represent undefined pixels (if any) in the image.
  - To create a compressed image HDU from scratch, simply construct a CompImageHDU object from an uncompressed image data array and its associated image header. From there, the HDU can be treated just like any image HDU.
```
>>> hdu=pyfits.CompImageHDU(imageData,imageHeader)
>>> hdu.writeto('compressed_image.fits')
```
  - The signature for the CompImageHDU initializer method describes the possible options for constructing a CompImageHDU object:
```
  def __init__(self, data=None, header=None, name=None,
               compressionType='RICE_1',
               tileSize=None,
               hcompScale=0.,
               hcompSmooth=0,
               quantizeLevel=16.):
  """data:            data of the image
     header:          header to be associated with the
                      image
     name:            the EXTNAME value; if this value
                      is None, then the name from the
                      input image header will be used;
                      if there is no name in the input
                      image header then the default name
                      'COMPRESSED_IMAGE' is used
     compressionType: compression algorithm 'RICE_1',
                      'PLIO_1', 'GZIP_1', 'HCOMPRESS_1'
     tileSize:        compression tile sizes default
                      treats each row of image as a tile
     hcompScale:      HCOMPRESS scale parameter
     hcompSmooth:     HCOMPRESS smooth parameter
     quantizeLevel:   floating point quantization level;
  """
```
- Added two new convenience functions. The setval function allows the setting of the value of a single header card in a fits file. The delval function allows the deletion of a single header card in a fits file.
- A modification was made to allow the reading of data from a fits file containing a Table HDU that has duplicate field names. It is normally a requirement that the field names in a Table HDU be unique. Prior to this change a ValueError was raised, when the data was accessed, to indicate that the HDU contained duplicate field names. Now, a warning is issued and the field names are made unique in the internal record array. This will not change the TTYPEn header card values. You will be able to get the data from all fields using the field name, including the first field containing the name that is duplicated. To access the data of the other fields with the duplicated names you will need to use the field number instead of the field name. (CNSHD737193)
- An enhancement was made to allow the reading of unsigned integer 16 values from an ImageHDU when the data is signed integer 16 and BZERO is equal to 32784 and BSCALE is equal to 1 (the standard way for scaling unsigned integer 16 data). A new optional keyword argument (uint16) was added to the open convenience function. Supplying a value of True for this argument will cause data of this type to be read in and scaled into an unsigned integer 16 array, instead of a float 32 array. If a HDU associated with a file that was opened with the uint16 option and containing unsigned integer 16 data is written to a file, the data will be reverse scaled into an integer 16 array and written out to the file and the BSCALE/BZERO header cards will be written with the values 1 and 32768 respectively. (CHSHD736064) Reference the following example:
```
>>> import pyfits
>>> hdul=pyfits.open('o4sp040b0_raw.fits',uint16=1)
>>> hdul[1].data
array([[1507, 1509, 1505, ..., 1498, 1500, 1487],
       [1508, 1507, 1509, ..., 1498, 1505, 1490],
       [1505, 1507, 1505, ..., 1499, 1504, 1491],
       ...,
       [1505, 1506, 1507, ..., 1497, 1502, 1487],
       [1507, 1507, 1504, ..., 1495, 1499, 1486],
       [1515, 1507, 1504, ..., 1492, 1498, 1487]], dtype=uint16)
>>> hdul.writeto('tmp.fits')
>>> hdul1=pyfits.open('tmp.fits',uint16=1)
>>> hdul1[1].data
array([[1507, 1509, 1505, ..., 1498, 1500, 1487],
       [1508, 1507, 1509, ..., 1498, 1505, 1490],
       [1505, 1507, 1505, ..., 1499, 1504, 1491],
       ...,
       [1505, 1506, 1507, ..., 1497, 1502, 1487],
       [1507, 1507, 1504, ..., 1495, 1499, 1486],
       [1515, 1507, 1504, ..., 1492, 1498, 1487]], dtype=uint16)
>>> hdul1=pyfits.open('tmp.fits')
>>> hdul1[1].data
array([[ 1507.,  1509.,  1505., ...,  1498.,  1500.,  1487.],
       [ 1508.,  1507.,  1509., ...,  1498.,  1505.,  1490.],
       [ 1505.,  1507.,  1505., ...,  1499.,  1504.,  1491.],
       ...,
       [ 1505.,  1506.,  1507., ...,  1497.,  1502.,  1487.],
       [ 1507.,  1507.,  1504., ...,  1495.,  1499.,  1486.],
       [ 1515.,  1507.,  1504., ...,  1492.,  1498.,  1487.]], dtype=float32)
```
- Enhanced the message generated when a ValueError exception is raised when attempting to access a header card with an unparsable value. The message now includes the Card name.
The following bugs were fixed:
- Corrected a bug that occurs when appending a binary table HDU to a fits file. Data was not being byteswapped on little endian machines. (CNSHD737243)
- Corrected a bug that occurs when trying to write an ImageHDU that is missing the required PCOUNT card in the header. An UnboundLocalError exception complaining that the local variable 'insert_pos' was referenced before assignment was being raised in the method _ValidHDU.req_cards. The code was modified so that it would properly issue a more meaningful ValueError exception with a description of what required card is missing in the header.
- Eliminated a redundant warning message about the PCOUNT card when validating an ImageHDU header with a PCOUNT card that is missing or has a value other than 0.

Version 1.4.1; November 4 2008

Updates described in this release are only supported in the NUMPY version of pyfits.
The following enhancements were made:
- Enhanced the way import errors are reported to provide more information.
The following bugs were fixed:
- Corrected a bug that occurs when a card value is a string and contains a colon but is not a record-valued keyword card.
- Corrected a bug where pyfits fails to properly handle a record-valued keyword card with values using exponential notation and trailing blanks.

Version 1.4; July 7 2008

Updates described in this release are only supported in the NUMPY version of pyfits.
The following enhancements were made:
- Added support for file objects and file like objects.
  - All convenience functions and class methods that take a file name will now also accept a file object or file like object. File like objects supported are StringIO and GzipFile objects. Other file like objects will work only if they implement all of the standard file object methods.
  - For the most part, file or file like objects may be either opened or closed at function call. An opened object must be opened with the proper mode depending on the function or method called. Whenever possible, if the object is opened before the method is called, it will remain open after the call. This will not be possible when writing a HDUList that has been resized or when writing to a GzipFile object regardless of whether it is resized. If the object is closed at the time of the function call, only the name from the object is used, not the object itself. The pyfits code will extract the file name used by the object and use that to create an underlying file object on which the function will be performed.
- Added support for record-valued keyword cards as introduced in the "FITS WCS Paper IV proposal for representing a more general distortion model".
  - Record-valued keyword cards are string-valued cards where the string is interpreted as a definition giving a record field name, and its floating point value. In a FITS header they have the following syntax:
```
keyword= 'field-specifier: float'
```
    where keyword is a standard eight-character FITS keyword name, float is the standard FITS ASCII representation of a floating point number, and these are separated by a colon followed by a single blank.
    The grammer for field-specifier is:
```
field-specifier:
    field
    field-specifier.field
```
```
field:
    identifier
    identifier.index
```
    where identifier is a sequence of letters (upper or lower case), underscores, and digits of which the first character must not be a digit, and index is a sequence of digits. No blank characters may occur in the field-specifier. The index is provided primarily for defining array elements though it need not be used for that purpose.
    Multiple record-valued keywords of the same name but differing values may be present in a FITS header. The field-specifier may be viewed as part of the keyword name.
    Some examples follow:
```
DP1     = 'NAXIS: 2'
DP1     = 'AXIS.1: 1'
DP1     = 'AXIS.2: 2'
DP1     = 'NAUX: 2'
DP1     = 'AUX.1.COEFF.0: 0'
DP1     = 'AUX.1.POWER.0: 1'
DP1     = 'AUX.1.COEFF.1: 0.00048828125'
DP1     = 'AUX.1.POWER.1: 1'
```
  - As with standard header cards, the value of a record-valued keyword card can be accessed using either the index of the card in a HDU's header or via the keyword name. When accessing using the keyword name, the user may specify just the card keyword or the card keyword followed by a period followed by the field-specifier. Note that while the card keyword is case insensitive, the field-specifier is not. Thus, hdu['abc.def'], hdu['ABC.def'], or hdu['aBc.def'] are all equivalent but hdu['ABC.DEF'] is not.
  - When accessed using the card index of the HDU's header the value returned will be the entire string value of the card. For example:
```
>>> print hdr[10]
NAXIS: 2
>>> print hdr[11]
AXIS.1: 1
```
  - When accessed using the keyword name exclusive of the field-specifier, the entire string value of the header card with the lowest index having that keyword name will be returned. For example:
```
>>> print hdr['DP1']
NAXIS: 2
```
  - When accessing using the keyword name and the field-specifier, the value returned will be the floating point value associated with the record-valued keyword card. For example:
```
>>> print hdr['DP1.NAXIS']
2.0
```
  - Any attempt to access a non-existent record-valued keyword card value will cause an exception to be raised (IndexError exception for index access or KeyError for keyword name access).
  - Updating the value of a record-valued keyword card can also be accomplished using either index or keyword name. For example:
```
>>> print hdr['DP1.NAXIS']
2.0
>>> hdr['DP1.NAXIS'] = 3.0
>>> print hdr['DP1.NAXIS']
3.0
```
  - Adding a new record-valued keyword card to an existing header is accomplished using the Header.update() method just like any other card. For example:
```
>>> hdr.update('DP1', 'AXIS.3: 1', 'a comment', after='DP1.AXIS.2')
```
  - Deleting a record-valued keyword card from an existing header is accomplished using the standard list deletion syntax just like any other card. For example:
```
>>> del hdr['DP1.AXIS.1']
```
  - In addition to accessing record-valued keyword cards individually using a card index or keyword name, cards can be accessed in groups using a set of special pattern matching keys. This access is made available via the standard list indexing operator providing a keyword name string that contains one or more of the special pattern matching keys. Instead of returning a value, a CardList object will be returned containing shared instances of the Cards in the header that match the given keyword specification.
  - There are three special pattern matching keys. The first key '*' will match any string of zero or more characters within the current level of the field-specifier. The second key '?' will match a single character. The third key '...' must appear at the end of the keyword name string and will match all keywords that match the preceding pattern down all levels of the field-specifier. All combinations of ?, *, and ... are permitted (though ... is only permitted at the end). Some examples follow:
```
>>> cl=hdr['DP1.AXIS.*']
>>> print cl
DP1     = 'AXIS.1: 1'
DP1     = 'AXIS.2: 2'
>>> cl=hdr['DP1.*']
>>> print cl
DP1     = 'NAXIS: 2'
DP1     = 'NAUX: 2'
>>> cl=hdr['DP1.AUX...']
>>> print cl
DP1     = 'AUX.1.COEFF.0: 0'
DP1     = 'AUX.1.POWER.0: 1'
DP1     = 'AUX.1.COEFF.1: 0.00048828125'
DP1     = 'AUX.1.POWER.1: 1'
>>> cl=hdr['DP?.NAXIS']
>>> print cl
DP1     = 'NAXIS: 2'
DP2     = 'NAXIS: 2'
DP3     = 'NAXIS: 2'
>>> cl=hdr['DP1.A*S.*']
>>> print cl
DP1     = 'AXIS.1: 1'
DP1     = 'AXIS.2: 2'
```
  - The use of the special pattern matching keys for adding or updating header cards in an existing header is not allowed. However, the deletion of cards from the header using the special keys is allowed. For example:
```
>>> del hdr['DP3.A*...']
```
  - As noted above, accessing pyfits Header object using the special pattern matching keys will return a CardList object. This CardList object can itself be searched in order to further refine the list of Cards. For example:
```
>>> cl=hdr['DP1...']
>>> print cl
DP1     = 'NAXIS: 2'
DP1     = 'AXIS.1: 1'
DP1     = 'AXIS.2: 2'
DP1     = 'NAUX: 2'
DP1     = 'AUX.1.COEFF.1: 0.000488'
DP1     = 'AUX.2.COEFF.2: 0.00097656'
>>> cl1=cl['*.*AUX...']
>>> print cl1
DP1     = 'NAUX: 2'
DP1     = 'AUX.1.COEFF.1: 0.000488'
DP1     = 'AUX.2.COEFF.2: 0.00097656'
```
  - The CardList keys() method will allow the retrivial of all of the key values in the CardList. For example:
```
>>> cl=hdr['DP1.AXIS.*']
>>> print cl
DP1     = 'AXIS.1: 1'
DP1     = 'AXIS.2: 2'
>>> cl.keys()
['DP1.AXIS.1', 'DP1.AXIS.2']
```
  - The CardList values() method will allow the retrivial of all of the values in the CardList. For example:
```
>>> cl=hdr['DP1.AXIS.*']
>>> print cl
DP1     = 'AXIS.1: 1'
DP1     = 'AXIS.2: 2'
>>> cl.values()
[1.0, 2.0]
```
  - Individual cards can be retrieved from the list using standard list indexing. For example:
```
>>> cl=hdr['DP1.AXIS.*']
>>> c=cl[0]
>>> print c
DP1     = 'AXIS.1: 1'
>>> c=cl['DP1.AXIS.2']
>>> print c
DP1     = 'AXIS.2: 2'
```
  - Individual card values can be retrieved from the list using the value attribute of the card. For example:
```
>>> cl=hdr['DP1.AXIS.*']
>>> cl[0].value
1.0
```
  - The cards in the CardList are shared instances of the cards in the source header. Therefore, modifying a card in the CardList also modifies it in the source header. However, making an addition or a deletion to the CardList will not affect the source header. For example:
```
>>> hdr['DP1.AXIS.1']
1.0
>>> cl=hdr['DP1.AXIS.*']
>>> cl[0].value = 4.0
>>> hdr['DP1.AXIS.1']
4.0
>>> del cl[0]
>>> print cl['DP1.AXIS.1']
Traceback (most recent call last):
File "", line 1, in 
File "NP_pyfits.py", line 977, in __getitem__
  return self.ascard[key].value
File "NP_pyfits.py", line 1258, in __getitem__
  _key = self.index_of(key)
File "NP_pyfits.py", line 1403, in index_of
  raise KeyError, 'Keyword %s not found.' % `key`
KeyError: "Keyword 'DP1.AXIS.1' not found."
>>> hdr['DP1.AXIS.1']
4.0
```
  - A FITS header consists of card images. In pyfits each card image is manifested by a Card object. A pyfits Header object contains a list of Card objects in the form of a CardList object. A record-valued keyword card image is represented in pyfits by a RecordValuedKeywordCard object. This object inherits from a Card object and has all of the methods and attributes of a Card object.
  - A new RecordValuedKeywordCard object is created with the RecordValuedKeywordCard constructor: RecordValuedKeywordCard(key, value, comment). The key and value arguments may be specified in two ways. The key value may be given as the 8 character keyword only, in which case the value must be a character string containing the field-specifier, a colon followed by a space, followed by the actual value. The second option is to provide the key as a string containing the keyword and field-specifier, in which case the value must be the actual floating point value. For example:
```
>>> c1 = pyfits.RecordValuedKeywordCard('DP1', 'NAXIS: 2', 'Number of variables')
>>> c2 = pyfits.RecordValuedKeywordCard('DP1.AXIS.1', 1.0, 'Axis number')
```
  - RecordValuedKeywordCards have attributes .key, .field_specifier, .value, and .comment. Both .value and .comment can be changed but not .key or .field_specifier. The constructor will extract the field-specifier from the input key or value, whichever is appropriate. The .key attribute is the 8 character keyword.
  - Just like standard Cards, a RecordValuedKeywordCard may be constructed from a string using the fromstring() method or verified using the verify() method. For example:
```
>>> c1 = pyfits.RecordValuedKeywordCard().fromstring(
         "DP1     = 'NAXIS: 2' / Number of independent variables")
>>> c2 = pyfits.RecordValuedKeywordCard().fromstring(
         "DP1     = 'AXIS.1: X' / Axis number")
>>> print c1; print c2
DP1     = 'NAXIS: 2' / Number of independent variables
DP1     = 'AXIS.1: X' / Axis number
>>> c2.verify()
Output verification result:
Card image is not FITS standard (unparsable value string).
```
  - A standard card that meets the criteria of a RecordValuedKeywordCard may be turned into a RecordValuedKeywordCard using the class method coerce. If the card object does not meet the required criteria then the original card object is just returned.
```
>>> c1 = pyfits.Card('DP1','AUX: 1','comment')
>>> c2 = pyfits.RecordValuedKeywordCard.coerce(c1)
>>> print type(c2)
<class 'pyfits.NP_pyfits.RecordValuedKeywordCard'>
```
  - Two other card creation methods are also available as RecordVauedKeywordCard class methods. These are createCard() which will create the appropriate card object (Card or RecordValuedKeywordCard) given input key, value, and comment, and createCardFromString which will create the appropriate card object given an input string. These two methods are also available as convenience functions.
```
>>> c1 = pyfits.RecordValuedKeywordCard.createCard('DP1','AUX: 1','comment)
or
>>> c1 = pyfits.createCard('DP1','AUX: 1','comment)
>>> print type(c1)
<class 'pyfits.NP_pyfits.RecordValuedKeywordCard'>

>>> c1 = pyfits.RecordValuedKeywordCard.createCard('DP1','AUX 1','comment)
or
>>> c1 = pyfits.createCard('DP1','AUX 1','comment)
>>> print type(c1)
<class 'pyfits.NP_pyfits.Card'>

>>> c1 = pyfits.RecordValuedKeywordCard.createCardFromString \
         ("DP1 = 'AUX: 1.0' / comment")
or
>>> c1 = pyfits.createCardFromString("DP1     = 'AUX: 1.0' / comment")
>>> print type(c1)
<class 'pyfits.NP_pyfits.RecordValuedKeywordCard'>
```
The following bugs were fixed:
- Corrected a bug that occurs when writing a HDU out to a file. During the write, any Keyboard Interrupts are trapped so that the write completes before the interrupt is handled. Unfortunately, the Keyboard Interrupt was not properly reinstated after the write completed. This was fixed. (CNSHD711138)
- Corrected a bug when using ipython, where temporary files created with the tempFile.NamedTemporaryFile method are not automatically removed. This can happen for instance when opening a Gzipped fits file or when open a fits file over the internet. The files will now be removed. (CNSHD718307)
- Corrected a bug in the append convenience function's call to the writeto convenience function. The classExtensions argument must be passed as a keyword argument.
- Corrected a bug that occurs when retrieving variable length character arrays from binary table HDUs (PA() format) and using slicing to obtain rows of data containing variable length arrays. The code issued a TypeError exception. The data can now be accessed with no exceptions. (CNSHD718749)
- Corrected a bug that occurs when retrieving data from a fits file opened in memory map mode when the file contains multiple image extensions or ASCII table or binary table HDUs. The code issued a TypeError exception. The data can now be accessed with no exceptions. (CNSHD707426)
- Corrected a bug that occurs when attempting to get a subset of data from a Binary Table HDU and then use the data to create a new Binary Table HDU object. A TypeError exception was raised. The data can now be subsetted and used to create a new HDU. (CNSHD723761)
- Corrected a bug that occurs when attempting to scale an Image HDU back to its original data type using the _ImageBaseHDU.scale method. The code was not resetting the BITPIX header card back to the original data type. This has been corrected.
- Changed the code to issue a KeyError exception instead of a NameError exception when accessing a non-existent field in a table.

Version 1.3; February 22 2008

Updates described in this release are only supported in the NUMPY version of pyfits.
The following enhancements were made:
- Provided support for a new extension to pyfits called stpyfits.
  - The stpyfits module is a wrapper around pyfits. It provides all of the features and functions of pyfits along with some STScI specific features. Currently, the only new feature supported by stpyfits is the ability to read and write fits files that contain image data quality extensions with constant data value arrays. See stpyfits for more details on stpyfits.
- Added a new feature to allow trailing HDUs to be deleted from a fits file without actually reading the data from the file.
  - This supports a JWST requirement to delete a trailing HDU from a file whose primary Image HDU is too large to be read on a 32 bit machine.
- Updated pyfits to use the warnings module to issue warnings. All warnings will still be issued to stdout, exactly as they were before, however, you may now suppress warnings with the -Wignore command line option. For example, to run a script that will ignore warnings use the following command line syntax:
```
python -Wignore yourscript.py
```
- Updated the open convenience function to allow the input of an already opened file object in place of a file name when opening a fits file.
- Updated the writeto convenience function to allow it to accept the output_verify option.
  - In this way, the user can use the argument output_verify='fix' to allow pyfits to correct any errors it encounters in the provided header before writing the data to the file.
- Updated the verification code to provide additional detail with a VerifyError exception.
- Added the capability to create a binary table HDU directly from a numpy.ndarray. This may be done using either the new_table convenience function or the BinTableHDU constructor.
The following performance improvements were made:
- Modified the import logic to dramatically decrease the time it takes to import pyfits.
- Modified the code to provide performance improvements when copying and examining header cards.
The following bugs were fixed:
- Corrected a bug that occurs when reading the data from a fits file that includes BZERO/BSCALE scaling. When the data is read in from the file, pyfits automatically scales the data using the BZERO/BSCALE values in the header. In the previous release, pyfits created a 32 bit floating point array to hold the scaled data. This could cause a problem when the value of BZERO is so large that the scaled value will not fit into the float 32. For this release, when the input data is 32 bit integer, a 64 bit floating point array is used for the scaled data.
- Corrected a bug that caused an exception to be raised when attempting to scale image data using the ImageHDU.scale method.
- Corrected a bug in the new_table convenience function that occurred when a binary table was created using a ColDefs object as input and supplying an nrows argument for a number of rows that is greater than the number of rows present in the input ColDefs object. The previous version of pyfits failed to allocate the necessary memory for the additional rows.
- Corrected a bug in the new_table convenience function that caused an exception to be thrown when creating an ASCII table.
- Corrected a bug in the new_table convenience function that will allow the input of a ColDefs object that was read from a file as a binary table with a data value equal to None.
- Corrected a bug in the construction of ASCII tables from Column objects that are created with noncontinuous start columns.
- Corrected bugs in a number of areas that would sometimes cause a failure to improperly raise an exception when an error occurred.
- Corrected a bug where attempting to open a non-existent fits file on a windows platform using a drive letter in the file specification caused a misleading IOError exception to be raised.

Version 1.1; June 15 2007

Modified to use either NUMPY or NUMARRAY.
New file writing modes have been provided to allow streaming data to extensions without requiring the whole output extension image in memory. See documentation on StreamingHDU.
Improvements to minimize byteswapping and memory usage by byteswapping in place.
Now supports ':' characters in filenames.
Handles keyboard interrupts during long operations.
Preserves the byte order of the input image arrays.

Version 1.0.1; March 24 2006

The changes to PyFITS were primarily to improve the docstrings and to reclassify some public functions and variables as private. Readgeis and fitsdiff which were distributed with PyFITS in previous releases were moved to pytools. This release of PyFITS is v1.0.1. The next release of PyFITS will support both numarray and numpy (and will be available separately from stsci_python, as are all the python packages contained within stsci_python). An alpha release for PyFITS numpy support will be made around the time of this stsci_python release.

Updated docstrings for public functions.
Made some previously public functions private.

Version 1.0; November 1 2005

Major Changes since v0.9.6:

Added support for the HEIRARCH convention
Added support for iteration and slicing for HDU lists
PyFITS now uses the standard setup.py installation script
Add utility functions at the module level, they include:

getheader
getdata
getval
writeto
append
update
info

Minor changes since v0.9.6:

Fix a bug to make single-column ASCII table work.
Fix a bug so a new table constructed from an existing table with X-formatted columns will work.
Fix a problem in verifying HDUList right after the open statement.
Verify that elements in an HDUList, besides the first one, are ExtensionHDU.
Add output verification in methods flush() and close().
Modify the the design of the open() function to remove the output_verify argument.
Remove the groups argument in GroupsHDU's contructor.
Redesign the column definition class to make its column components more accessible. Also to make it conducive for higher level functionalities, e.g. combining two column definitions.
Replace the Boolean class with the Python Boolean type. The old TRUE/FALSE will still work.
Convert classes to the new style.
Better format when printing card or card list.
Add the optional argument clobber to all writeto() functions and methods.
If adding a blank card, will not use existing blank card's space.

PyFITS Version 1.0 REQUIRES Python 2.3 or later.

Version 0.9.6; 11 November 2004

Major changes since v0.9.3:

Support for variable length array tables.
Support for writing ASCII table extensions.
Support for random groups, both reading and writing.

Some minor changes:

Support for numbers with leading zeros in an ASCII table extension.
Changed scaled columns' data type from Float32 to Float64 to preserve precision.
Made Column constructor more flexible in accepting format specification.

Version 0.9.3; 2 July 2004

Changes since v0.9.0:

Lazy instanciation of full Headers/Cards for all HDU's when the file is opened. At the open, only extracts vital info (e.g. NAXIS's) from the header parts. This change will speed up the performance if the user only needs to access one extension in a multi-extension FITS file.
Support the X format (bit flags) columns, both reading and writing, in a binary table. At the user interface, they are converted to Boolean arrays for easy manipulation. For example, if the column's TFORM is "11X", internally the data is stored in 2 bytes, but the user will see, at each row of this column, a Boolean array of 11 elements.
Fix a bug such that when a table extension has no data, it will not try to scale the data when updating/writing the HDU list.

Version 0.9; 27 April 2004

Changes since v0.8.0:

Rewriting of the Card class to separate the parsing and verification of header cards
Restructure the keyword indexing scheme which speed up certain applications (update large number of new keywords and reading a header with larger numbers of cards) by a factor of 30 or more
Change the default to be lenient FITS standard checking on input and strict FITS standard checking on output
Support CONTINUE cards, both reading and writing
Verification can now be performed at any of the HDUList, HDU, and Card levels
Support (contiguous) subsection (attribute .section) of images to reduce memory usage for large images

Version 0.8.0; August 19, 2003

NOTE: This version will only work with numarray Version 0.6. In addition, earlier versions of PyFITS will not work with numarray 0.6. Therefore, both must be updated simultaneously.

Changes since 0.7.6:

Compatible with numarray 0.6/records 2.0
For binary tables, now it is possible to update the original array if a scaled field is updated.
Support of complex columns
Modify the __getitem__ method in FITS_rec. In order to make sure the scaled quantities are also viewing ths same data as the original FITS_rec, all fields need to be "touched" when __getitem__ is called.
Add a new attribute mmobject for HDUList, and close the memmap object when close HDUList object. Earlier version does not close memmap object and can cause memory lockup.
Enable 'update' as a legitimate memmap mode.
Do not print message when closing an HDUList object which is not created from reading a FITS file. Such message is confusing.
remove the internal attribute "closed" and related method (__getattr__ in HDUList). It is redundant.

Version 0.7.6; November 22, 2002

NOTE: This version will only work with numarray Version 0.4.

Changes since 0.7.5:

Change x*=n to numarray.multiply(x, n, x) where n is a floating number, in order to make pyfits to work under Python 2.2. (2 occurrences)
Modify the "update" method in the Header class to use the "fixed-format" card even if the card already exists. This is to avoid the mis-alignment as shown below:
After running drizzle on ACS images it creates a CD matrix whose elements have very many digits, e.g.:
```
CD1_1   =  1.1187596304411E-05 / partial of first axis coordinate w.r.t. x
CD1_2   = -8.502767249350019E-06 / partial of first axis coordinate w.r.t. y
```
with pyfits, an "update" on these header items and write in new values which has fewer digits, e.g.:
```
CD1_1   =        1.0963011E-05 / partial of first axis coordinate w.r.t. x
CD1_2   =          -8.527229E-06 / partial of first axis coordinate w.r.t. y
```

Change some internal variables to make their appearance more consistent:

        old name                new name
        
        __octalRegex            _octalRegex
        __readblock()           _readblock()
        __formatter()           _formatter().
        __value_RE              _value_RE
        __numr                  _numr 
        __comment_RE            _comment_RE 
        __keywd_RE              _keywd_RE 
        __number_RE             _number_RE.
        tmpName()               _tmpName()
        dimShape                _dimShape
        ErrList                 _ErrList

Move up the module description. Move the copywright statement to the bottom and assign to the variable __credits__.
change the following line:
```
self.__dict__ = input.__dict__
```
to
```
self.__setstate__(input.__getstate__())
```
in order for pyfits to run under numarray 0.4.
edit _readblock to add the (optional) firstblock argument and raise IOError if the the first 8 characters in the first block is not 'SIMPLE ' or 'XTENSION'. Edit the function open to check for IOError to skip the last null filled block(s). Edit readHDU to add the firstblock argument.

Version 0.7.5; 16 August 2002

Changes since v0.7.3:

Memory mapping now works for readonly mode, both for images and binary tables.
Usage: pyfits.open('filename', memmap=1)
Edit the field method in FITS_rec class to make the column scaling for numbers use less temporary memory. (does not work under 2.2, due to Python "bug" of array *=)
Delete bscale/bzero in the ImageBaseHDU constructor.
Update bitpix in BaseImageHDU.__getattr__ after deleting bscale/bzero. (bug fix)
In BaseImageHDU.__getattr__ point self.data to raw_data if float and if not memmap. (bug fix).
Change the function get_tbdata() to private: _get_tbdata().

Version 0.7.3; 12 July 2002

Changes since v0.7.2:

It will scale all integer image data to Float32, if BSCALE/BZERO != 1/0. It will also expunge the BSCALE/BZERO keywords.
Add the scale() method for ImageBaseHDU, so data can be scaled just before being written to the file. It has the following arguments:

type:

destination data type (string), e.g. Int32, Float32, UInt8, etc.

option:

scaling scheme. if 'old', use the old BSCALE/BZERO values. if 'minmax', use the data range to fit into the full range of specified integer type. Float destination data type will not be scaled for this option.

bscale/bzero:

user specifiable BSCALE/BZERO values. They overwrite the "option".
Deal with data area resizing in 'update' mode.
Make the data scaling (both input and output) faster and use less memory.
Bug fix to make column name change takes effect for field.
Bug fix to avoid exception if the key is not present in the header already. This affects (fixes) add_history(), add_comment(), and add_blank().
Bug fix in __getattr__() in Card class. The change made in 0.7.2 to rstrip the comment must be string type to avoid exception.

Version 0.7.2.1; June 25, 2002

A couple of bugs were addressed in this version.

Fix a bug in _add_commentary(). Due to a change in index_of() during version 0.6.5.5, _add_commentary needs to be modified to avoid exception if the key is not present in the header already. This affects (fixes) add_history(), add_comment(), and add_blank().
Fix a bug in __getattr__() in Card class. The change made in 0.7.2 to rstrip the comment must be string type to avoid exception.

Version 0.7.2; June 19, 2002

The two major improvements from Version 0.6.2 are:

support reading tables with "scaled" columns (e.g. tscal/tzero, Boolean, and ASCII tables)
a prototype output verification.

Other changes include:

Implement the new HDU hierarchy proposed earlier this year. This in turn reduces some of the redundant methods common to several HDU classes.
Add 3 new methods to the Header class: add_history, add_comment, and add_blank.
The table attributes _columns are now .columns and the attributes in ColDefs are now all without the underscores. So, a user can get a list of column names by: hdu.columns.names.
The "fill" argument in the new_table method now has a new meaning:
If set to true (=1), it will fill the entire new table with zeros/blanks. Otherwise (=0), just the extra rows/cells are filled with zeros/blanks. Fill values other than zero/blank are now not possible.
Add the argument output_verify to the open method and writeto method. Not in the flush or close methods yet, due to possible complication.
A new copy method for tables, the copy is totally independent from the table it copies from.
The tostring() call in writeHDUdata takes up extra space to store the string object. Use tofile() instead, to save space.
Make changes from _byteswap to _byteorder, following corresponding changes in numarray and recarray.
Insert(update) EXTEND in PrimaryHDU only when header is None.
Strip the trailing blanks for the comment value of a card.
Add seek(0) right after the __buildin__.open(0), because for the 'ab+' mode, the pointer is at the end after open in Linux, but it is at the beginning in Solaris.
Add checking of data against header, update header keywords (NAXIS's, BITPIX) when they don't agree with the data.
change version to __version__.

Version 0.6.2; February 12, 2002

This version requires numarray version 0.2.

Things not yet supported but are part of future development:

Verification and/or correction of FITS objects being written to disk so that they are legal FITS. This is being added now and should be available in about a month. Currently, one may construct FITS headers that are inconsistent with the data and write such FITS objects to disk. Future versions will provide options to either a) correct discrepancies and warn, b) correct discrepancies silently, c) throw a Python exception, or d) write illegal FITS (for test purposes!).
Support for ascii tables or random groups format. Support for ASCII tables will be done soon (~1 month). When random group support is added is uncertain.
Support for memory mapping FITS data (to reduce memory demands). We expect to provide this capability in about 3 months.
Support for columns in binary tables having scaled values (e.g. BSCALE or BZERO) or boolean values. Currently booleans are stored as Int8 arrays and users must explicitly convert them into a boolean array. Likewise, scaled columns must be copied with scaling and offset by testing for those attributes explicitly. Future versions will produce such copies automatically.
Support for tables with TNULL values. This awaits an enhancement to numarray to support mask arrays (planned). (At least a couple of months off).