Version 3.29 - 2 December 2011
Enhancements
- modified Makefile.in to allow configure to override the lib and include
destination directories.
- added (or restored actually) support for tile compression of 1-byte integer
images in imcomp_compress_tile. Support for that data type was overlooked
during recent updates to this routine.
- modified the fits_get_token command-line parsing routine to perform more
rigorous checks to determine if the token can be interpreted as a number
or not.
- made small modification to fpack.c to not allow the -i2f option (convert
image from integer to floating point) with the "-g -q 0" option (do lossless
gzip compression). It is more efficient to simply use the -g option alone.
- made modifications to fitsio.h and drvrfile.c to support reading and
writing large FITS files (> 2.1 GB) when building CFITSIO using
Microsoft Visual C++ on Windows platforms.
- added new WCS routine (ffgicsa) which returns the WCS keyword values
for a particular WCS version ('A' - 'Z').
Bug Fixes
- fixed a problem with multi-threaded apps that open/close FITS files
simultaneously by putting mutex locks around the call to
fits_already_open and in fits_clear_Fptr.
- fixed a bug when using the 'regfilter' function to select a subset of the
rows in a FITS table that have coordinates that lie within a specified
spatial region on the sky. This bug only affects the rarely used panda
(and epanda and bpanda) region shapes in which the region is defined by
the intersection of an annulus and a pie-shaped wedge. The previous code
(starting with version 3.181 of CFITSIO where support for the panda region
was first introduced) only worked correctly if the 2 angles that define
the wedge have values between -180 and +180. If not, then fewer rows than
expected may have been selected from the table.
- fixed the extended filename parser so that when creating a histogram by
binning 2 table columns, if a keyword or column name is given as the
weighting factor, then the output histrogram image will have a floating
point datatype, not the default integer datatype as is the case when no
weigth is specified (e.g. with a filename like
"myfile.fits[bin x,y; weight_column]"
- added fix to the code in imcompress.c to work around a problem with
dereferencing the value of a pointer, in cases where the address of
that pointer has not been defined (e.g., the nulval variable).
- modified the byte shuffling algorithm in fits_shuffle_8bytes to work
around a strange bug in the proprietary SunStudioExpress C compiler
under OpenSolaris.
- removed spurious messages on the CFITSIO error stack when opening a
FITS file with FTP (in drvrnet.c);
Version 3.28 - 12 May 2011
- added an enhancement to the tiled-image compression method when compressing
floating-point image using the standard (lossy) quantization method. In
cases where an image tile cannot be quantized, The floating-point pixel values
will be losslessly compressed with gzip before writing them to the tile-
compressed file. Previously, the uncompressed pixel values would have
been written to the file, which obviously requires more disk space.
- made significant internal changes to the structure of the tile compression
and uncompression routines in imcompress.c to make them more modular and
easier to maintain.
- modified configure.in and configure to force it to build a Universal
binary on Mac OS X.
- modified the ffiter function in putcol.c to properly clean up allocated
memory if an error occurs.
- in quantize.c, when searching for the min and max values in a float array,
initialize the max value to -FLT_MAX instead of FLT_MIN (and similarly
for double array).
Version 3.27 - 3 March 2011
Enhancements
- added new routines fits_read_str and fits_delete_str which read or
delete, respectively, a header keyword record that contains a specified
character string.
- added a new routine call fits_free_key_longstr which frees the memory
that fits_read_key_longstr allocated for the long string keyword value.
- enhanced the ffmkky routine in fitscore.c to not put a space before the
equals sign when writing long string-valued keywords using the ESO
HIERARCH keyword convension, if that extra character is needed to
fit the length of the keyword name + value string within the 80-character
FITS keyword record.
- made small change to fits_translate_keyword to support translation of
blank keywords (where the name = 8 blank chracters)
- modified fpack so that it uses the minimum of the 2nd, 3rd, and 5th order
MAD noise values when quantizing and compressing a floating point image.
This is more conservative than just using the 3rd order MAD value alone.
- added new routine imcomp_copy_prime2img to imcompress.c that is used by
funpack to copy any keywords that may have been added to the primary
array of the compressed image file (a null image) back into the header of
the uncompressed image.
- enhanced the fits_quantize_float and fits_quantize_double routines in
quantize.c to also compress the tile if it is completely filled with
null values. Previously, this type of tile would have been written
to the output compressed image without any compression.
- enhanced imcomp_decompress_tile to support implicit datatype conversion
when reading a losslessly compressed (with gzip) real*4 image into an
array of real*8 values.
Version 3.26 - 30 December 2010
Enhancements
- defined 2 new macros in fitsio.h:
#define CFITSIO_MAJOR 3
#define CFITSIO_MINOR 26
These may be used within other macros to detect the CFITSIO
version number at compile time.
- modified group.c to initialize the output URL to a null string in
fits_url2relurl. Also added more robust tests to see if 2 file
pointers point to the same file.
- enhanced the template keyword parsing code in grparser.c to support
the 'D' exponent character in the ASCII representation of floating
point keyword values (as in TVAL = 1.23D03). Previously, the parser
would have writen this keyword with a string value (TVAL = '1.23D03').
- modified the low-level routines that write a keyword record to a FITS
header so that they silently replace any illegal characters (ASCII
values less than 32 or greater than 126) with an ASCII space character.
Previously, these routines would have returned with an error when
encountering these illegal characters in the keyword record (most commonly
tab, carriage return, and line feed characters).
- made substantial internal changes to imcompress.c in preparation for
possible future support for compression methods for FITS tables analogous
to the tiled image compression method.
- replaced all the source code in CFITSIO that was distributed under the
GNU General Public License with freely available code. In particular,
the gzip file compression and uncompression code was replaced by the
zlib compression library. Thus, beginning with this version 3.26 of CFITSIO,
other software applications may freely use CFITSIO without necessarily
incurring any GNU licensing requirement. See the License.txt file for
the CFITSIO licensing requirements.
- added support for using cfitsio in different 'locales' which use a
comma, not a period, as the decimal point character in ASCII
representation of a floating point number (e.g., France). This
affects how floating point keyword values and floating point numbers
in ASCII tables are read and written with the 'printf' and 'strtod'
functions.
- added a new utility routine called fits_copy_rows/ffcprw that copies
a specified range of rows from one table to another.
- enhanced the test for illegal ASCII characters in a header (fftrec) to
print out the name of the offending character (e.g TAB or Line Feed) as
well as the Hex value of the chracter.
- modified ffgtbc (in fitscore.c) to support nonstandard vector variable
length array columns in binary tables (e.g. with TFORMn = 2000PE(500)').
- modified the configure file to add "-lm" when linking CFITSIO on
Solaris machines.
- added new routine, fits_get_inttype, to parse an integer keyword value
string and return the minimum integer datatype (TBYTE, TSHORT, TLONG,
TLONGLONG) required to store the integer value.
- added new routine, fits_convert_hdr2str, which is similar to fits_hdr2str
except that if the input HDU is a tile compressed image (stored
in a binary table) then it will first convert that header back to
that of a normal uncompressed FITS image before concatenating the header
keyword records.
- modified the file template reading routine (ngp_line_from_file in
grparser.c) so that it ignores any carriage return characters (\r)
in the line, that might be present, e.g. if the file was created on a
Windows machine that uses \r\n as end of line characters.
- modified the ffoptplt routine in cfileio.c to check if the PCOUNT
keyword in the template file has a non-zero value, and if so, resets
it to zero in the newly created file.
Bug Fixes
- fixed a bug when uncompressing floating-point images that contain Nan
values on some 64-bit platforms.
- fixed a bug when updating the value of the CRPIXn world coordinate
system keywords when extracting a subimage from larger FITS image, using the
extended CFITSIO syntax (e.g. myimage[1:500:2, 1:500:2]). This bug only
affects casee where the pixel increment value is not equal to 1, and caused
the coordinate grid to be shifted by between 0.25 pixels (in the case of
a pixel increment of 2) and 0.5 pixels (for large pixel increment values).
- fixed a potential string buffer overflow error in the ffmkls routine
that modifies the value and comment strings in a keyword that uses
the HEASARC long string keyword convention.
- fixed a bug in imcompress.c that could cause programs to abort on 64-bit
machines when using gzip to tile-compress images. Changed the declaration
of clen in imcomp_compress_tile from int to size_t.
Version 3.25 - 9 June 2010
- fixed bug that was introduced in version 3.13 that broke the ability
to reverse an image section along the y-axis with an image section
specifier like this: myimage.fits[*,-*]. This bug caused the output
image to be filled with zeros.
- fixed typo in the definition of the ftgprh Fortran wrapper routine
in f77_wrap3.c.
- modified the cfitsio.pc.in configuration file to make the lib path
a variable instead of hard coding the path. The provides more
flexibility for projects such as suse and fedora when building CFITSIO.
- fixed bug in imcomp_compress_tile in imcompress.c which caused
null pixel values to be written incorrectly in the rare case where
the floating-point tile of pixels could not be quantized into integers.
- modified imcompress.c to add a new specialized routine to uncompress
an input image and then write it to a output image on a tile by tile basis.
This appears to be faster than the old method of uncompressing the
whole image into memory before writing it out. It also supports
large images with more than 2**31 pixels.
- made trivial changes to 2 statements in drvrfile.c to suppress
nuisance compiler warnings.
- some compilers define CLOCKS_PER_SEC as a double instead of an integer,
so added an explicted integer type conversion to 2 statements in
imcompress.c that used this macro.
- removed debugging printf statements in drvrnet.c (15 July)
Version 3.24 - 26 January 2010
- modified fits_translate_keywords so that it silently ignores any
illegal ASCII characters in the value or comment fields of the input
FITS file. Otherwise, fpack would abort without compressing input
files that contained this minor violation of the FITS rules.
- added support for Super H cpu in fitsio2.h
- updated funpack to correctly handle the -S option, and to use a
more robust algorithm for creating temporary output files.
- modified the imcomp_compress_tile routine to support the NOCOMPRESS
debugging option for real*4 images.
Version 3.23 - 7 January 2010
- reduced the default value for the floating point image quantization
parameter (q) from 16 to 4. This parameter is used when tile compressing
floating point images. This change will increase the average compression
ratio for floating point images from about 4.6 to about 6.5 without losing
any significant information in the image.
- enhanced the template keyword parsing routine to reject a header
template string that only contains a sequence of dashes.
- enhanced the ASCII region file reading routine to allow tabs as well
as spaces between fields in the file.
- got rid of bogus error message when calling fits_update_key_longstr
- Made the error message more explicit when CFITSIO tries to write
to a GZIP compressed file. Instead of just stating "cannot write
to a READONLY file", it will say "cannot write to a GZIP compressed
file".
Version 3.22 - 28 October 2009
- added an option (in imcompress.c) to losslessly compress floating
point images, rather than using the default integer scaling method.
This option is almost never useful in practice for astronomical
images (because the amount of compression is so poor), but it has
been added for test comparison purposes.
- enhanced the dithering option when quantizing and compressing
floating point images so that a random dithering starting point
is used, so that the same dithering pattern does not get used for
every image.
- modified the architecture setup section of fitsio2.h to support the
64-core 8x8-architecture Tile64 platform (thanks to Ken Mighell, NOAO)
Fixes
- fixed a problem that was introduced in version 3.13 of CFITSIO
in cases where a program writes it own END keyword to the header
instead of letting CFITSIO do it, as is strongly recommended. In
one case this caused CFITSIO to rewrite the END keyword and any
blank fill keywords in the header many times, causing a
noticeable slow-down in the FITS file writing speed.
Version 3.21 - 24 September 2009
- fixed bug in cfileio.c that caused CFITSIO to crash with a bus error
on Mac OS X if CFITSIO was compiled with multi-threaded support (with
the --enable-reentrant configure option). The Mac requires an
additional thread initialization step that is not required on Linux
machines. Even with this fix, occasional bus errors have been seen on
some Mac platforms, The bus errors are seen when running the
thread_test.c program. The bus errors are very intermittent, and occur
less than about 1% of the time, on the affected platforms.
These bus errors have not been seen on Linux platforms.
- fixed invalid C comment delimiter ("//*" should have been "/*")
in imcompress.c.
- Increased the CFITSIO version number string length
in fpackutil.c, to fix problem on some platforms when running
fpack -V or funpack -V. Also modified the output format of the
fpack -L command.
Version 3.20 - 31 August 2009
- modified configure.in and configure so that it will build the Fortran
interface routines by default, even if no Fortran compiler is found
in the user's path. Building the interface routines may be disabled
by specifying FC="none". This was done at the request of users who
obtained CFITSIO from some other standard linux distributions, where
CFITSIO was apparently built in an environment that had no Fortran
compiler and hence did not build the Fortran wrappers.
- modified ffchdu (close HDU) so that it calls the routine to update
the maximum length of variable length table columns in the TFORM
values in all cases where the values may have changed. Previously
it would not update the values if a value was already specified in
the TFORM value.
- added 2 new string manipulation functions to the CFITSIO parser
(contributed by Craig Markwardt): strmid extracts a substring
from a string, and strstr searches for a substring within a string.
- removed the code in quantize.c that treated "floating-point integer"
images as a special case (it would just do a datatype conversion from
float to int, and not otherwise quantize the pixel values). This
caused complications with the new subtractive dithering feature.
- enhanced the code for converting floating point images to quantized
scaled integer prior to tile-compressing them, to apply a random
subtractive dithering, which improves the photometric accuracy
of the compressed images.
- added new internal routine, iraf_delete_file, for use by fpack to
delete a pair of IRAF format header and pixel files.
- small change in cfileio.c in the way it recognizes an IRAF format
.imh file. Instead of just requiring that the filename contain the
".imh" string, that string must occur at the end of the file name.
- fixed bug in the code that is used when tile-compressing real*4 FITS
images, which quantizes the floating point pixel values into
integer levels. The bug would only appear in the fairly rare
circumstance of tile compressing a floating point image that contains
null pixels (NaNs) and only when using the lossy Hcompress algorithm
(with the s parameter not equal to 1). This could cause underflow of
low valued pixels, causing them to appear as very large pixel values
(e.g., > 10**30) in the compressed image
- changed the "if defined" blocks in fitsio.h, fitsio2.h and f77_wrap.h
to correctly set the length of long variables on sparc64 machines.
Patch contributed by Matthew Truch (U. Penn).
- modified the HTTP file access code in drvrnet.c to support basic
HTTP authentication, where the user supplies a user name and
password. The CFITSIO filename format in this case is:
"http://username:password@hostname/..."
Thanks to Jochen Liske (ESO) for the suggestion and the code.
Version 3.181 (BETA) - 12 May 2009
- modified region.c and region.h to add support for additional
types of region shapes that are supported by ds9: panda, epanda,
and bpanda.
- fixed compiler error when using the new _REENTRANT flag, having to
do with the an attempted static definition of Fitsio_Lock in
several source files, after declaring it to be non-static in fitsio2.h.
Version 3.18 (BETA) - 10 April 2009
- Made extensive changes to make CFITSIO thread safe. Previously,
all opened FITS files shared a common pool of memory to store
the most recently read or written FITS records in the files.
In a multi-threaded environment different threads could
simultaneously read or write to this common area causing
unpredictable results. This was changed so that every opened
FITS file has its own private memory area for buffering the
file. Most of the changes were in buffers.c, fitsio.h, and
fitsio2.h. Additional changes were made to cfileio.c, mainly
to put locks around small sections of code when setting up the
low-level drivers to read or write the FITS file. Also, locks
were needed around the GZIP compression and uncompression code
in compress.c., the error message stack access routine in
fitscore.c, the encode and decode routines in fits_hcompress.c
and fits_hdecompress.c, in ricecomp.c, and the table row
selection and table calculator functions. Also, removed the
'static' declaration of the local variables in pliocomp.c
which did not appeared to be required and prevented the
routines from being thread safe.
As a consequence of having a separate memory buffer for every
FITS file (by default, about 115 kB per file), CFITSIO may now
allocate more memory than previously when an application
program opens multiple FITS files at once. The read and write
speed may also be slightly faster, since the buffers are not
shared between files.
- Added new families of Fortran wrapper routines to read and
write values to large tables that have more than 2**31 rows.
The arguments that define the first row and first element to
read or write must be I*8 integers, not ordinary I*4
integers. The names of these new routines have 'LL' appended
to them, so for example, ftgcvb becomes ftgcvbll.
Fixes
- Corrected an obscure bug in imcompress.c that would have incorrectly
written the null values only in the rare case of writing a signed
byte array that is then tile compressed using the Hcompress or PLIO
algorithm.
Version 3.14 - 18 March 2009
Enhancements
- modified the tiled-image compression and uncompression code to
support compressing unsigned 16-bit integer images with PLIO.
FITS unsigned integer arrays are offset by -32768, but the PLIO
algorithm does not work with negative integer values. In this
case, an offset of 32768 is added to the array before compression,
and then subtracted again when reading the compressed array.
IMPORTANT NOTE: This change is not backward compatible, so
these PLIO compressed unsigned 16-bit integer images will not be
read correctly by previous versions of CFITSIO; the pixel values
will have an offset of +32768.
- minor changes to the fpack utility to print out more complete
version information with the -V option, and format the report
produced by the -T option more compactly.
Fixes
- Modified imcomp_compress_image (which is called by fpack) so that
it will preserve any null values (NaNs) if the input image has
a floating point datatype (BITPIX = -32 or -64). Null values in
integer datatype images are handled correctly.
- Modified imcomp_copy_comp2img so that it does not copy the
ZBLANK keyword, if present, from the compressed image header
when uncompressing the image.
- Fixed typo in the Fortran wrapper macro for the ftexist function.
Version 3.13 - 5 January 2009
Enhancements
- updated the typedef of LONGLONG in fitsio.h and cfortran.h to
support the Borland compiler which uses the __int64 data type.
- added new feature to the extended filename syntax so that when
performing a filtering operation on specified HDU, if you add
a '#' character after the name or number of the HDU, then ONLY
that HDU (and the primary array if the HDU is a table) will be
copied into the filtered version of the file in memory. Otherwise,
by default CFITSIO copies all the HDUs from the input file into
memory.
- when specifying a section, if the specified number of dimensions
is less than the number of dimensions in the image, then CFITSIO
will use the entire dimension, as if a '*' had been specified.
Thus [1:100] is equivalent to [1:100,*] when specifying a section
of 2 dimensional image.
- modified fits_copy_image_section to read/write the section 1 row
at a time, instead of the whole section, to reduce memory usage.
- added new stream:// drivers for reading/writing to stdin/stdout.
This driver is somewhat fragile, but for simple FITS read and
write operations this driver streams the FITS file on stdin
or stdout without first copying the entire file in memory, as is
done when specifying the file name as "-".
- slight modification to ffcopy to make sure that the END keyword
is correctly written before copying the data. This is required
by the new stream driver.
- modified ffgcprll, so that when writing data to an HDU, it first
checks that the END keyword has been written to the correct place.
This is required by the new stream driver.
Fixes
- fixed bug in ffgcls2 when reading an ASCII string column in binary
tables in cases where the width of the column is greater than 2880
characters and when reading more than 1 row at a time. Similar
change was made to ffpcls to fix same problem with writing to
columns wider than 2880 characters.
- updated the source files listed in makepc.bat so that it can be
used to build CFITSIO with the Borland C++ compiler.
- fixed overflow error in ffiblk that could cause writing to Large Files
(> 2.1 GB) to fail with an error status.
- fixed a bug in the spatial region code (region.c) with the annulus
region. This bug only affected specialized applications which
directly use the internal region structure; it does not affect
any CFITSIO functions directly.
- fixed memory corruption bug in region.c that was triggered if the
region file contained a large number of excluded regions.
- got rid of a harmless error message that would appear if filtering
a FITS table with a GTI file that has zero rows. (eval_f.c)
- modified fits_read_rgnfile so that it removes the error messages
from the error stack if it is unable to open the region file as
a FITS file. (region.c)
Version 3.12 - 8 October 2008
- modified the histogramming code so that the first pixel in the binned
array is chosen as the reference pixel by default, if no other
value is previously defined.
- modified ffitab and ffibin to allow a null pointer to the
EXTNAME string, when inserting a table with no name.
Version 3.11 - 19 September 2008
- optimized the code when tile compressing real*4 images (which get
scaled to integers). This produced a modest speed increase. For
best performance, one must specify the absolute q quantization
parameter, rather than relative to the noise in the tile (which
is expensive to compute).
- modified the FITS region file reading code to check for NaN values,
which signify the end of the array of points in a polygon region.
- removed the test for LONGSIZE == 64 from fitsio.h, since it may
not be defined.
- modified imcompress.c to support unconventional floating point FITS
images that also have BSCALE and BZERO keywords. The compressed
floating point images are linearly scaled twice in this case.
This changes the buildlink3.mk files to use an include guard for the
recursive include. The use of BUILDLINK_DEPTH, BUILDLINK_DEPENDS,
BUILDLINK_PACKAGES and BUILDLINK_ORDER is handled by a single new
variable BUILDLINK_TREE. Each buildlink3.mk file adds a pair of
enter/exit marker, which can be used to reconstruct the tree and
to determine first level includes. Avoiding := for large variables
(BUILDLINK_ORDER) speeds up parse time as += has linear complexity.
The include guard reduces system time by avoiding reading files over and
over again. For complex packages this reduces both %user and %sys time to
half of the former time.
developer is officially maintaining the package.
The rationale for changing this from "tech-pkg" to "pkgsrc-users" is
that it implies that any user can try to maintain the package (by
submitting patches to the mailing list). Since the folks most likely
to care about the package are the folks that want to use it or are
already using it, this would leverage the energy of users who aren't
developers.
in the process. (More information on tech-pkg.)
Bump PKGREVISION and BUILDLINK_DEPENDS of all packages using libtool and
installing .la files.
Bump PKGREVISION (only) of all packages depending directly on the above
via a buildlink3 include.
All library names listed by *.la files no longer need to be listed
in the PLIST, e.g., instead of:
lib/libfoo.a
lib/libfoo.la
lib/libfoo.so
lib/libfoo.so.0
lib/libfoo.so.0.1
one simply needs:
lib/libfoo.la
and bsd.pkg.mk will automatically ensure that the additional library
names are listed in the installed package +CONTENTS file.
Also make LIBTOOLIZE_PLIST default to "yes".
out of date - it was based on a.out OBJECT_FMT, and added entries in the
generated PLISTs to reflect the symlinks that ELF packages uses. It also
tried to be clever, and removed and recreated any symbolic links that were
created, which has resulted in some fun, especially with packages which
use dlopen(3) to load modules. Some recent changes to our ld.so to bring
it more into line with other Operating Systems also exposed some cracks.
+ Modify bsd.pkg.mk and its shared object handling, so that PLISTs now contain
the ELF symlinks.
+ Don't mess about with file system entries when handling shared objects in
bsd.pkg.mk, since it's likely that libtool and the BSD *.mk processing will
have got it right, and have a much better idea than we do.
+ Modify PLISTs to contain "ELF symlinks"
+ On a.out platforms, delete any "ELF symlinks" from the generated PLISTs
+ On ELF platforms, no extra processing needs to be done in bsd.pkg.mk
+ Modify print-PLIST target in bsd.pkg.mk to add dummy symlink entries on
a.out platforms
+ Update the documentation in Packages.txt
With many thanks to Thomas Klausner for keeping me honest with this.
