|
|
|
| |

|
| |
This page provides a forum for feedback on
bugs in the HEALPix package. We will publish contributions here if they
are of general interest.
Please submit to
(no attachment larger than 4 MB).
NOTE: If you encounter internal compiler errors when compiling the HEALPix
source code, we strongly encourage you to report these problems to your
compiler manufacturer, or to let your system administrators do this for you,
even if workarounds for these compiler bugs exist. This is very important
to improve the quality of compilers, and thus to make life easier for programmers.
NOTE: Version 4.1.0 of the gfortran compiler has a bug which causes silent
miscompilation of the current HEALPix Fortran packages. This typically
leads to segmentation faults when the HEALPix facilities are run. No workaround
is known at this time.
The problem has been reported to the gfortran developers and will
hopefully be fixed in version 4.1.1.
gfortran version 4.0.3 does not exhibit this problem.
Detected 03-16-2006 by M. Reinecke
Update: The bug has indeed been fixed in gfortran version 4.1.1.
- In the IDL routine
write_fits_cut4 the default extension was not properly defined. To fix this problem,
replace src/idl/fits/write_fits_cut4.pro with the one found here.
2006-11-09, reported by Jiang, B. Z.
- Version 4.1.1 of the
gfortran compiler has a bug
preventing the correct compilation and usage of the Healpix routine
fatal_error. To fix this problem,
replace src/f90/mod/misc_utils.f90 with the one found here.
2006-10-31, reported and fixed by V. Stolyarov
- The IDL routine
median_filter
had several bugs affecting its FITS file output and did not deal properly with
filtering radii provided in degrees. A new problem related to the /NESTED
keyword also has been detected. To fix all these problems, replace src/idl/toolkit/median_filter.pro with the one found here.
2006-03-09, reported and fixed by M.A. Miville-Deschenes
Updated 2006-07-11, reported by Kwangil Seon
- The IDL visualisation routines
cartview, gnomview, mollview and orthview
did not deal properly with polarised data read from multi-row arrays (Online option) and with cut-sky polarised data.
To fix both these problems, replace
src/idl/visu/loaddata_healpix.pro with the one found here.
Updated 2006-04-27, reported by T. Banday & M. Landriau
- One of Sun's Fortran compilers has a bug leading to an internal compiler
error when compiling
fitstools.f90. If you encounter this,
you should change line 863 of
src/f90/mod/fits_template.f90 from
tform(1:ncl) = trim(adjustl(form))
to
tform(1:ncl) = form
Updated 03-23-2006, reported by Tristan Smith
- The F90 function
median
has been edited to compile properly with Portland Group Compiler (pgf90 5.2-4) and to fix a memory leak. To enjoy this upgrade, replace src/f90/mod/statistics.f90 with this one.
Updated 03-09-2006, reported by M. Alvarez
- The F90 module
alm_tools
has been slightly edited to facilitate the compilation of the MPI based Spherical Harmonics routines (ie, making the select_rings subroutine public). Replace src/f90/mod/alm_tools.f90 with this one.
Updated 03-09-2006, reported and fixed by D. Larson
- The F90 routine
write_alms (called by alms2fits)
has a bug and will not let you write polarised alms (as is the case when the alteralm facility is used on polarised alms with error bars). To fix this problem, replace src/f90/mod/fits_template.f90 with the one found here.
Updated 01-23-2006, reported by D. Maino
- To compile the Fortran package under Windows, you should proceed as follows:
Updated 12-08-2005
- on Nov 05, 2005, version
2.01 was released to address the
bugs and problems found
in version 2.00 (click here
for
details)
- If you run Healpix on Mac OS X with the g95
compiler, you will have to
-- install the latest cctools available from g95.sf.net
-- add -lSystemStubs to the LDFLAGS in the
Makefile created by configure
Updated 9-27-2005, reported by Andrew Jaffe
- If you run Healpix on Mac OS X with the IBM XL
Fortran compiler, you will have to
-- change the cfitsio installation (click here for
details),
-- add -lSystemStubs to the LDFLAGS in the
Makefile created by configure
Updated 9-27-2005, reported by Hiranya Peiris
- If you compile Healpix with the GNU Fortran 95 (gfortran)
compiler, you will have to replace
Healpix_2.00/src/f90/mod/healpix_fft.F90
with the one found here.
Updated 9-27-2005 --
FIXED in v2.01 --
- The C++ method
Healpix_Base::neighbors() has
a bug and will return incorrect neighbor lists in most cases. This bug
can be fixed by replacing the following lines in src/cxx/Healpix_cxx/healpix_base.cc:
line 585:
result[m] =
xyf2ring(ix+xoffset[m],iy+xoffset[m],face_num);
should be changed to
result[m] =
xyf2ring(ix+xoffset[m],iy+yoffset[m],face_num);
line 588:
result[m] =
xyf2nest(ix+xoffset[m],iy+xoffset[m],face_num);
should be changed to
result[m] =
xyf2nest(ix+xoffset[m],iy+yoffset[m],face_num);
Updated 11-03-2005, reported by Marshall
Roth --
FIXED in v2.01 --

- The g95 compiler is a Fortran90/95 compiler
available for free on Linux, Mac
OSX, Windows, Sun and HP platforms. It is available for download (as a
self-contained precompiled executable) at http://g95.sf.net.
Using it with Healpix requires some minimal editions to the configure
script and to the src/f90/mod/healpix_fft*.F90 source
files. A patch available here
will modify those 3 files.
Notes:
-- we have only tested this compiler on Linux, and could not assess its
overall performances yet,
-- the editions performed in the patch above remain perfectly
compatible with all the other compilers.
Posted March 3, 2005
- Some versions of the Intel F90 compiler (in
particular version 8.1.021) may choke during the compilation of the
healpix_types
module, with an error message similar to
fortcom: Severe: healpix_types.f90, line 24: Please
report this error along with the circumstances in which it occurred in
a Software Problem Report
INTEGER, PARAMETER, public :: max_i1b = HUGE(1_i1b)
^
[ Aborting due to internal error. ]
compilation aborted for healpix_types.f90 (code 1)
If this happens, edit the file src/f90/mod/healpix_types.f90 as
follows:
replace (in line #24)
INTEGER, PARAMETER, public ::
max_i1b = HUGE(1_i1b)
with
INTEGER, PARAMETER, public ::
max_i1b = 127
(Problem reported by Anna Bonaldi, Monica Relaño Pastor and
Licia Verde.)
Item added Feb 2, 2005
- If you run Healpix on Mac OS X with the IBM XL Fortran
compiler, you will have to change the cfitsio installation. Click here for details.
- The patch #02 has been send to all
registered users on March 26, 2004. It fixes the bugs reported since
March 2003, and includes some code enhancements. Installation on Mac OS
is now supported. Click here for
details.
- A problem occuring (under very limited circumstances) with
the gaussian deviate random generator
randgauss_boxmuller
has been pointed out. Click
here for details and for a corrected source file.
- Note about overwritting existing FITS files in C or
Fortran 90: CFITSIO (the library used for FITS I/O) will crash if you
attempt to write a FITS file that already exists, unless you
prepend it with
'!' (exclamation mark) (in version
2.0 or newer) (e.g., outfile='!out.fits' will write 'out.fits',
whether or not it already exists).
Note that the HEALPix facilities synfast, anafast and smoothing test
for the existence of the file before accepting an output file name as
valid. A file whose name starts with '!' is considered as
new and will be passed on to CFITSIO.
(For the experts: beware of the effect of '!' in shell
scripts. You may have to protect it with '\').
- The patch #01 has been send to all registered users on
March 1st, 2003. It fixes all the bugs reported to us so far, and
includes some code enhancements. Click here for
details.
- We recommend that the GNU
'make' be used for
compilation of HEALPix (ftp://ftp.gnu.org/gnu/make/). If you
are using a different 'make' (specially under SunOS) you
can get the message
make: Fatal error in reader:
Makefile, line 51: Unexpected
end of line seen during compilation.
If the compilation fails for this reason and the GNU make is not
available, replace the files Makefile.in, src/f90/mod/Makefile
and src/f90/lib/Makefile with those found in this tar
file. You can then run the commands './configure' and 'make'
to install the package.
(Problem reported by Matias Zaldarriaga)
- Some Fortran 90 compilers do not support 8 byte (or 64
bit) integers and either fail during the compilation of
healpix_types
or choke over the usage of the i8b parameter. If this
happens, edit the file src/f90/mod/healpix_types.f90 as
follows:
replace line 11, which currently reads
INTEGER, PARAMETER, public ::
i8b = SELECTED_INT_KIND(18)
with
INTEGER, PARAMETER, public ::
i8b = SELECTED_INT_KIND(9)
(Problem reported by Viktor Eisler)

- Because of non-standard behaviour of native awk, the
HEALPix configure script produces an error message on at least some
versions of SunOS. If that happens to you, apply the following patch to
configure.
Replace:
test `f90 -V 2>&1 | \
awk '{if (/Compilers/) {print $0}}' |\
with:
test `f90 -V 2>&1 |
head -1 |\
(Problem reported by Adam Amara)

- The HEALPix Fortran90 Subroutines documentation on page 37
should read "double precision" for the kind of theta and phi, not
integer. Also,
create_alm requires the header_PS
argument to be declared as
character(len=80),
dimension(1:60) :: header_PS
and the filename and windowfile arguments
as
character(len=128) ::
filename,windowfile
(E. Pierpaoli)
- The HEALPix Fortran90 Subroutines documentation contains
an inaccuracy and an error relating to
the use of
cos_theta_cut in the section on map2alm.
(G. Hinshaw)
- The IDL routine
index2lm.pro in the src/idl/misc
directory gives a wrong m vector for large l. Download
the corrected version.
(B.D. Wandelt)
- On some Sun machines there appears to be a problem with
the subroutine which reads in FITS files when running the
hotspot
facility. A workaround is now
available.
(A. Jaffe)
- IDL routines need recent version of IDL (e.g., IDL 5.0.3
chokes on "
today()").
(K. Ganga)
- We found that DEC compilers have a non-standard definition
of the default format "
*". This makes this
change necessary.
(F.K. Hansen)
- Line 239 in
Healpix/src/f90/plmgen/plmgen.f90
should be removed.
(F.K. Hansen)
- Compiler bug workaround for DEC.
(A. Connolly)
- (Now solved (Version 1.0)) The standard implementation of
P_lm
recursion, may cause floating point exceptions on some platforms at l
>> 2000; a new P_lm evaluation, applicable
at high l-values is included with the new version of the
package.
(R. Stompor)
- Owing to the limitations of 4 byte integers, the number of
pixels in a map currently cannot exceed 805,306,368. This limits the
smallest resolvable scale to 0.429 arcminutes.
(B.D. Wandelt)
|
| |
|
|
|