[bananas] First patches for 15APR98
Pat Murphy
pmurphy at nrao.edu
Thu Apr 23 22:14:58 EDT 1998
NOTE: this address is used only for usenet posting. All mail to it is
deleted UNREAD. See the .signature below for contact information.
There are now two patches to 15APR98 AIPS. These are:
Task Description of problem Date
-------------------------------------------------------------------
FITLD makes incorrect source tables and numbers. 1998-04-19
TEKSRV error for SGI and DEC computers. 1998-04-23
Below are details of the FITLD problem as described by Eric Greisen. The
TEKSRV problem is also described (at the very end). The FITLD fix is the
more important of the two, especially if you have used it to process any
VLBA data during the relevant period. Please read carefully.
- Pat Murphy
----- begin included text -----
From: Eric Greisen
Date: Thu, 23 Apr 1998 18:45:47 -0400 (EDT)
FITLD has had problems with source numbers and tables for some time.
The VLBA correlator writes consistent source numbers within each file,
but changes the relation between source name and number from file to
file. Furthermore, it distinguishes between identical sources with
different CalCodes - a case that arises when non-VLBA antennas (with
no CalCodes) are combined with VLBA antennas (with CalCodes). FITLD
has had code to translate the source numbers of incoming data sets to
match those already present in the file. It left the distinction
between CalCodes even though the rest of AIPS has trouble
distinguishing sources with identical names and different CalCodes.
Users complained about the multiple occurrences of their sources within
the data since they are SPLIT as separate sources, treated as
different calibrators at times and confused at other times.
Accordingly, on 27 January, 1998, I changed FITLD to ignore CalCode
when deciding whether an incoming source matched one already in the
data set. Unfortunately, the code in FITLD that renumbered sources
had a backwards sort of logic that allowed only one incoming source
number to be translated into any one output number. In general, one
output number may receive two (or more) input numbers (one with and
one without the CalCode). When this happened, one of the two incoming
source numbers was correctly translated but the other was passed
through unchanged although it was reported as being translated in the
message file (from the wrong number). The output source table also
would have more than one source with the same value in the source
number column.
This is serious and, if you are affected, you should start over. The
bug applies only to 15APR98 FITLD from 27 Jan 1998 until 20 April 1998
(or later if the patch is not applied) and probably only to data
containing both VLBA and non-VLBA antennas. To find out if you are
affected, run LISTR in the OPTYPE='SCAN' mode. A part of the output
is a Source summary like the one below. If there is more than one
occurrence of the same ID number then there is a problem, e.g., number
1 occurs for both 0420-014 and 3C120 as does number 2. The occurrence
of two numbers of 3C120 is a nuisance (that we were trying to fix),
but not a serious error. Note also the confusion about number of
visibilities.
BAD:
Source summary
Velocity type = 'GEOCENTR' Definition = 'OPTICAL '
ID Source Qual Calcode RA(2000.0) Dec(2000.0) IFlux QFlux UFlux VFlux No. vis
1 0420-014 :00000 04:23:15.8007 -01:20:33.064 0.000 0.000 0.000 0.000 7340
IF( 2) 0.000 0.000 0.000 0.000
2 3C120 :00000 04:33:11.0956 05:21:15.620 0.000 0.000 0.000 0.000 8619
IF( 2) 0.000 0.000 0.000 0.000
3 3C120 :00000 N 04:33:11.0956 05:21:15.620 0.000 0.000 0.000 0.000 4613
IF( 2) 0.000 0.000 0.000 0.000
1 3C120 :00000 N 04:33:11.0956 05:21:15.620 0.000 0.000 0.000 0.000 7340
IF( 2) 0.000 0.000 0.000 0.000
2 0420-014 :00000 04:23:15.8007 -01:20:33.064 0.000 0.000 0.000 0.000 8619
IF( 2) 0.000 0.000 0.000 0.000
GOOD:
Source summary
Velocity type = 'GEOCENTR' Definition = 'OPTICAL '
ID Source Qual Calcode RA(2000.0) Dec(2000.0) IFlux QFlux UFlux VFlux No. vis
1 0420-014 :00000 04:23:15.8007 -01:20:33.064 0.000 0.000 0.000 0.000 2325
IF( 2) 0.000 0.000 0.000 0.000
2 3C120 :00000 04:33:11.0956 05:21:15.620 0.000 0.000 0.000 0.000 18247
IF( 2) 0.000 0.000 0.000 0.000
I am sorry about this problem and hope that you are not too seriously
affected by it. The repaired version of FITLD is in 15OCT98 for
midnight-job sites and is available from the patch area on the WWW for
15APR98.
Eric W. Greisen
20 April, 1998
Further bugs which should be patched have been found 23 April 1998
thanks to Andrew Lobanov at MPI. One subroutine used a
single-precision 0.0 where a double was required and a buffer in FITLD
was too small for table I/O. Both of these caused the task to die on
HPs and the latter may have caused troubles with CQ tables on other
computers.
Also on 23 April, 1998, a patch was released for SGI and DEC
computers. The $APLSGI and $APLDEC versions of ZTKSRV.C were not
changed when the image catalog was moved into the Tektronix server
($AIPNOT/TEKSRV). The $APLUNIX version was changed and works well,
but the other two versions were not changed until today. Managers of
DEC and SGI machines will need to patch their 15APR98 versions for
this problem.
Eric W. Greisen
23 April, 1998
----- end -----
--
\|/ ____ \|/ Pat Murphy; email http://orangutan.cv.nrao.edu/kippure.html
@~ / oO \ ~@ Division Head, NRAO/Charlottesville Computing
/_( \__/ )_\ Plan, PGP, Pages: http://orangutan.cv.nrao.edu/plan.html
\_U__/ "I don't believe in the no-win scenario" -- James T. Kirk
More information about the Bananas
mailing list