gijzelaerr
commented
10 years ago Original comment thread migrated from bugzilla
at 2006-07-26 21:51:54 Tony Willis replied:
Hi Ger and Oleg
A MeqTrees <-> aips++ Tables interface problem seems to have recently surfaced. In order to speed up the CLAR simulation system which requires a lot of AzEl calculations, we first run a 'pre-processor' program which stores the fitted AzEl parameters as a MeqParm in what's called a 'mep' table but which is just an aips++ Table underneath. Then in succeeding steps, we read these Parameters back in directly from the Table rather than having to recalculate, etc. This system seems to work fine when the MeqTrees system is linked against aips++ 1.9 Build 1353.
Recently, however I tried upgrading to aips++ stable 1.9 Build 1460. But something is now incorrect, because when I now link the identical MeqTrees code against this version of aips++, a parameter which is based on contents of the 'mep' table, and should have a value of roughly 1e-7, is given an incorrect value of 1. Oleg, have a look at MeqComposer E:S10:1 3:ediag:S10:1 exp(multiply(V_GAIN, ihpbw_sq)):S10:1 multiply((V_GAIN, ihpbw_sq):S10:1 V_GAIN:S10:1 ihpbw_sq
The problem lies with V_GAIN. It should be roughly 1e-7, but with the bad table interface it is somehow ending up with a scalar value 1. Since ihpbw_sq is about 4e5 in value, you can imagine that exp(4e5) causes a serious sytem implosion!
For the moment I shall retreat to aips++ V 1353!
at 2006-07-29 00:36:47 Tony Willis replied:
Hi Oleg
I don't think the 1460 version of aips++ has been installed on your site. When you want to track this problem down. I can relink things to the 1460 aips++ on the atic1 computer here so you can find the problem.
at 2006-11-14 19:07:36 Tony Willis replied:
Well, things seem to be working fine with aips++ stable 19.1556 so who knows - maybe I had some obscure link issue or something. With 1556 I can read in paramters from my beam_locations.mep file used for focal plane arrays with no problems. Anyway, lets forget about this one - not worth tracking down.
at 2007-06-01 18:20:34 Oleg Smirnov replied:
Reassigned to MeqKernel component, since I want to eliminate unnecessary components
at 2006-07-26 21:51:54 Tony Willis reported:
MeqTrees <-> aips++ Table problem with aips++ V1.9 1460 'stable'