[daip] Test of VBMRG

[Jon Romney]

OK!  Preliminary tests (all I can do before the VLBA Test
meeting) indicate this recipe works *perfectly*:

--  Use FITLD to load data from multiple station-subset
correlation passes, with concatenation ON.  Output datasets
can be either compressed or uncompressed.

--  Use UVSRT to sort into BT order.

--  Use VBMRG to select unique records.

--  Use UVSRT to re-sort into TB order (for most purposes).

--  Run INDXR to re-index the final dataset.  (I also did
this after the first sort but it's not clear to me whether
that was necessary.)

That's the end of the recipe.  I used UVPRT to list bits
of the pre- and post-merge datasets, and found the selec-
tion of unique records appeared to be done perfectly.

Many thanks for the quick action on VBMRG to fix the short
buffer and add the handling of compressed data.

Jon


Eric Greisen wrote:
> Jon Romney writes:
>  > 
>  > --  Load all three passes sequentially using FITLD, with
>  > concatenation ON and compression OFF.  This seemed to
>  > work.  Output dataset: STNSUBTEST.UVDATA.1.
>  > 
>  > --  Sort into BT order using MSORT.  It notes a "Truely
>  > brute force disk copy to be done", but also "Appears to
>  > have ended successfully".  Output dataset:
>  > STNSUBTSRT.UVDATA.1.
> 
>       UVSRT is probably better than MSORT for such sorts.
> 
>  > 
>  > --  Run VBMRG on the sorted dataset.  This failed, with
>  > error message: "RECORD LENGTH=  6154 GREATER THAN  2060 /
>  > VMRGUV: VMERGE ERROR  8".
> 
>        Corrected the issue of a small internal buffer.
> 
>  > 
>  > The datasets mentioned above are on 'thuban' disk 2,
>  > user #1891.
> 
> I have also made it work on compressed data.  But you should note that
> the results on a compressed data set are not the same as on an
> uncompressed data set.  I have added to the help file an attempt to
> explain why this is so.
> 
> The task should be available Tuesday via the MNJ.
> 
> Eric Greisen
>