[fitswcs] Paper 1 comment

[Mark Calabretta]

On Thu 2002/01/31 14:23:56 CDT, William Pence wrote
and copied to: fitswcs at NRAO.EDU

Hi Bill,

Note that I'm taking your questions in a slightly different order.

>2.  At the end of section 3.3 "Keyword naming convention", a new paragraph
>has been added which describes how the Coordinate Parameters, PVj_ms, could
>be written to a table column instead of storing them as keyword values (and
>hence limited to at most only 9 parameters).

Actually, you can store up to 100 parameters in the header (numbered 0 to
99) provided that the column number (n) does not exceed 99.  E.g. jPVn_m
could be "9PV99_99", or in alternate form "9P99_99A".

The "as few as 9 parameters" comment in the new paragraph refers to column
numbers exceeding 99 as in the definition of "m" which precedes it.

(Also note that while the PV are explicitly limited to 0 - 99, there is
scope for up to 1000 parameters if the axis number does not exceed 9.
However, we felt that this would be too restrictive so introduced the
jPVn_X construct.  It's an important consideration for Paper IV.)

>   How does this work?

Refer to Sect. 3.4 following (the "Greenbank Convention") and Paper II,
Sect. 7.4.1 for specific example headers.

The jPVn_X keyword (the "X" is literal) only works as a vector table
column.  Set TTYPEm = 'jPVn_X', the length of the vector is set by the
corresponding TFORMm and can be fixed or variable.

This issue was discussed in my email to this exploder dated 2001/08/09.

>In the BINTABLE vector case, the table may only contain a few rows, or even
>just a single row, where each row contains an image in a vector column.

Yes (and it could even contain multiple image columns).

>   Is
>it assumed that all the images (rows) in the table use the same non-linear
>projection algorithm, and all have the same values for the coordinate
>parameters.

No, c.f. Sect. 3.4.

>  If that table only has a single row, how can up to 99
>parameters be written to the table?

(I assume you mean 100 parameters numbered 0 to 99.)  In this simple case
you could use any of the following methods:

  1) Record 100 parameters in the header as above.

  2) Record one parameter in each of 100 scalar columns using the
     Greenbank Convention (the image column no. must not exceed 99).

  3) Record 100 parameters in one vector column using jPVn_X, the use of
     which always requires the Greenbank Convention.

>  In the
>Pixel List case, does this mean the table would contain columns named
>'TPV1', 'TPV2', etc. to hold the parameter values (up to 99 of them)?

That should be TPVn_0, TPVn_1,... TPVn_99 where, "n" is the column number.

>  What
>if the pixel list table contains millions of rows?  How does one tell how
>many Parameter Values are in the column, and what value is written to the
>column in all the higher numbered rows.  Note that these new columns could
>significantly increase the size of the pixel list table.

I take your point; pixel lists only define one image so it doesn't make
sense to use the Greenbank Convention - the parameters should always be
in the header, never the table.  Consequently, the TPVn_X and TVn_Xs
should be removed from Table 2.

Note that Sect. 3.4 does correctly refer only to binary tables.

Cheers, Mark