Dinner prospect, and another philosphical tirade

From: Mario Roederer (Roederer@Beadle.Stanford.EDU)
Date: Mon Jun 30 1997 - 17:14:27 EST


Ray asks:

>   On the FCS side, are there any programs that do cover
>   all of the permutations of any version? Are there any
>   that actually conform yet?

The answer to the first part is definitely no.  The FCS standard is
unfortunately much too flexible--thus writing code to read all possible variants
is nearly impossible, not to mention simply not worth the effort!  For
example-who now stores bivariate histograms in data files?  Anyone?  And yet,
this is part of the FCS standard.  And I will buy dinner, with wine, for the
first person who shows me a legitimate, non-hypothetical use for the $LOST
keyword.

In particular, the problem becomes considerably more difficult when you consider
fully bit-packed data (i.e., putting 10 bit data in only 10 bits instead of 16).
For those who thought byte order was a problem for FCS, this is a nightmare by
comparison.  Even for those who didn't think byte order was difficult, this is
still a nightmare.

It also seems that the evolution of FCS is only going towards greater
flexibility rather than less.  This is like the evolution of CPU's--for a while,
manufacturers were putting greater and greater power into each--supporting
thousands of different instructions.  Then someone discovered that supporting
LESS instructions made for MORE efficient processers (i.e., RISC processors).  I
hope that the FCS standard will soon evolve in a similar direction--simplify!
Especially, in a way that encourages rigorous data analysis (e.g., not allowing
the storage of non-listmode data).

 Why not make the new FCS actually only allow those formats which are currently
supported by a vast majority of software?  The distilled version would be far
easier to support programmatically, and would result in far fewer problems for
cross-compatibility.

mr



This archive was generated by hypermail 2b29 : Wed Apr 03 2002 - 11:49:53 EST