[Edserplo_users] Statistics improved, mouse option available

[Steve Loch]

Those who don't deal with tides needn't read beyond Temporary Files. Note the request for comment.

Button swapping for mice
----------------------------------------
For people who swap their mouse buttons, horizontal panning using mouse clicks happens in a counter-intuitive manner. There is now an option at invocation to correct this. It can also be changed on the Port and Series pages by pressing the 'm' button to get a pop-up. This is Bugzilla enhancement 721. If you think we haven't got it right let us know by reporting through that Bugzilla entry.

Tutorial
------------
This is not working at the moment (and needs to be expanded of course).

Flags
--------
I have added blank to the list of acceptable flags. I have done this for BODC but not for Argo. If I should do it for Argo as well let me know.

Range Selection of Ports and Series (and Channels)
----------------------------------------------------------------------------------
This should work with the shift key now but it is still temperamental with Linux and I'm fairly sure it can be improved but for the record it goes something like this. Suppose you want to select a range of series. Click on the relevant line to define an anchor point. This makes the series (port) the current one highlighted in yellow. Make sure it is selected by clicking again if need be. Move to the other end of the range (above or below) and with the shift key down, click the mouse and the items in the range (but not including the highlighted anchor) shoudl toggle. The highlight remains on the anchor. Use 'C' to  unselect all items beforehand if needed ('c' for channels).
  
Temporary Files
------------------------
On Windows  Edserplo creates a temporary copy of QXF (and some other types of file) to expedite IO. If the filename occurrs more than once in the list of series to be visualised  the necessary steps hadn't been taken to ensure the copy of one file didn't overwrite the copy of the other. This is bug 880 and has been fixed. Even if the series share the same internal name, incidentally, the colour coding for the series will tell you which is which though you will have to go the Series page to find out.

NB Further investigation into the matter also revealed that the use of temporary files is not in line with expectation. In particular it wasn't performing the intended role of shortening the time to update files with flag edits (occurs at Cntl S or shift X) . Please let me know if you think flag update is a slow process.  

Tidal Analysis
---------------------
A review of the Analysis page revealed it wasn't doing things as per spec. Muhammad is currently working on it. Bug#769 which inhibited longer analyses has been resolved. Ditto bug#897 (problem reading dump file prepared without benefit of  Porch files).  What I need to do next - for Sundar - is to allow the use of analyses stored in Oracle as an alternative source to POLAP's database to compute tide and residuals. I also want  to put the Constituent file inside a JAR file which will then make the analysis truly portable. Obfuscation of the code is another high priority.

Statistics
--------------
We can generate History for 3000+ site months without falling over with Too Many Open Cursors and the results largely agree with Edteva's. There are some remaining problems. One disagreement hinges on whether there should be entries for series which are completely flagged 'M'. Edteva says no but doesn't properly delete trailing sequences flagged 'M'. This is the one major difference that needs to be resolved - by consultation and adjustment in one or both programs.

My previous experience with TMOC - in the Matlab context - had led me to think that it was inherent in the (Matlab) code and careful checking of the Java didn't lead me to think otherwise. The main handicap in the debugging of course is the lack of a guide as to where the cursors are coming from. The answer was to devise a probe  which would tell us exactly how many cursors were open at any time. Having found the relevant SQL on the Web to do this I asked Karen to code it as an SQL function and its use  has resolved the problem. For future reference the function can be utilised as follows:-

select sessopcurs from dual;       


The MSL statistics now seem to agree with Edteva's - bar 2. The last problems related to the number of days judged capable of producing a valid statistic for MSL. Most of the time there's no problem but when the good data is bounded by bad data at month ends we were experiencing occasional differences between Edteva and Edserplo which turned on rounding error. I have adjusted Edserplo so it uses a 2-second margin to be more inclusive. The corresponding change has not been made in Edteva (yet). For the record one of the MSL differences is for Perin, IOM 200612.

There are one or two other points to be made of course.

1) Edteva has a problem with absent data. Thus if there is no MSL for a month it doesn't register zero days for that month and lock it as would be theoretically desirable. Edserplo does do this however setting the MSL to null; 

2) The choice of spline doesn't seem to be a big problem as least as far Dataring MSL is concerned because the values recorded are extremely close to the 1/4 hr so one is unlikely to pick up differences here. 

3) There remain messaging display problems on Linux - which are cosmetic to a degree and should be fixed sometime this year. 

Note that there is a facility for producing daily MSL for any channel ('s' option on the OutputPage). You have to select the dependent channel on the PortPage first. In Edteva this was kludged by writing to logical unit number 21 as part of the process of computing monthly MSL statistics (and therefore limited to the channel specified in PRIMOCH). This facility parallels the Hi-Low facility which locates high and low waters and is the reason why the former (the daily MSL) is restricted to a single dependent channel.  With further work we can remove this restriction.