Re: [cf-pointobsconvention] Draft 2 comments

Interesting. Presumably you keep the unzipped "file" in memory? One issue is that you need a naming convention for the internal files that quickly selects just the ones you want. And all this is outside the netcdf API. And as Brian points out, if you did have a request for some piece of data from each internal file, you'd be hosed. OTOH, the zipped disk file might be much smaller.

Might be interesting to do a bake-off with an "all in one file" design. Do you have a small sample we could look at?

Joe Sirott wrote:
Hi Bryan,
(not that I'm following this closely, would that I had the time ...)

1) I don't want to unzip a million files, and then do a million file open and closes to find the location of a million stations and plot them on a map ... :-( That's my vote for multiple things in one file ...

My point in my previous e-mail (which I guess wasn't very clear) is that you don't have to unzip the zip or jar file to get at the data. You open one file using the zip API and then you have access to all of the netCDF files in the archive. If you're familiar with Java, it works just like a Java jar file -- the jar file contains many individual Java class files which are contained in the single jar file. You don't have to unzip the jar file to get at the individual class files.

Cheers, Joe