I suppose for simplicity that records will for the most part will be within directories for the vendor and the midlet name. I suppose a vendor directory first followed by the name. All in lowercase naturally. However, some file systems do not support spaces in filenames such as DOS and have character limitations. So the file store will need to have a kind of manifest to specify the actual vendor and the name. Then I suppose the directory is just the hash code of the vendor/name. I suppose since there is a synchronization lock on open, I do not need to worry about multiple clusters being opened at the same time. I do however will need to handle cases where the filesystem is read-only and I cannot create new clusters. I should also potentially lazily create the clusters, but that could complicate things a bit. However, what I can do is store each record in its own individual file which means I do not have to handle my own database format. This would delay the need to have a block based hanlder for records, but a later implementation of that could be shared for example with discreet blocks of memory. It is also possible for me to model a filesystem on top of the record system in the event that none exists. It would basically be a record filesystem.


A record filesystem could be handy for some systems which lack a true filesystem, although it might not be capable of storing much data and it would only be accessible from SquirrelJME for the most part. This means that it would be best to write a utility that can be used to manage files, along with zipping and unzipping them. That would be a very useful utility to have.


I believe I need an actual TODO class, one which prints the stack trace then exits the SquirrelJME fatally. Otherwise when I am implementing support for some things such as third party applications or games, the exceptions that I need to figure what to implement are gone.


This will be the only means to know what the actually implement.


Ok for some reason, I read the data but the wanted CRC is incorrect.


So the length is 13 but the read count is actually 9.


Actually, I was reading the input data, I need to increase the length of 4 because the type is included also.


An enumeration would be the best course for the color type perhaps, but that would be somewhat painful to have.


Reading the image data will be fun perhaps. I suppose the easiest thing to do would be to have a large number of input streams which just read pixel data for the most part. That would be the simplest. Then that way I do not need giant switch statements and such.