Why does not RSF store a complete file history in the header (as SEPlib does)?
Not keeping complete file histories in header files was a design decision. The reasoning is that processing histories belong in other places (namely, the SConstruct files), and one should not rely on the header information for figuring out how the file was processed. A technical detail is that, since history parameters are stored in a hash table, they are outputed in random order. Please let us know if you can think of compelling reasons for changing this behavior.
Update: the behavior has changed following the suggestion from James Rickett. File histories are now stored in RSF files.
Sergey,
I think this blog is great btw.
However, the lack of history in file headers is one thing though that I dont particularly like: I want to use the RSF mains, but not scons. I dont have SConstruct files in which to keep processing histories. Datasets tend to get passed around to different people too. Keeping processing histories attached to the data means this doesnt get lost.
My understanding is that the hash table is just an implementation choice here, and is not driving the design.