query by label? With any database you need to use an index. Berkeley DB is no different. It is much lower level but has the same facilities which are required to implement all of these features in any RDBMS. Including partial row retrieval and storage – when you know the size of the row columns (which you cannot (easily) if you have variable length columns).

Given that querying by some indexing function (e.g. query by label) was the whole point of the disksummary branch to start with, it is definitely possible although it isn’t always as easy to write.

Though I did not measure the time, I felt, the column unification and splitting with berkeley db will take a lot of time. And I cannot query for individual fields. I have to get everything. For instance, if an application say, OpenOffice wants to query a certain labelled mail’s subjects alone, it may not be possible.

But as you said, your work is for fun / challenge and interest . So, there is no question of a berkeley db vs sqlite :-)

sankar: But you still need sql to object marshalling/demarshalling, and you have to mix languages in the code, so what’s the difference? The code is still there, but as well as the db code which does it you have to do it again yourself. I think sql-language queries is a valid reason to use it rather than something like libdb, but they also have significant execution and memory space overhead. Anyway it’s an exercise in trying to do it without those crutches as much as anything – I find this sort of code more challenging and thus interesting, and since I’m doing it for fun, I want it to be fun. SQL is not fun.

You might want to give a try to sqlite instead of libdb. It supports storing multiple columns and hence you don’t need all that columns-to-a-combined-value overhead code.