Hi the list,

I have some question relative to data migration.

Say you have a software in version A which save data in a file with format FA. Later, you update your software to version B, with data format FB.

Now, if you want your version B of the software to be able to read data saved by version A, you're obliged to include FA in version B, together with some functions to translate from FA to FB.

This is what I don't find elegant: you're obliged to keep old code (data formats) in your software.

-> do you know of any way to avoid keeping old data formats (for retro-compatiblilty) in your code?

In practice I use acid-state with safecopy:
http://acid-state.seize.it/safecopy

In this example you can see the problem: the author is obliged to keep old code (data structures) to maintain compatibility. Even worth, you might be obliged to suffix your data structure with a version number:

data MyType_V1 = MyType_V1 Int
data MyType_V2 = MyType_V2 Integer

Instead, I'm thinking of a process using GIT, or Cabal as a back-end.

The idea would be to have an additional program (or library) specialized in the data migration of your main software. It would extract both version A and B from the repo, and then would compile an application capable of handling migrations from FA to FB.

Does something like this exists (even outside of Haskell)?

Cheers,
Corentin