On Sun, Dec 21, 2014 at 1:50 AM, Evan Laforge wrote:

On Sat, Dec 20, 2014 at 3:29 AM, Corentin Dupont wrote:

...

Hi, I didn't had so much luck with this question. The question was: -> "How to avoid keeping old data formats in your code?" Indeed those old data formats are often needed for retro-compatibility with previous versions of your software.

Say you have a data structure that is serialized in a file, and then you add a field in a later version. You are obliged to keep both versions of the data structure if you want to be able to read both versions of the file. Potentially, if you update very often your structure between releases, you are obliged to keep N versions of the data structure in your code. See the exemple at: http://acid-state.seize.it/safecopy

I just keep both versions. It doesn't seem like too much hassle. I move the old version to the Serialize module so modules that aren't concerned with serialization aren't cluttered.

In the fully general case, I think there's no way around it. But if you are just adding a field, or or changing a field to a super-type (e.g. 'a' becomes 'Maybe a') then you can trivially upgrade in place in the deserialization code. Or you can use something like protobufs, which supports that kind of thing automatically.

What I'm thinking about is to put in the Setup.hs some code that will extract from GIT the older versions of the data structure. This way you are able to keep only the last version of it in the HEAD, and still be able to build migration code. Do you know of any such custom build?

Yeah, I agree with Evan, it does sound too complicated. I also have a SafeCopy module where I keep my SafeCopy instances and old versions of data types. On 21/12/14 07:29, Evan Laforge wrote:

I haven't heard of anything like that, it sounds too complicated for me. You still need to convert the old format to the new one, so it seems like you still need a bit of code for each version, which needs access to both old and new.

On Dec 21, 2014 9:10 AM, "Corentin Dupont" mailto:corentin.dupont@gmail.com> wrote:

On Sun, Dec 21, 2014 at 1:50 AM, Evan Laforge mailto:qdunkan@gmail.com> wrote:

On Sat, Dec 20, 2014 at 3:29 AM, Corentin Dupont mailto:corentin.dupont@gmail.com> wrote: > Hi, > I didn't had so much luck with this question. The question was: > -> "How to avoid keeping old data formats in your code?" > Indeed those old data formats are often needed for retro-compatibility with > previous versions of your software. > > Say you have a data structure that is serialized in a file, and then you add > a field in a later version. You are obliged to keep both versions of the > data structure if you want to be able to read both versions of the file. > Potentially, if you update very often your structure between releases, you > are obliged to keep N versions of the data structure in your code. > See the exemple at: http://acid-state.seize.it/safecopy

I just keep both versions. It doesn't seem like too much hassle. I move the old version to the Serialize module so modules that aren't concerned with serialization aren't cluttered.

In the fully general case, I think there's no way around it. But if you are just adding a field, or or changing a field to a super-type (e.g. 'a' becomes 'Maybe a') then you can trivially upgrade in place in the deserialization code. Or you can use something like protobufs, which supports that kind of thing automatically.

What I'm thinking about is to put in the Setup.hs some code that will extract from GIT the older versions of the data structure. This way you are able to keep only the last version of it in the HEAD, and still be able to build migration code. Do you know of any such custom build?

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe