Savefile compatibility

[Carnivean]

Originally Posted by Derakon

Someone has to write the code to do the conversion. If you have that code, you might as well just use it to automatically upgrade savefiles when run under the new version -- you don't gain anything by making it a separate process. The entire point of breaking savefile compatibility is to not have to write that code in the first place.

Ok, if we assume that to be true (and I have no reason to doubt it), then how hard is that process? Assume that only version X-1 is required.

[AnonymousHero]

Originally Posted by Carnivean

I get that.

However, you have 2 defined file types, with largely overlapping data. Outside of the core game, it seems like it should be possible to fill as much of the new file as exists in the old file, and fill the rest with an acceptable default value. This would leave you with a save file that loads in the new version, without having to change the core game code to facilitate it.

Is there a technical challenge to doing this that I'm not aware of?

Yes. It's -- as they usually say in techincal litterature -- a highly non-trivial undertaking. In this case for mostly practical reasons having to do with numbers and positions of those numbers changing all over the place in unpredictable ways.

You can certainly mitigate some of this by choosing a standardized and self-describing file format for savefiles (e.g. JSON or XML), but that's only 1% of the solution.

Originally Posted by Carnivean

Ok, if we assume that to be true (and I have no reason to doubt it), then how hard is that process? Assume that only version X-1 is required.

Incredibly hard, unfortunately -- even for a single version. You'd be suprised if you tried it.

(At bottom it all has to do with the non-linearity of changes to programs and the effects of those programs, but I won't bother you with the abstract details.)

EDIT: I should say, that avoiding breaking compatibility is a lot easier (if you have good testing in place and/or are extremely focused on it), but that constrains the things you can do enormously. It's always a trade-off.

[Nick]

Originally Posted by tumbleweed

As far as I can tell from glancing over the savefile howto, Angband should already have the facilities in place to support gracefully rejecting whole savefiles or even just the individual parts it can't load any more.

Which makes me wonder why current iterations of 4.0beta crash and burn horribly when attempting to load old 4.0beta saves from before the break.

You'll notice that the method in the howto relies on there being a new function to read items every time items are changed. My plan in the restructure process was to have no savefile compatibility with previous versions. So I blew away all the old item readers, and started with just one. Thing is, this one has changed during the rewriting process, so it is no longer compatible with savefiles form early in the rewrite, and there is no nice system of tiered item readers to allow it to fail gracefully.

So this behaviour is not unexpected; it shouldn't ever happen for savefiles from actual proper versions, though.

[Nick]

Originally Posted by AnonymousHero

However, there is one case where it might conceivably be desirable, namely Linux distributions where you might get updated from 8.1.3 to 8.2.0 semi-automatically. However, people using those usually know what they're doing -- as long as you get a reasonably clear error message from the game (and it doesn't corrupt the save file) people could just downgrade.

Thanks - this is a good point. I will make sure to make the error message helpful.

It's also good for me to see that sometimes it's Linux that makes my life more difficult rather that one of the other OSes

[Carnivean]

Originally Posted by AnonymousHero

Yes. It's -- as they usually say in techincal litterature -- a highly non-trivial undertaking. In this case for mostly practical reasons having to do with numbers and positions of those numbers changing all over the place in unpredictable ways.

Thanks. Without looking into the code, and I'm hopeless at reading C, I wouldn't have known that.

[tumbleweed]

Originally Posted by Nick

Thing is, this one has changed during the rewriting process, so it is no longer compatible with savefiles form early in the rewrite, and there is no nice system of tiered item readers to allow it to fail gracefully.

Yeah, fair enough.

Then again, I'd still call it "failing gracefully" if it at least told me to kindly fuck off with pre-230 (iirc) savefiles. But instead it just falls flat on its face with assertion failures and access violations.

[Derakon]

Originally Posted by tumbleweed

Yeah, fair enough.

Then again, I'd still call it "failing gracefully" if it at least told me to kindly fuck off with pre-230 (iirc) savefiles. But instead it just falls flat on its face with assertion failures and access violations.

Sounds like all that's really needed is to stick a version number in the header and error out if that version doesn't match the version of Angband currently running.

[Nick]

Originally Posted by tumbleweed

Then again, I'd still call it "failing gracefully" if it at least told me to kindly fuck off with pre-230 (iirc) savefiles. But instead it just falls flat on its face with assertion failures and access violations.

Yes, I think assertion failure just from loading something invalid is a bit rude. I'll fix that.

[AnonymousHero]

Originally Posted by Nick

It's also good for me to see that sometimes it's Linux that makes my life more difficult rather that one of the other OSes

Hey! I'm sure I should feel insulted, but somehow I don't...?!? Dude?!?

[AnonymousHero]

Originally Posted by Carnivean

Thanks. Without looking into the code, and I'm hopeless at reading C, I wouldn't have known that.

I don't mean to (metaphorically) apply a trounce to anyone, so please do ask if you have any questions about coding (or otherwise). EDIT: ... and I also want to sell you the true awesome of programming in case you're not acquainted already.