Rotate only if there will be a new <interval>.0 folder

[myrdd]

With this bugfix, gaps like a missing .0 folder will be prevented.
See bug #18.

[nkadel]

This is a reasonable but notable default behavior change. In some environments, you want to discard all backups past a certain age, even at the cost of braking hardlinking. But in general, I think it's a really good idea.

[bebehei]

@myrdd Thanks for the patch. Now after @nkadel's answer, I understood the reason for the patch. But: In the patch is checked for the highest interval and not for .0!?

[myrdd]

@bebehei see my comment on the commit.

I understand that there can be other use cases where rotating is wanted. Maybe this PR should be an opt-in, that is, we add a new option to rsnapshot? What do you think @bebehei?

even at the cost of braking hardlinking

@nkadel If I understand correctly, hardlinking will only be broken in case lowest_interval.0 disappears, no? By the way, removing lowest_interval.0 is prevented here (current master branch at line 4765).

[sgpinkus]

👍 . Even though its a simple thing to check that the last snapshot in the previous interval is there or not before rotating, it makes sense for rsnapshot to do this because it already has access to the number of snapshots in the previous interval.

In some environments, you want to discard all backups past a certain age, even at the cost of braking hardlinking.

@nkadel I'm interested in what the concrete use cases for not this commit are??

Regarding the confusion about $prev_interval_max. I think you need to add a test case for this change to verify its correct.

[bebehei]

@sam-at-github test case added in #92

[myrdd]

This PR has been replaced by #92, which is already merged. Closing.

[WKarel]

@myrdd has closed this PR, because seemingly, #92 had been merged. But this is not the case - see #92 (comment)

The code changes of this PR and #92 are basically identical. The original problem still persists.

Re-open this one?

[myrdd]

@WKarel yes, the code changes to rsnapshot-program.pl in #92 are identical, but #92 also adds a test. I suggest you to open another PR which basically copies #92. This way, you are the owner of the PR.