restore deleted files too

[alive]

context:
mkdir foo/, mkdir foo/bar/, touch foo/bar/file
gb backup foo/
rm -r foo/bar/
gb backup foo/

now:
gb restore foo/bar/ will fail without timestamp provided, even though it's pretty clear (?) what gb should restore - namely the last version of bar/

This is just a proof of concept of what a solution might look like.

[leijurv]

well wait, i thought the whole idea of your issue was that it's annoying to restore files one at a time with a timestamp for each. wouldn't this make much more sense if it was per-file rather than per-directory?

also i don't like the GLOB thing. https://github.com/leijurv/gb/blob/master/proxy/proxy.go#L101 in order to glob directory paths containing square brackets you have to do a stupid hack

[leijurv]

I think it can be much simpler tbh, very few lines need to change. I think this is an alternative for QUERY_BASE

const (
	QUERY_BASE = "SELECT files.hash, files.path, files.fs_modified, files.permissions, MAX(files.start) AS start, sizes.size FROM files INNER JOIN sizes ON files.hash = sizes.hash WHERE files.path GLOB ? GROUP BY files.path"
)

(it isn't quite right because of how I put the files.path GLOB ? in the query rather than appending it to the end, but this is just to get across the idea)

the trick here is that if you put in max(files.start) and group by files.path sqlite will do something hilarious which is to pick the row with the highest value for files.start for every files.path. the other columns like files.hash will be chosen to be the correct values for that row where files.start was maximized. how convenient, that's exactly what we want! (this is a sqlite only thing, see https://sqlite.org/lang_select.html#simple_select_processing section 2.5 Bare columns in an aggregate query if you don't believe me)

[5HT2]

this is a sqlite only thing

tfw postgresql user