Magic Folder: consider types of (size, ctime, mtime) more carefully

[daira]

These need to be compared, so it's important that the magic folder db be able to represent them precisely enough.

[daira]

​https://www.sqlite.org/datatype3.html

[daira]

The issue here is, for all possible floats f that can be returned by time.time(), does storing f in a database field and then retrieving it result in a float == to f? Since we compare timestamps using ==, if this were not the case, then we'd detect spurious timestamp mismatches.

[daira]

I don't know where the type NUMBER came from in the current schema. We probably want REAL.

[dawuud]

ok here's my solution: ​https://github.com/david415/tahoe-lafs/tree/2505.db-schema-types.0

i've replaced NUMBER and TIMESTAMP with REAL. According to this link, TIMESTAMP is useful if you sometimes want to have a default value of the current time; we do not.

​https://androidcookbook.com/Recipe.seam?recipeId=413

All tests pass. Review needed.

[meejah]

dawuud's changes look good to me; certainly an improvement over what's there now.

I do share the == concern about floats -- *especially* because Python's internal representation isn't necessarily an 8-byte IEEE as SQLite's is.

Also, although time.time() does return a float, we really only care about seconds so wouldn't lose anything by storing an INTEGER in the database and always doing int(time.time()) when comparing timestamps (that is, throwing away fractional seconds). Also, within Twisted, it's far better to use IReactorTime.seconds() anyway. So whenever it's convenient (e.g. whenever we have a reactor instance to call .seconds() on) we should probably prefer that. As a side-effect, this makes testing things that use that interface properly way easier.

Personally, I think the integer based approach is far less likely to have weird errors or rounding issues etc. and is thus overall way safer.

[daira]

Replying to dawuud:

ok here's my solution: ​https://github.com/david415/tahoe-lafs/tree/2505.db-schema-types.0

i've replaced NUMBER and TIMESTAMP with REAL. According to this link, TIMESTAMP is useful if you sometimes want to have a default value of the current time; we do not.

​https://androidcookbook.com/Recipe.seam?recipeId=413

All tests pass. Review needed.

Reviewed.

[dawuud]

pushed changes here: ​https://github.com/david415/tahoe-lafs/tree/2438.magic-folder-stable.9

[daira]

This is causing failures on Windows exactly because of the problem in comment 3. On Linux the problem is hidden because the events caused by writing the downloaded file are different (see #2737).

[daira]

We changed the types to integer nanoseconds (currently this is only on the 1431.windows-fixes.3 branch).