BUG: Problem with string vars when subsetting large df

[nickeubank]

Struggling with a very strange problem:

The data I'm using isn't really public, so I've been trying to come up with a MWE, but haven't had luck. The problem disappears if I take the first 10, 100, or 1000 rows. Only when I keep 10,000 rows (head -10000 the_file.csv >> mwe.csv) does the problem persist. It also goes away if I drop a lot of the columns. :/

Does this mean anything to anyone?

[nickeubank]

OK, while not sensitive, I'd rather not post file on github, but could share 10000 rows with an admin if it'd help.

[nickeubank]

ah, ok -- the weakrefs must be lost. Check out row 3 columns :trans. The letter is changing over time. It must be pointing to memory used by something else.

[EDIT:] Jumped again:

[nalimilan]

Yes, that's a CSV/WeakRefString bug. I think it's fixed by JuliaData/WeakRefStrings.jl#17. Feel free to comment there if it persists after the fix is merged.