Fix incorrect ordering in StrL comparison functions

[gorcha]

Hi @evanmiller,

While investigating an unrelated issue with stata files crashing R in haven (tidyverse/haven#600) I noticed that readstat isn't implementing StrL ordering correctly.

In the dta spec it says:

3. GSOs must appear in "ascending" order of (v,o). Ascending order is defined as the same order as they appeared in <data>...</data>: ascending v for o==1, followed by ascending v for o==2, ....

The comparison function used for searching the StrL array is currently assuming that it's ordered by v then o, i.e. ascending o for v == 1, followed by ascending o for v == 2 etc. As a result the bsearch thinks that a lot of StrL references don't exist so they're missing from the imported file.

The same assumption has been made for writing (as one would expect 🙂), so files written by readstat roundtrip successfully.

This PR fixes the comparison functions for reading and writing.

Test data

I was double checking results against another R library that has an independently implemented parser and noticed that haven/readstat was missing a bunch of string values in the imported file.
For reference, the file I was using for testing was linked by the issue creator, and can be found in this repo:
https://github.com/sjkiss/ces19/raw/main/2019%20Canadian%20Election%20Study%20-%20Online%20Survey%20v1.0.dta

[evanmiller]

Nice catch, simple fix! Will merge.

[gorcha]

No worries, always happy to help! 🙂