Explicitly set locale for some test files

[MichaelChirico]

I was seeing naming mismatch errors without setting locale to C here.

[SebKrantz]

Thanks @MichaelChirico! Wondering what brings me the honor of you making these edits to collapse's testing facilities?

Also wondered, did CRAN force you to do something about

Result: NOTE
  File ‘data.table/libs/data_table.so’:
    Found non-API calls to R: ‘LEVELS’, ‘NAMED’, ‘SETLENGTH’,
      ‘SET_GROWABLE_BIT’, ‘SET_S4_OBJECT’, ‘SET_TRUELENGTH’,
      ‘UNSET_S4_OBJECT’

? I have some of the same notes now.

[MichaelChirico]

Wondering what brings me the honor of you making these edits to collapse's testing facilities?

Finally got around to importing {collapse} as a third-party package internally. We have a "unique" testing environment different from CRAN which can dredge up these sorts of edge cases.

Also wondered, did CRAN force you to do something about

We have not submitted to CRAN yet, not sure what will be enforced. Follow along at Rdatatable/data.table#6180 :)

[SebKrantz]

Ok, thanks! And cool that collapse can now be used at Google :)

[SebKrantz]

Probably not the way you'll go about it @MichaelChirico, but for the radixsort I implemented a custom TRUELENGTH/SET_TRUELENGTH, which avoids an ALTREP check that made the function noticeably more expensive: https://github.com/SebKrantz/collapse/blob/master/src/base_radixsort.h.

[MichaelChirico]

Thanks for sharing! I'd still like to strive to use the public API if possible, let's see how it goes :)

[SebKrantz]

@MichaelChirico further to this, I've now done things the dirty way again. CRAN was not responding anymore to my submissions, which I guess is now the way they start cracking down on all non-API compliant smaller packages (not helpful). Just sharing in case it is useful (I can see that the ALTREP solution of aitap (Ivan) is cool but would of course overhaul a large part of your architecture, don't allow chmatch() to continue working the smart way you are doing it now etc.). Turns out defining (SET_)TRUELENGTH is rather easy:

#define STDVEC_TRUELENGTH(x) (((VECSEXP) (x))->vecsxp.truelength)
// No method to set ALTREP_TRUELENGTH (gives error)
#define SET_TRULEN(x, v) (STDVEC_TRUELENGTH(x)=(v))
// ALTREP_TRUELENGTH is 0: https://github.com/wch/r-source/blob/48f06c1071fea6a6e7e365ad3d745217268e2175/src/main/altrep.c#L345
#define TRULEN(x) (ALTREP(x) ? 0 : STDVEC_TRUELENGTH(x))

SETLENGTH on the other hand is complicated (see here). Better get rid of that one if you can. Anyway, just sharing in case you are also getting serious problems with CRAN. If the plan is still to do everything in ALTREP I will follow along... . For now, honestly, I just want my software to continue working the way it did and I don't want to deal with CRAN (or rather, I had no leverage to convince them to do anything about this).