Ensure we're not UNPROTECT()ing 'x' in gsumm

[MichaelChirico]

rchk is still throwing issues around here:

Function gmean
  [UP] allocating function gather may destroy its unprotected argument (x <arg 1>), which is later used. /rchk/packages/build/CaNpi0PW/data.table/src/gsumm.c:600
  [UP] calling allocating function gather with a fresh pointer (x <arg 1>) /rchk/packages/build/CaNpi0PW/data.table/src/gsumm.c:600
  [UP] unprotected variable x while calling allocating function Rf_allocVector /rchk/packages/build/CaNpi0PW/data.table/src/gsumm.c:601

The only thing I can think of is that the UNPROTECT() is applying to x, not as:

The protection mechanism is stack-based, so UNPROTECT(n) unprotects the last n objects which were protected.

IINM that means we have to keep as protected as long as x is? Or we have to clutter the code to do UNPROTECT(2) (x, as), followed immediately by PROTECT(x) to get it back under protection? Not even sure that will work, and probably not worth the clutter.

[github-actions]

Generated via commit 5c7527f

Download link for the artifact containing the test results: ↓ atime-results.zip

Time taken to finish the standard R installation steps: 11 minutes and 33 seconds

Time taken to run atime::atime_pkg on the tests: 3 minutes and 45 seconds

[ben-schwen]

IINM that means we have to keep as protected as long as x is? Or we have to clutter the code to do UNPROTECT(2) (x, as), followed immediately by PROTECT(x) to get it back under protection? Not even sure that will work, and probably not worth the clutter.

That should work and apparently some packages are doing it that way.

https://github.com/search?q=org%3Acran+%2FUNPROTECT%5C%28%5B2-9%5D%2B%5C%29%3B%5Cs*PROTECT%5C%28%5Cw%2B%5C%29%3B%2F&type=code

[MichaelChirico]

That should work and apparently some packages are doing it that way.

Nice find! Great that GitHub regex search works across lines. Applied here now, with comment.