Drop direct use of NAMED and LEVELS #6420
Conversation
Since data.table now depends on R >= 3.3, the backports are no longer needed. Moreover, MAYBE_SHARED is currently a function, while MAYBE_REFERENCED expands to !NO_REFERENCES (which is a function). In debugging output, show MAYBE_REFERENCED (NAMED > 0) instead of NAMED.
|
thanks for the great resource! would you mind
re:LEVELS, is there a performance penalty for the new approach? |
|
Generated via commit d626df2 Download link for the artifact containing the test results: ↓ atime-results.zip Time taken to finish the standard R installation steps: 11 minutes and 21 seconds Time taken to run |
Currently, Will split and update the PR soon-ish. Thanks for the comments! |
getCharCE appeared in R-2.7, making it possible to check for strings _marked_ as UTF-8 or Latin-1. There is no marking as ASCII, so fixing IS_ASCII will have to wait for R >= 4.5.
|
nice NEWS item! |
* Drop direct use of NAMED Since data.table now depends on R >= 3.3, the backports are no longer needed. Moreover, MAYBE_SHARED is currently a function, while MAYBE_REFERENCED expands to !NO_REFERENCES (which is a function). In debugging output, show MAYBE_REFERENCED (NAMED > 0) instead of NAMED. * Almost drop direct use of LEVELS getCharCE appeared in R-2.7, making it possible to check for strings _marked_ as UTF-8 or Latin-1. There is no marking as ASCII, so fixing IS_ASCII will have to wait for R >= 4.5. * NEWS entry for NAMED, LEVELS reduction * fine-tune NEWS --------- Co-authored-by: Michael Chirico <[email protected]>
These are the remaining easy pickings for #6180.
Costs:
verbose = TRUE,assignnow only prints whetherREFCNT()is>= 1and>= 2, not its value directly.Rf_charIsASCIIis only an "eapi" function, not a bona-fide "api" function, and it is currently only in R-devel. But it doesn't raise NOTEs.NAMED/REFCNTversion information gathered here (shameless plug).