This blog is updated daily.
A general description is here.
‘tools/fetch-recommended’ can be used instead of ‘tools/rsync-recommended’ to fetch recommended packages into R sources using ‘curl’ on systems without ‘rsync’ or behind firewalls.
C++ standard specifications (‘CXX_STD =’ in ‘src/Makevars*’ and in the ‘SystemRequirements’ field of the ‘DESCRIPTION’ file) are now checked more thoroughly. Invalid values are still ignored but now give a warning, as do contradictory specifications.
(Preliminary) support for C++26 has been extended to Windows.
A non-zero exit status from ‘cleanup’, ‘cleanup.win’ or ‘cleanup.ucrt’, if requested via options ‘--clean’ or ‘--preclean’, is now reported with a warning.
‘all.equal(obj, simple, check.class=FALSE)’ now is true, also when ‘simple’ is a bare atomic vector and ‘obj’ has a simple class, fixing the first part of PR#18971 thanks to Jan Gorecki.
‘str(x, give.attr=FALSE)’ no longer shows attributes when ‘x’ is a zero length ‘"Date"’ or ‘"POSIXt"’ object.
Tweaks to binning for ‘bw.SJ()’ and ‘bw.ucv()’ in the very rare case of data which have an extremely small range compared to their absolute values, e.g., ‘x <- 6e9 + 6:9’.
Formatting book-type ‘bibentry’ objects now converts LaTeX accents also in the publisher and series fields.
‘model.frame(~1, list(), na.action=na.pass)’ and similar “border-line” uses no longer produce invalid data frames, fixing PR#18977, reported with patch by Mikael Jagan.
‘length(<POSIXlt>) <- v’ is more careful about balancing, notably when ‘v’ is not integer, thanks to Suharto Anggono's remarks on the mailing list R-devel.
‘approx(<x_with_ties>, <y_with_NA>, na.rm = FALSE)’ now should always call the ‘ties()’ function, fixing PR#17604 reported by Bill Dunlap.
‘besselJ(1, 1e-15)’ and similar now give correct results, thanks to Leo Mada and other “R-help”ers.
‘vignette(<pkg>::<topic>)’ is now a documented usage variant and confines vignette retrieval to the specified package.
‘pretty(ch)’ again correctly works with ‘ch’ a character vector of numbers.
‘persp()’ labels the three axes correctly also when C level ‘atan2pi()’ is available, fixing PR#19007 by Klaus Schliep.
Large (tall) subscripts in plotmath expressions are now positioned correctly. They were being positioned too high.