UTILITIES:

• R CMD check now handles archives with extension .tar or .tar.zstd
  (where zstd compression is supported by the R build).

BUG FIXES:

• t.test(c(1:3, Inf)) and similar no longer produce an error but
  return a (still not so useful) "htest" result, fixing PR#18901,
  thanks to Jesse Alderliesten.

• attr(., "tsp") <- val now uses getOption("ts.eps") instead of
  hardwired 1e-5; consequently, ts(.., ts.eps=*) now passes ts.eps
  to the "tsp" setting C code; both fixing a long-standing ‘FIXME’.

• insertSource() now ignores the internal .packageName object,
  avoiding a superfluous message.

• In static HTML help, links to vignette files from the default
  doc/index.html page now also work for packages not installed in
  the default .Library, thanks to a report by Patrice Kiener.

• fixInNamespace("<S3method>") failed to update the S3 methods
  table when the generic was not on the search path.

• In plain-text help (Rd2txt), an initial newline from an Rd inline
  \eqn no longer breaks the paragraph.

• hist(*, log = "x") now works without a warning, thanks to Martin
  Smith's PR#18921.

• Subassigning "POSIXlt", i.e., <tdat>[i] <- val and <tdat>[[i]] <- val
  now rebalance as they should, thanks to Mikael Jagan's
  PR#18919.

• <POSIXlt>[*] (re-)setting "balanced", fixing PR#18681 comment wch#7,
  thanks to Mikael Jagan.

• All four of {col,row}{Sums,Means}(Z, na.rm=TRUE) now correctly
  work with complex Z where is.na(Re(Z)) differs from is.na(Im(Z)),
  fixing PR#18942, unearthed by Dirk Eddelbuettel.

• Fix for glyph rendering on the quartz() device when there is
  other (“normal”) text drawn on the device. The problem was that
  the text transformation matrix was not reset so glyphs would be
  rendered incorrectly (often completely outside the device, i.e.,
  not visible).

• Functions install.packages() and download.packages() again
  consult option download.file.method when the download method is
  unspecified.

• Tanguy Barthelemy and colleagues at the ‘R Dev Day’ following
  Rencontres R in May 2025 extended the help page of lm(), fixing
  PR#18058. As suggested by Thomas Soeiro, such not es were also
  added to the glm(), poly() and splines::bs() and ns() pages.

• lbeta(1i, 1) now signals an error, as lbeta() is not implemented
  for complex, fixing PR#18946 from Ben Bolker and Kasper
  Kristensen.

• The Cairo-based SVG device uses pt as the default document unit
  also with Cairo >= 1.17.8 (PR#18912).

• pretty(*, eps.correct = 2) has been fixed, e.g., to avoid over 1
  million length result for pretty(c(0, 1e-322), eps.correct = 2).

• When "POSIXlt" date-time objects are NA-padded from subsetting or
  increasing length in the [ or length<- methods, the
  "balanced" attribute is set to NA, now; noted in PR#18681, thanks
  to Mikael Jagan.