Movatterモバイル変換

Rmpfr News

Changes in version 1.1-2 [2025-10-21, r437]

NEW FEATURES

• New.mpfrSizeof() providing the small integersizes of the platform's MPFR library configuration.

BUG FIXES

• Formpfr(x, precBits=*) it is now documented thatprecBits must not be larger than largest integer, i.e.,2^31 - 1 = 2147483648. An error message for larger valuesis now easier to understand.

• Correct ‘configure.ac’, ‘Makevars’, etc needed whenMPFR exponent size does not match limb size (as on r-universe Windows),fixing R-forge Rmpfr bug #6865 by Mikael Jagan.

Changes in version 1.1-1 [2025-07-18, r430]

NEW FEATURES

• addeddchisq() for completeness, simply callingdgamma(x, df/2, scale = 2).

BUG FIXES

• dgamma() accidentally used double precision instead of"mpfr" in some cases.

Changes in version 1.1-0 [2025-05-08, r426]

NEW FEATURES

• Newpgamma(), simply viaigamma(), to becomemore visible to probability and statistics.
  Note that in some regions of its domain, it is currently severelylimited (both slow and inaccurate)as by its implementation in MPFR, see the ‘Warning’ in?igamma.

BUG FIXES

• The default for ourdbinom()'suseLog argumentwas determined alone byany(abs(x) > 1e6); it is now alsotrue, for smallerx and largesize or extremeprob, e.g., indbinom(7L, mpfr(1e10, 96), prob = 1/4).

  Notably a bad thinko has been fixed which lead to only double precision accuracyresults foruseLog = TRUE.

Misc

• Requiring MPFR >= 3.2.0 such thatgamma_inc() isavailable for ourigamma() andpgamma().

Changes in version 1.0-0 [2024-11-15, r422]

USER-VISIBLE CHANGES

• The low-level bug fix about conversion has slightly changed theABI and the low-levelvalidity method on Windows: Non-finitempfr-numbers, e.g.,NA,Inf, now have differentmantissa slot@ d, now starting with0L instead ofpreviously-1L.

NEW FEATURES

• Newnum2bigq(x) finds “small” denominatorbigq / ‘bigRational’ approximation to numeric or <mpfr>x.Basically aMASS::fractions() version forRmpfr andgmp.

BUG FIXES

• format(mpfr()) no longer warns.

• informatDec(): usep*log(2, 10) instead oflog(2^p, 10)!

• low-level C fixes in ‘src/convert.c’ thanks to analysisand proposition by Mikael Jagan. Entails update ofvalidity()for"mpfr1".

• add\link[pkg]{...} where needed.

Misc

• new ‘../tests/special-fun-dgamma.R’ file; partly from ‘..../special-fun-ex.R’:accuracy checking for more accuratestirlerr() inR 4.4.0 and later.

Changes in version 0.9-5 [2024-01-20, r407]

NEW FEATURES

• New.mpfr2bigq(m) transforms <mpfr>-number vectors tobig rational akabigq ones (from packagegmp).

• New low-level utility functions.mpfr2d() and.mpfr2i().

BUG FIXES

• x == y and other comparisons of two"mpfr"operands now returnNaN when an operand isNaN (mpfrnumbers are neverNA).

• dbinom(x, *) anddnbinom(x, *) now use theprecision ofx when it is an (integer valued)"mpfr"object, and then notice or even error when precision is lost asx is coerced to integer.

• (erange-related comments in examples)

• chooseMpfr(a, n) gives0 whenn < 0, asR'schoose() and whena is integer valued, may usethe equivalent ofchooseMpfr(a, a-n) whena-n < n to bemore accurate and faster.

• .mpfr2bigz(m) now also works for really largem.

Changes in version 0.9-4 [2023-12-04, r399]

BUG FIXES

• Fixed Windows-only (long = 32 bit) bug; adapt mpfr1-validity() check.

• allow ‘.Platform$endian != "little" to ’work' in validity().

• format ("%ld" etc) fixes in ‘src/utils.c’

• fix "lost braces"

Changes in version 0.9-3 [2023-07-27, r394]

BUG FIXES

• any() andall() no longer damageR's internalFALSE or TRUE, fixing R-forge Rmpfr bug #6764 by Andrew Manderson.

Changes in version 0.9-2 [2023-04-21, r392]

USER-VISIBLE CHANGES

• formatMpfr() and hence theformat() method for"mpfr" etc now uses ascientific = <number> the same asbaseformat.default(); accidentally it had thesign of the number interpreted differently and failed to usee.g.,getOption("scipen") correctly.

NEW FEATURES

• log(<mpfr>, base=10) now works.

BUG FIXES

• igamma(),hypot(), andatan2() no longeronly return 53 bit (or more generally, mpfr default precision)results.

• Newmax2_prec() utility instead ofimax2(),also curing compilation warning.

• Internal (exported) objects are now (somewhat) documented, aseventually required byR.

Changes in version 0.9-1 [2023-01-30, r386]

BUG FIXES

• ldexpMpfr(f, E) now returns an"mpfr" vector(instead of a"mpfr1").

Changes in version 0.9-0 [2023-01-16, r382]

NEW FEATURES

• Add theextendInt="*" option tounirootR()adopting its addition in baseRuniroot().
  Also add an optionalverbDigits argument only used in caseverbose is true.

• NewqnormI(), the normal quantile function, computed viainversion of (our potentially arbitrarily accurate)pnorm(),usingunirootR() root finding.

• sapplyMpfr() getsdrop_1_ = TRUE option,changing default behaviour to return an"mpfr" vectorinstead of a 1-column"mpfrMatrix".

BUG FIXES

• sapplyMpfr() now gets the correct dimension when it returnsan"mpfrMatrix" (or"*array").

• Fix\eqn{(\gamma)}{} - as requested by KH (June 9).

• .mpfr_gmp_numbbits() is documented now (notably as Isee it is only 32 on Winbuilder (!)).

Changes in version 0.8-9 [2022-06-02, r363]

BUG FIXES

• Embarrassing thinko indpois (inuseLog=TRUEcase) fixed.

Changes in version 0.8-8 [2022-06-01, r362]

NEW FEATURES

• Our “mpfr-ized”all.equal() is now based onhiddenall.equalMpfr() which uses a smart defaulttolerance for all methods, and shows much less digits if thereare differences (compatibly toall.equal.numeric()).

BUG FIXES

• Ourdpois(x, lambda, *) now worksR-compatibly forlambda=+Inf giving 0 or-Inf whenlog=TRUE.

• formatMpfr(x, scientific=FALSE) now “works”.

Misc

• ‘man/*.Rd’: get rid of some\ escapes (needed in past; wrong now)

Changes in version 0.8-7 [2021-10-27, r353]

BUG FIXES

• In ‘Ops.c’'sR_mpfr_mod(), no longer allocatebut never use nor freerr - fixing YAL (yet another leak).

Changes in version 0.8-6 [2021-10-08, r351]

BUG FIXES

• In ‘convert.c’'sR_mpfr_frexp(), dompfr_clear(*), fixing a valgrind-detectable leak.

• additionally use single mpfr_exp_t and its pointer (and assignto R allocated vector inside main loop, fixing a valgrind"unitialized value" case.

Changes in version 0.8-5 [2021-10-05, r349]

NEW FEATURES

• New mpfr-izeddt(x, df, ..); noncentralityncpnot yet supported.

• New arithmetic functionsfrexpMpfr() andldexpMpfr(), corresponding to C's (and CRAN packageDPQ's)ldexp() andfrexp() functions.

• sapplyMpfr() now also returns"mpfrMatrix" or"mpfrArray" when appropriate.

BUG FIXES

• Improved`[.Ncharacter` method forformatHex() etc.

Misc

• No longer include ‘Rdefines.h’ as it is somewhat deprecated.

• update both ‘configure{,.ac}’ (forautoconf 2.71).

Changes in version 0.8-4 [2021-03-24, r341]

NEW FEATURES

• dpois(x, *) now also gets an explicit optional argumentuseLog instead of just switching to log-scale by the sametest as definesuseLog's default.

• The default method ofmpfr(r) now also works whenr is alist of"mpfr1" objects, e.g., resultingfromVectorize(.) or similar applied to mpfr-vectors.

BUG FIXES

• fixed problem detected by clang-UBSAN testing in ‘src/Ops.c’(from arithmetic bug fix in 0.8-3).

Changes in version 0.8-3 [2021-03-22, r340]

NEW FEATURES

• dnbinom(x, *) anddbinom(x, *) now also work whenx is too large to be coerced tointegerand theyget a new optional argumentuseLog (with a smart default) inorder to choose log-scale computation also forlog=FALSE.

BUG FIXES

• For arithmetic (`+`,`*`, etc), the check to seeif a numeric can be validly coerced to a long has been amended suchthat 9223372036854775808 is no longer accidentally coerced intonegative.

• dpois(x, lambda) now works via log-scale in caseexp(-lambda) orlambda^x would under- or overflow evenfor mpfr-numnbers.

  Analogously,dbinom() anddnbinom() work via log-scalein case even mpfr-arithmetic would overflow.

Changes in version 0.8-2 [2020-11-10, r337]

NEW FEATURES

• The workhorse for all matrix multiplication (%*%,crossprod(), andtcrossprod()), i.e.,.matmult.R(x, y, *)(not exported), gets new optional argumentsfPrec = 1 andprecBits, which defaults to the maximum ofgetPrec(x)andgetPrec(y).To get correctcrossprod() and tcrossprod() generics (with aformal... argument), need a newgmp release, aswe get these two generics fromgmp.

• matmult(x,y) is identical tox %*% y, butmatmult() has further optional argumentsfPrec andprecBits, see above.

• Newis.mpfr(x) function; simple, with fast pre-test.

• InpbetaI(q, ..), whenq is a bigrational, i.e.,of class"bigq" (packagegmp), the computations areexact now, using big rational arithmetic.

• Newdnbinom() function (with corresponding newconflicts() with thestats package baseRfunction), providing an"mpfr"-number version of the negativebinomial probabilities.

• .mpfr_erange_set() now can setboth exponentrange limits simultaneously, and now returns invisiblyTRUE ifthe change succeeded.

• Newlog1mexp() andlog1pexp(), as "from"copula, notably as the vignette has been here, and authoredby me.

BUG FIXES

• mpfr(mm) (and similar) now works for"bigq" or"bigz" matrices, thanks to a report by Jerry Lewis.

• Arithmetic and other"Ops" between"mpfr"and"bigq" aka bigrational numbers, now use the“inherent” precision of the bigrational.

• chooseMpfr(a, n) and similar now “work” whenn is of length zeroand when an experimental version ofstopifnot(*, allow.logical0=FALSE) is used.

• Ourcbind() andrbind() methods withsignature = "mNumber" now keep and construct column and rownames as the corresponding base functions, obeyingdeparse.level.

• Fixed "not-yet"-as-cran ‘NOTE’Undeclared packagesdfoptim,pracma in Rd xrefs.

Changes in version 0.8-1 [2020-01-14, r323]

BUG FIXES

• Provide dummy C function in the case MPFR library is older than3.2.0, thanks to Brian Ripley. Should work around checkNOTEs.

Changes in version 0.8-0 [2019-12-05, r321]

USER-VISIBLE CHANGES

• Formatting inclprint()ing by default uses a"+"after the exponential character ("e" by default).

NEW FEATURES

• Provide new low-level utilities.mpfr_formatinfo(),.mpfr2exp(), and.mpfr_erange_is_int().

• Renamed low-level utility functions to use_ instead of. keeping initial".", e.g.,.mpfr.gmp.numbbits() to.mpfr_gmp_numbbits().

• formatMpfr() gets a new optionaldecimal.plus = TRUE whichadds a"+" before positive exponents when in exponential (aka“scientific”) representation.

  Thempfr andmpfrArrayprint() methods geta correspondingdecimal.plus argument with a default that canbe set byoptions(Rmpfr.print.decimal.plus = *) to allowstrict back compatibility where needed.

• For MPFR (C library) version >= 3.2.0 (not by default in Fedora 30!),provide theincomplete gamma functionigamma(a,x)which is closely related topgamma(x,a), see help page.

• Now also export S3 method ofunique() for"mpfr",such thatbasefactor(<mpfr>) “works”.

BUG FIXES

• formatMpfr() and hence allprint()ing sufferedfrom an integer overflow bug with very large (base 2) exponents.

• .mpfr2str(x, *) is no longer dependent on theorder of the elements inx; consequentlyformat() andprint() may use less digits in case theprecision decreases alongx.

• adapt to new C compiler default behavior-fno-common,usingextern #include <Syms.h> in most ‘*.c’ files.

Changes in version 0.7-3 [2019-08-06, r305]

USER-VISIBLE CHANGES

• Decreased the default formax.digits to 999.

NEW FEATURES

• Providedgamma(x, shape) version, e.g., for small shapeparameter where most of the mass is on very smallx notrepresentable as double precision numbers.

• Low-level formatting function.mpfr2str() — called byformat() and henceprint() methods for"mpfr"objects — now finally obeys itsmaybe.full argument when itisFALSE, internally in C'smpfr2str().

BUG FIXES

• pnorm(<large>, log.p=TRUE) no longer underflows much tooearly, thanks to reports by Jerry Lewis.

• print.mpfrArray() now also uses a finitemax.digits default, preventing, e.g.,cbind(x, y) touse too many digits.

Changes in version 0.7-2 [2019-01-18, r299]

BUG FIXES

• str(<mpfr>) no longer callsformatMpfr(x, digits, *)with adigitsvector of the same length asx(which never worked correctly).

• seqMpfr(1, length.out=8) now works correctly.

NEW FEATURES

• unirootR() gets an option tonot warn onnon-convergence.

• Provide asummary() method for"mpfr" numbersclosely modeled aftersummary.default for numeric.

• mpfr(NULL) now works, returningmpfr(logical()).

• a simplesapplyMpfr() function, showing how to workaround the fact thatsapply() does typically not work with"mpfr" numbers.

Changes in version 0.7-1 [2018-07-24, r291]

BUG FIXES

• formatMpfr() for large low-precision numbers now usesscientific representation, fixing the bug RMH was reporting March17 already.

• outer() is “imported” from base, so itbehaves as anRmpfr function which dispatches e.g., whencallingtcrossprod().

Changes in version 0.7-0 [2018-01-12, r284]

NEW FEATURES

• .mpfr2list() andmpfrXport() gain an optionnames (for nicer output).

• formatMpfr() and theprint() method get a newoptionmax.digits with default9999 for the printmethods, to limit the number of digits printed for high precisionnumbers.

BUG FIXES

• For non-“regular” mpfr numbers, thed slot inthe"mpfr1" representation is now empty instead of“random”. This also eliminates valgrind warnings aboutuninitialized values in C.

Changes in version 0.6-2 [2017-05-29, r264], never on CRAN

NEW FEATURES

• The S3 classes"Hcharacter" and"Bcharacter"resulting fromformatHex() andformatBin() now“inherit from”"character" formally.

• They also got a`[` method, so subsetting shouldwork, including forarrays of these.

• The"mpfr" method ofstr() gains optioninternal.

BUG FIXES

• whenprint()ing mpfr numbers, the result no longer sometimesloses the last digit.

• dnorm() now works correctly with mpfr numbers;similarlydbinom() anddpois() should work in allcases, now.

• in ‘NAMESPACE’, alsoexportMethods(apply), soSNscan works.

• print(formatHex(..)),formatBin() andformatDec() now look better and are more correct; the firsttwo get a new optionexpAlign indicating to use the samenumber of digits for exponents (in “scientific” cases).

  Notably,mpfr(formatBin(mpx)) works for morempxobjects (of class"mpfr").

• format(mpfr(3,7), digits = 1, base = 2) no longercrashes (from inside MPFR).

• formatDec(mpfr(NA, 7)) now works.

• For non-“regular” mpfr numbers, thed slot inthe"mpfr1" representation is now empty instead of“random”. This also eliminates valgrind warnings aboutuninitialized values in C.

Changes in version 0.6-1 [2016-11-15, r249]

NEW FEATURES

• head() andtail() methods for"mpfrMatrix".

BUG FIXES

• C-levelmpfr2str() no longer calls S_realloc() withwrong "old size" (thanks to Bill Dunlap).

• c() now also works when its result is a length-0"mpfr" object.

Changes in version 0.6-0 [2015-12-04, r237]

NEW FEATURES

• mpfr() now is S3 generic with several methods, notably a"mpfr" method allowing to change precision or rounding mode.

• mpfr(),formatMpfr(), etc, now work with bases from 2to 62 (using digits, upper and lower case ASCII letters,62 == 10 + 2*26 characters), as this has been in MPFR since version3.0.0 (seempfrVersion), which is hence (implicitly)required forbase greater than 36.

• formatMpfr() gets a new argumentbase = 10 and canbe used to produce in other bases, notably binary (base = 2)or hexadecimal (base = 16).

• str(<mpfr>, ....) is now based onformatMpfr()and nicely shows numbers also out of the double precision range.

  Further, it now chooses a smart default for optional argumentvec.len.

• matrix(mp, ..) now also works whenmp is ofclass"mpfr".

• new matrixnorm() for severalkinds.

• new functionsformatHex() andformatBin()thanks to Rich Heiberger.

• mpfr(x) also works asinverse offormatBin andformatHex.

• roundMpfr() and mathematical functions such asjn, orchooseMpfr() get new optional argumentrnd.mode passed to the corresponding MPFR function.

• median(x),mean(x, trim) fortrim > 0now work fine for"mpfr" x, andquantile(x, *) nolonger needsnames=FALSE to avoid a warning.

BUG FIXES

• pnorm(.),j0() and similar special functionsnow preservempfrMatrix andmpfrArray classes.

• similarly,is.finite() etc keep thedim()ensionality for"mpfrArray" arguments.

• mpfr("0xabc", base=16) andmpfr("0b101",base=2) and correspondinggetPrec() now give the correctprecBits instead of too many.

• str(<0-length mpfr>) now works correctly.

Changes in version 0.5-7 [2014-11-27, r205]

NEW FEATURES

• .

BUG FIXES

• as.integer() now rounds “to zero” as forregularR numbers (it accidentally did round “to nearest”previously).

Changes in version 0.5-6 [2014-09-05, r203]

NEW FEATURES

• experimentalmpfrImport(),mpfrXport()utilities – as we found cases, where save()"mpfr" objectswerenot portable between different platforms.

• as(*,"mpfr") now also supports rounding mode"A" (“Away from zero”).

• Several hidden low level utilities also get arnd.mode option.

Changes in version 0.5-5 [2014-06-19, r190]

NEW FEATURES

• The result ofintegrateR() now prints even if awarning happened.

• pbetaI(x, a,b), the arbitrarily accuratepbeta() computation forintegera andb,now works for larger(a,b).

• Newly providingmpfr-versions ofdbinom(),dpois(), anddnorm().

• New utility functionsmpfr_default_prec(),.mpfr.minPrec(), etc, to get, check, set default exponentranges and precision.

• Newsinpi(),cospi() etc, notably forR >= 3.0.1.

Changes in version 0.5-4 [2013-10-22, r173]

NEW FEATURES

• .

BUG FIXES

• .

Changes in version 0.1-5 [2009-08-06]

NEW FEATURES

• First CRAN release on 'Publication:' 2009-08-14 20:24:02

• newpmin() andpmax(), improvingseqMpfr().

• new"Math" and"Math2" group methods, notably forround() andsignif().

• as(. , "integer") now works (via Cmpfr2i).

More details for old versions up to Feb. 2015:

• See file ‘ChangeLog’