Skip to contents

Changelog

Source: NEWS.md

nonmem2rx 0.1.11

  • Regenerate the rxSolve.nonmem2rx() method so it no longer passes the order solver argument, which was removed from rxode2::rxSolve(). Passing it caused solving a translated model to fail with unused argument: 'order' against the current CRAN rxode2.

  • nonmem2rx()’s inputData argument now also accepts a data.frame of the already read-in NONMEM input dataset (in addition to a file path). This is useful when importing a model from a different system where you have the data to validate against but the file paths in the control stream do not match (#186). The columns are assumed to be in $INPUT order and the usual $INPUT names, DROP, IGNORE/ACCEPT filters and record subsetting are applied.

  • ADVAN5/ADVAN7 general linear models (which NONMEM itself solves with matrix exponentials) are now translated to rxode2’s native matrix-exponential matExp() model (cmt() declarations plus k_<from>_<to> rate constants) by default, instead of explicit d/dt() ODEs. The previous ODE translation is retained and can be selected with the new matexp=FALSE argument (or options(nonmem2rx.matexp=FALSE)). Other model types are unaffected, and if the installed rxode2 does not support matExp() the ODE translation is used with a warning.

  • NONMEM mixture models ($MIX) now translate to the native rxode2/nlmixr2 mixture support (mix()), replacing the previous rxord() simulation of the sub-population. When the mixture probabilities are simple parameters (e.g. P(1)=THETA(5)), the imperative MIXNUM/MIXEST branching in $PK/$PRED is collapsed into readable mix() calls (e.g. V <- mix(VCM, p1, VCF)), the probabilities are registered on the model (ui$mixProbs) so the model estimates natively under focei and saem, and MIXEST/MIXNUM map to the reserved mixest component. Models whose probabilities are not simple parameters fall back to the previous rxord() translation. Simulation-based validation remains gated for mixtures because NONMEM’s per-subject sub-population assignment is not recoverable for a faithful prediction comparison.

  • Support the NONMEM $DATA numeric-comparison operators .EQN. and .NEN. in IGNORE=/ACCEPT= filters (#195). These request that the data item be converted to numeric before being compared, so they now translate to as.numeric(.data$COL) == value / != value. .NEN. was previously unparseable (silently dropping the filter), and .EQN. was treated as a plain == without the numeric coercion.

  • Support assignment to the NONMEM COM(#) communication array (declared with $ABBREVIATED COMRES=#). Previously only reading COM(#) was translated, so control streams that assigned COM(#) (e.g. IF(NEWIND.LE.1) COM(1)=-1) failed to parse (#228). A COM(#) assignment now translates to the rxode2 sticky variable rxCOM_#_, which retains its value from record to record like the NONMEM COM array (see the rxode2 sticky variable vignette).

  • Add optional LLM-assisted residual error detection to as.nonmem2rx() (via the ellmer package) for models imported without a residual error specification ($predDf). When no chat engine is supplied, the default engine is now chosen dynamically: getOption( "nonmem2rx.llmProvider") is honored first (accepting any exported ellmer::chat_* engine by name), otherwise the first provider with a configured API key is auto-detected (e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY), instead of always requiring Claude. Any ellmer chat object may also be passed directly through the new chat argument.

  • Add integer overflow guards in the C-level string buffer (src/sbuf.c). sAppendN, sAppend, and addLine previously computed the new allocation size as sbb->o + 2 + n + SBUF_MXBUF (or analogous expression). When the user-controlled n was large enough this expression overflowed int to a negative value, which R_Realloc then converted to a huge unsigned size and crashed. The guard converts this into a clean R error.

  • Document known (int)strlen(gBuf) cast in all 10 NONMEM-record parser entry-points (src/abbrec.c, src/abbrev.c, src/data.c, src/input.c, src/lst.c, src/model.c, src/omega.c, src/sub.c, src/tab.c, src/theta.c). Inputs at or above INT_MAX bytes cause silent length truncation passed to dparse(). A long-term fix will switch each call site to udparse() once dparser-R ships that symbol to CRAN.

  • Fix implicit ptrdiff_t to int truncation in rc_dup_str (src/records.c). When the parser passes a string segment longer than INT_MAX bytes (or a NUL-terminated string of that length), the pointer difference / strlen result was silently cast to int, truncating the length to a wrong (often negative) value. The new guard rejects such inputs with an informative R error.

nonmem2rx 0.1.10

  • Bug fix for covariance matrices that span multiple FORTRAN output pages

  • Bug fix for multiple errors are imported using nonmem2rx()

  • Change NONMEM import for mixture models to prepend NM to MIXNUM and related import items so they will not interact with rxode2/nlmixr2’s handling of mixture models.

nonmem2rx 0.1.9

CRAN release: 2025-11-27

  • Try to make sure all the values that can be numeric are numeric #208

  • Use qs2 since qs is being archived

nonmem2rx 0.1.8

CRAN release: 2025-08-29

  • Better handling of ytype

  • Added work-around for new Rstudio completion

nonmem2rx 0.1.7

CRAN release: 2025-07-15

  • Added $ERROR (ONLY OBS) support
  • if statements can be upper or lower case #196

nonmem2rx 0.1.6

CRAN release: 2024-11-28

  • Add more flexible replacement of ETA and other values #188

  • Add ability to use nonmem2rx with a NONMEM control stream text input

nonmem2rx 0.1.5

CRAN release: 2024-09-18

  • Be more forgiving in the validation and remove IDs without observations when solving the IPRED problem.

  • Binary linkage to dparser changed to structure only, meaning nonmem2rx may not have to be updated if dparser is updated.

nonmem2rx 0.1.4

CRAN release: 2024-05-29

  • When reading NONMEM results from xml will try nm: prefixed tags and non-nm: prefixed tags.

  • Omega and Sigma prior estimates are currently ignored (theta priors were already ignored)

  • Improve reading in theta values from the xml

  • Read all NONMEM files using latin1 encoding to allow single byte parser to work

  • When lines in the NONMEM input dataset start with # they are now ignored.

  • When all IDs are zero, NONMEM assumes restarting time gives different IDs; this is now reflected in NONMEM translation of IDs.

  • With linCmt() parsing, expand the scope of conflicting parameters that will be renamed with an import.

  • Added better parsing for ELSE where there is another IF on the next line.

  • Prefixed conflicting VP with rxm. when linCmt() models to be more accommodating when importing linear compartment models.

nonmem2rx 0.1.3

CRAN release: 2023-12-12

  • Added explicit requirement for rxode2 2.0.13

  • Added support of DADT(#) statements on the right side of the equation, i.e. DADT(3) = DADT(1) + DADT(2) (#164)

  • Added support of ADVAN#, TRANS# (#161)

  • Added more NONMEM-specific solving options

  • Fixed security related format issues as requested by CRAN #167

  • Now omega, thetaMat, dfObs and dfSub are incorporated into model function (by default). You can change this with the nonmem2rx keep argument

  • Using the rxode2 2.0.13 makes sure that the solves for models where the endpoint is not determined in the typical nlmixr2 style will validate more often (due to a bug in solving in rxode2).

nonmem2rx 0.1.2

CRAN release: 2023-07-03

  • Added support for ADVAN5 and ADVAN7 models

  • Add parsing of accept/ignore characters for example IGNORE=(C='C') (See Issue #140)

  • Add more robust reading of NONMEM information (and add source) in nminfo() (See issue #142)

  • Since NONMEM does not protect divide by zeros by default, the default for solveZero is changed to solveZero = TRUE for nonmem2rx objects.

  • Fixed bug for renaming eta and theta when they are renamed so that the ui$iniDf does not match the theta# or eta# (Issue #153)

  • Turned off testing of the as.nonmem2rx example since it took too much time (according to CRAN)

nonmem2rx 0.1.1

CRAN release: 2023-06-01

  • Fix internal memory issue (LTO, valgrind etc)

nonmem2rx 0.1.0

CRAN release: 2023-05-26

  • Added a NEWS.md file to track changes to the package.