Changelog

Source: NEWS.md

constructive 1.3.02025-09-19

1.2.0 couldn’t be installed on some OS, R versions, or combinations of the latter, this is addressed in 1.3.0 along with a few additional fixes and features so it ends up being a new minor version.

  • We construct external pointers differently, from a global store in the package’s namespace rather than from an address. It’s more accurate and the previous way seemed to be problematic on Windows. xml2::read_html
  • matrix() and array() are not used anymore when calling construct_dput(), this is consistent with dput().
  • A new “prototype” constructor for objects of class “S4” allows faithful construction of S4 objects that have an initialize() method
  • We construct object of the “xml_document” class using xml2::read_html() or xml2::read_xml()
  • .cstr_new_class() and .cstr_new_constructor() work with VSCode and VSCode based editors.
  • We’re a bit more careful when using recycling on data frame columns. Previously we were not accurate when recycled columns had some attributes.

constructive 1.2.02025-08-28

cynkra (https://cynkra.com) is officially a funder, already was from the start in practice, thanks cynkra!

Features and improvements:

  • We now support the “object” type, that many S7 objects are built on. The code to build an empty “object” is either getClass("S4")@prototype |> asS3(complete = FALSE) or S7::S7_object() |> structure(class = NULL)
  • We now support classes from the S7 package:
    • “S7_Object”
    • “S7_union”
    • “S7_S3_class”
    • “S7_generic”
    • “S7_class”
    • “S7_base_class”
    • “S7_any”
  • We now support classes from the ellmer classes :
    • “ellmer::TypeObject”
    • “ellmer::TypeJsonSchema”
    • “ellmer::TypeEnum”
    • “ellmer::TypeBasic”
    • “ellmer::TypeArray”
  • We now support the new ggplot2 classes:
    • “ggplot2::element_blank”
    • “ggplot2::element_geom”
    • “ggplot2::element_line”
    • “ggplot2::element_point”
    • “ggplot2::element_polygon”
    • “ggplot2::element_rect”
    • “ggplot2::element_text”
    • “ggplot2::ggplot”
    • “ggplot2::label”
    • “ggplot2::margin”
    • “ggplot2::theme”
    • “Guide”
    • “GuideAxisLogticks”
    • “GuideAxisStack”
    • “GuideAxisTheta”
    • “GuideBins”
    • “GuideColourBar”
    • “GuideColourSteps”
    • “GuideCustom”
    • “GuideLegend”
    • “GuideNone”
    • “Guides”
    • “labels”
    • “FacetGrid” (an old class that had been forgotten!)
  • We don’t signal a reconstruction inaccuracy when ggplot objects are reconstructed accurately except for stored language elements, e.g. we consider ggplot2::geom_point and geom_point to be equivalent in stored calls. It’s not fully satisfying because an aim of constructive is to spot such small details, but we also want to be practical and assessing the practical equivalence of the generated code seems more valuable than perfect accuracy there. Please open a ticket if this upsets you and we’ll work something out.
  • {cli} >= 3.1.0 is now required and cli::code_highlight() is now used, removing the dependency on {prettycode}. Turn it off with options(constructive_pretty = FALSE). The style argument of print.constructive_code() is deprecated in favour of the new argument code_theme. Thanks @wurli!

Bug fixes:

  • deparse_call() supports more corner cases (mostly for artificial cases you probably won’t encounter)
  • The construction of complex numbers now accounts for negative zeroes in imaginary and real parts
  • Buggy arg forwarding in some internal functions was fixed
  • The “list2env” constructor for environments was improved
  • calls to the function function (of type “language”) are correctly constructed
  • We don’t recycle with data.frame() when a column has attributes that don’t support it
  • S4 gains a “prototype” constructor, more general, that we fall back to by default where the class as an “initialize” method

constructive 1.1.02025-01-10

Features:

  • Implemented construction of R6 objects
  • Implemented construction of zoo and xts objects
  • Implemented "cbind" and "rbind" constructors for matrices
  • Now there’s no need to namespace opts_*() functions if the package is not attached, this mean we can call for instance constructive::construct(x, opts_tbl_df("tribble")) rather than constructive::construct(x, constructive::opts_tbl_df("tribble"))
  • New classes and constructors can now be implemented in your package without importing the package (adding to “Suggests” instead)
  • Added a hex logo

Fixes and improvements:

  • Fixed a display issue in construct_diff()
  • construct_clip() now shows a console message in addition to writing to the clipboard
  • Improved support of data frames, tibbles, and data.tables regarding non syntactic, empty, NA or duplicate names
  • Improved recycling in data frames
  • Implemented support for dates in tribble() calls when calling construct(x, opts_tbl_df("tribble"))
  • Improved support of classed matrices
  • Fixed timestamp precision issue
  • Improved support of "POSIXlt" class
  • Improved encoding support
  • Fixed construction of vectors with numeric_version(), package_version(), and R_system_version()
  • Fixed construction of edge case calls like (`fun<-`)() or (`a b`)()
  • Fixed construction of environments with NULL parents

constructive 1.0.12024-07-07

  • We fixed a typo that was breaking .cstr_new_class(, commented = TRUE)
  • We fixed a bug regarding high precedence infix operators that was breaking deparse_call()
  • “simpleUnit” is considered as a “grid” package class, not a “base” class and we fixed the check for corrupted “simpleUnit” objects
  • The construction of “ScalesList” objects is improved, it failed in some cases
  • The construction of “CoordCartesian” is improved and considers the “default” argument
  • Improved vignette consistency

constructive 1.0.02024-07-02

Internals, extension system, cosmetics

  • The internals have changed quite a bit, and a new extension system has been developed.
  • The vignette “extend-constructive” details it and how constructive works.
  • The package {constructive.example} was updated to reflect the new system.
  • We remove non API calls in the C code, to comply with new CRAN rules.
  • The pkgdown website categorizes the functions so the package can be explored more conveniently (Thanks Maelle Salmon @maelle for the suggestion).
  • The doc was improved overall
  • The README was reworked

New functions

New classes

  • Expression vectors are now supported (For some reason we had missed it!).
  • We implement constructors for the following base R classes:
    • citationFooter
    • citationHeader
    • hexmode
    • octmode
    • person
    • difftime
    • simpleError
    • simpleWarning
    • simpleMessage
    • simpleCondition
    • errorCondition
    • warningCondition
  • We support the class “integer64” from the {bit64} package. It was important because we can’t recreate NA or negative integer64 objects using base R only.
  • We support the class “blob” from the {blob} package.

New features and deprecation

  • Many constructive functions gain the classes argument that generalize construct_dput() and construct_base(), so users can enable or disable the idiomatic construction of some classes.
  • construct_reprex() and construct_multi() gain the include_dotted = TRUE argument so we can optionally disable the construction of objects such as .Random.seed int he global environment or .Class in the execution environment of S3 methods.s
  • construct() and construct_multi() gain the arguments unicode_representation and escape previously used by opts_atomic() and these are now not only applied on strings but also on element names and variable names.
  • We look at the encoding when constructing character vectors, so an UTF-8 “é” is not constructed like a latin-1 “é” anymore. Hopefully this will help some users out of encoding hell faster.
  • opts_tbl_df() gains a justify argument to control the justification of columns with constructor =“tribble”` (Thanks Jacob Scott @wurli for the implementation).
  • The imports and lazydata environments of packages are constructed with parent.env(asNamespace("pkg")) and getNamespaceInfo("pkg", "lazydata"). Before that they were constructed as regular environments.
  • When constructing environments we now lock environments and bindings when relevant.
  • We construct negative zeroes as -0. identical(0, -0)) is TRUE but 1/-0 is -Inf so it made sense to support them.
  • opts_environment() gains a "predefine" constructor and opts_environment(predefine = TRUE) is deprecated. The old way still works but warns and is not documented anymore.
  • In opts_atomic() the arguments unicode_representation and escape are deprecated, use the new opts_character() function or set them in the main function directly instead so they also affect symbols and argument names. The old way still works but warns and is not documented anymore.

Fixes

  • We solve some operator precedence issues in deparse_call()
  • Named vectors of length 1 are constructed properly
  • Objects are constructed properly if their names have attributes, contain NAs, or are named like c()’s arguments recursive and use.names
  • Circularity is detected when attempting to construct an environment refering to itself with an inappropriate constructor. It now fails explicitly rather than trigger an infinite loop or a low level error.
  • NAs and NaNs are not conflated anymore when compressing double vectors
  • complex numbers are constructed properly regarding the different combination of NA values in their real and imaginary parts.
  • We check for the S4 bit using isS4() and use asS4() when necessary
  • In opts_numeric_version(), opts_package_version() and opts_R_system_version() the incorrectly named “atomic” constructor is replaced by a “list” constructor
  • data frames, data tables and tibbles are now properly constructed when they contain columns that their idiomatic constructors cannot handle, such as “row.names”, or “stringsAsFactors” for data frames.
  • POSIXlt are constructed according to the R version, to account for the changes in R 4.3.0
  • We can construct objects with a “.Data” attribute, this used to fail because structure() has a .Data argument.
  • The default “row.names” attribute is built with the c(NA, -<nrow>) form, as in dput(), this solves some rare corner cases.
  • We fix some issues with raw strings when constructing character vectors
  • We fix some issues with NA, NaN, Inf dates and NULL timezones
  • We fix the environment attribute repair
  • Functions with non syntactic formals are constructed properly

constructive 0.3.02024-03-05

  • A new debugging tool, the construct_reprex() function, is introduced. It can be called in any function and will construct all variables and arguments where it’s called.
  • construct_multi() now constructs promises, in practice this is useful so we can construct the evaluation environment of a function, including the uneavaluated and potentially NSE args of a function
  • Set options(constructive_print_mode = <character>) where <character> is a vector of strings among "console", script, "clipboard" and "reprex". The default behavior is “console”. See `?constructive-global_options`
  • opts_atomic(escape = FALSE) (the default) now correctly uses surrounding single quotes for strings that contain double quotes and no single quotes.
  • deparse_call() is more robust and gains the arguments escape and unicode_representation that were already present in opts_atomic()
  • The ggplot object generation supports the internal changes of ggplot2 3.5.0, and the resulting construction is nicer.
  • Data frames can be reconstructed when their columns don’t have a data.frame() method
  • The “read.table” constructor for data frames supports the one_liner argument
  • roxygen2 is Suggested (not Imported anymore), the ellipsis dependency is removed
  • formulas have a “next” constructor, useful to see what formulas are at a low level
  • classes with S3 methods for length, [, [[ etc are handled better
  • .env() doesn’t crash anymore when provided a wrong or obsolete memory address
  • Integer vectors are constructed properly when they feature consecutive elements differing by more than .Machine$integer.max
  • Classed objects of types “…” and “externalptr” and are constructed properly
  • S4 construction uses the slot argument rather than the representation arg

constructive 0.2.02023-11-13

  • We don’t use {styler} anymore, performance is sensibly enhanced as a consequence.
  • The package works without having {prettycode} installed, but uses it if it’s installed
  • We default to using the magrittr pipe %>% for R versions that don’t support |>
  • NA levels are supported for the classes “factor” and “ordered”
  • Environment construction includes variables prefixed with a dot
  • When we don’t attach the package, we don’t need to use constructive:: in the ... before opts_* functions, for instance we can call constructive::construct(cars, opts_data.frame("read.table")).
  • quote({}) is now constructed as "{ }" rather than "`{`()"

constructive 0.1.02023-06-06

  • First CRAN release
  • construct() generates the code to build an object using idiomatic code, it wraps the lower level .cstr_construct() S3 generic.
  • We currently support 66 classes/types: “array”, “AsIs”, “atomic”, “classGeneratorFunction”, “classPrototypeDef”, “classRepresentation”, “constructive_options”, “CoordCartesian”, “CoordFixed”, “CoordFlip”, “CoordMap”, “CoordMunch”, “CoordPolar”, “CoordQuickmap”, “CoordSf”, “CoordTrans”, “data.frame”, “data.table”, “Date”, “default”, “dm”, “dots”, “element_blank”, “element_grob”, “element_line”, “element_rect”, “element_render”, “element_text”, “environment”, “externalptr”, “FacetWrap”, “factor”, “formula”, “function”, “ggplot”, “ggproto”, “grouped_df”, “labels”, “language”, “Layer”, “list”, “margin”, “matrix”, “mts”, “numeric_version”, “ordered”, “package_version”, “pairlist”, “POSIXct”, “POSIXlt”, “quosure”, “quosures”, “R_system_version”, “rel”, “rowwise_df”, “S4”, “Scale”, “ScalesList”, “simpleUnit”, “tbl_df”, “theme”, “ts”, “uneval”, “vctrs_list_of”, “waiver” and “weakref”.
  • A set of functions prefixed with opts_ can be used to choose various constructors and apply parameters to tweak the output.
  • A special constructor named “next” can be used to fall back on the next .cstr_construct() method. This is useful to explore objects at a level one step lower than the idiomatic constructor.
  • When a corrupted object is encountered the next method is used.
  • Users can extend the package using a set of exported functions prefixed with .cstr_, a vignette describes how to proceed.
  • construct_issues() is used without arguments to check what were the issues encountered with the last reconstructed object, it can also be provided a specific constructive object.
  • construct_diff() highlights the differences in the code used to produce 2 objects.
  • construct_multi() constructs several objects from a named list,
  • construct_dump() is similar to base::dump(), it’s a wrapper around construct_multi() that writes to a file.
  • construct_signature() constructs a function signature such as the one we see in the “usage” section of a function’s help file. outputs the code produced
  • deparse_call() is an alternative to base::deparse() and rlang::expr_deparse() that handles additional corner cases and fails when encountering tokens other than symbols and syntactic literals .

constructive 0.0.1

  • {constructive} produces code that can be used to recreate R objects. In a sense it is similar to base::dput() or base::deparse() but {constructive} strives to use “natural” constructors (factor for factors, as.Date() for dates, data.frame() for data frames etc), in order to get output readable by humans.