Tolerance is also taken into account when displaying differences (#173).
NA_real_ and NaN are only treated as non-equal when tolerance is
non-null. That means that testthat::expect_equal(NaN, NA_real_) will
pass but testthat::expect_identical(NaN, NA_real_) will fail (#174).
You can opt-out of quoting strings with quote_strings = FALSE (#145).
Improvements to missing value handling:
NA_character_ and "NA" are no longer treated as equal (#162).
NA_real_ and NaN are no longer treated as equal (@sorhawell, #150).
Leading and trailing NAs are no longer omitted from output when the
lengths of x and y are unequal (#109).
The balanced attribute used by some POSIXlt objects in R 4.3 and greater
is now ignored (#160).
3d (and greater) numeric arrays no longer cause an error (#148).
Support for complex numbers is improved (#146).
ignore_attr = "class" now works for more types of input (#143).
Atomic S3 classes with format methods now use those methods when displaying comparisons (#98). If the printed representation is the same, they fallback to displaying the underlying data.
Rowwise data frame comparisons are now much much faster (#116),
and respect the max_diffs argument (@krlmlr, #110).
Unnamed environments now compare by value, not by reference (i.e. if two environments contain the same values, they compare the same, even if they're different environments) (#127). Environments that contain self-references are handled correctly (#117). Differences between pairs of environments are only ever reported once.
In the unlikely event that you have bare CHARSXP objects, waldo now handles them (#121).
S4 objects are labelled with their class, not all superclasses (#125).
compare_proxy() ignores the "index" attribute for data tables
(@krlmlr, #107), and works again for RProtoBuf objects
(@MichaelChirico, #119)
Infinite values can be compared with a tolerance (@dmurdoch, #122).
compare()ing data frames now works independently of option(max.print)
(#105).
Fixed regression when comparing vectors with missing values (#102).
compare() is now considerably faster when comparing complex objects that
don't have any differences (thanks to strategic use of identical()) (#86).
compare() gains two improvements to low-level diffs:
Structurally identical data frames (#78) and numeric matrices (#76) gain a row-by-row diff that makes it easier to see where exactly values differ.
An element-by-element diff will be automatically used if it's shorter than the "smart" diff. This improves diff quality when comparing two vectors that aren't really related (#68).
compare() gains a list_as_map argument thanks to an idea from @dmurdoch.
It allows you to compare the behaviour of two lists when they are used to
connect names to values (i.e. the list is operating as a map or dictionary).
It removes NULLs and sorts named components (#72).
The objects involved in compare() (as opposed to the caller of compare())
gained much greater ability to control the comparison.
Objects can now contain a waldo_opts attribute, a list with the same
names and valid values as the arguments to compare(), which overrides
the default comparisons (@dmurdoch).
compare_proxy() is now called earlier (before type comparison) making
it more flexible (#65).
compare_proxy() gains a second argument, path, used to report how the
proxy changed the object. This makes it easier to see when and how a proxy
is used (#73).
Proxies now exist for comparing RProtoBuf objects, converting them to proto text format (#82, @michaelquinn32).
Comparing a list with symbol to a list without that element no longer errors (@mgirlich, #79).
Additional arguments to compare() generate a more informative warning
(#58).
Numbers use a better algorithm for picking the number of decimal places to show (#63).
ASTs with identical deparsed strings now show exactly how the AST differs.
Source references are now more comprehensively stripped using
rlang::zap_srcrefs()
S3 objects now show the base type, and no longer fails when the types are incompatible.
compare() gains a new max_diffs argument that allows you to control
the maximum number of differences shown. Set max_diffs = Inf to
see all differences (#49)
Logical vectors fall back to element-by-element comparison in more cases (#51).
Long-form diff no longer confuses additions and deletions (#52, @krlmlr).
Handle S4 objects that have attributes that are not slots.
Additions are now coloured blue and deletions yellow (instead of the opposite).
compare() now labels output as old and new, since that's the most
natural way to use it.
compare() can selectively ignore attributes by providing vector to
ignore_attr (#45).
print() method gets n argument to allow explicitly specifying number of
differences to show (@mnazarov).
Improvements to comparison display:
Zero length vectors compare robustly (#39)
Line-by-line comparisons show modifications as deletion then addition, rather than addition then deletion (#44).
Differences between numeric vectors are more robust, particularly in the presence of missing values (#43). The number of digits selected has also been slightly improved so that you're more likely to get exactly the number of digits needed.
All objects: class (#26) and names (#31) are ignored when ignoring attributes.
Numeric and logical vectors: clearer display of differences. Numbers are right-aligned, and we show the numbers not the differences.
Character vectors: a trailing newline is no longer ignored (#37).
Lists: all elements of the unnamed lists are compared, not just the last! (#32)
Lists: unclassed prior to comparison (#21).
Data frames: The internal representation of row names is no longer used;
instead we use the same result of rownames() (#23).
Environments: New ignore_formula_env and ignore_function_env arguments to
ignore formula and function environments for compatibility with all.equal()
(#24).
Expression objects: can now be compared (#29).
Calls: srcrefs and attributes are ignored.
compare_proxy() is now exported so that you can provide methods if your
objects need special handling (particularly needed for objects that contain
external pointers) (#22).
Fixed a partial argument name in as.list().
NEWS.md file to track changes to the package.