CRAN release 0.14.0

• addEventLines() and addLegend() now draw on multiple panels when on is a vector. Thanks to Ethan Smith for the report. (#420)

• Replace SET_TYPEOF() in merge.c because it will error when it tries to convert a REAL to an INTEGER. Thanks to Kurt Hornik for the report! (#419)

• Fix crash when 'j' is not an integer and in [0, 1) (e.g. j = 0.1). Also throw a warning when 'j' is not an integer. (#413, #415)

• Fix plot header when layout() is used to draw multiple plots on a single device. Thanks to Dirk Eddelbuettel for the report and testing! (#412)

• Fix plot legend location when the y-axis is log scale. (#407)

CRAN release 0.13.2

• Print a message when period.apply() is called with FUN = mean because it calculates the mean for each column, not all the data in the subset like it does for all other functions. The message says to use FUN = colMeans for current behavior and FUN = function(x) mean(x) to calculate the mean for all the data. This information is also included in the help files. The option xts.message.period.apply.mean = FALSE suppresses the message. (#124)

• Fix error when print.xts() is called 'quote' or 'right' arguments. (#401)

• Fix addPolygon() so it renders when observation.based = TRUE. (#403)

• Print trailing zeros for index value with fractional seconds, so every index value has the same number of characters. (#404)

• Add ability to log scale the y-axis in plot.xts(). (#103)

• Actually change the underlying index values when 'tclass' is changed from a class with a timezone (e.g. POSIXct) to one without a timezone (e.g. Date). Add a warning when this happens, with a global option to always suppress the warning. (#311)

• Significantly refactor the internals of plot.xts(). (#408)

CRAN Release 0.13.1

• Ignore attribute order in all.equal(). Attribute order shouldn't matter. That can be checked with identical().

• Add instructions to update old objects. Old xts objects do not have tclass and tzone attributes on the index. Add a function to update the object attributes and add a note to the warning to show how to use it. Also, only call tzone() and tclass() once in check.TZ(). Calling these functions multiple times throws multiple warnings for xts objects created before the tclass and tzone were attached to the index instead of the xts object. (#306)

• Return 'POSIXct' if object has no 'tclass'. An empty string is not a valid 'tclass', so it can cause an error.

• Add notes on plot.xts() nomenclature and structure. Also add ASCII art to illustrate definitions and layout. (#103)

• Register missing S3 methods and update signatures. With R-devel (83995-ish), R CMD check notes these S3 methods are not registered. It also notes that the signatures for as.POSIXct.tis() and str.replot_xts() do not match the respective generics. R CMD check also thinks time.frequency() is a S3 method because time() is a generic. The function isn't exported, so renaming won't break any external code. Thanks to Kurt Hornik for the report. Issues with functionality for 'tis' were also identified. I removed 'tis' support entirely because the implementation was not even a bare minimum, and it's not clear it even worked correctly. (#398)

• Format each column individually before printing. The top/bottom rows could have a different number of decimal places and there are often multiple variying spaces between columns. For example:

                              close      volume          ma
    2022-01-03 09:31:00     476.470  803961.000          NA
    2022-01-03 09:32:00     476.700  179476.000          NA
    2022-01-03 09:33:00     476.540  197919.000          NA
                    ...
    2023-03-16 14:52:00    394.6000  46728.0000    392.8636
    2023-03-16 14:53:00    394.6500  64648.0000    392.8755
    2023-03-16 14:54:00    394.6500  69900.0000    392.8873

  There are 4 spaces between the index and the 'close' column, 2 between 'close' and 'volume', and 4 between 'volume' and 'ma'. There should be a consistent number of spaces between the columns. Most other classes of objects print with 1 space between the columns. The top rows have 3 decimals and the bottom rows have 4. These should also be the same. (#321)

• Only convert printed index values to character. Converting the entire index to character is time-consuming for xts objects with many observations. It can take more than a second to print an xts object with 1mm observations.

• Make column names based on number of columns. The original code was a lot more complicated because it tried to account for truncating the number of printed columns. That functionality was removed because of how complicated it was. So now we can simply create printed column names from the number of columns. (#395)

• Fix xts() for zero-row data.frame. The xts() constructor would create an object with a list for coredata when 'x' is a data.frame with no rows. It needs to convert 'x' to a matrix and throw an error if 'x' is a list. (#394)

• Reduce instances when dplyr::lag() warning is shown. The warning was shown whenever it detected dplyr is installed, even if the user wasn't actively using dplyr. That caused an excessive amount of noise when other packages attached xts (e.g. quantmod). Thanks to Duncan Murdoch for the report and suggested fix! (#393)

• Keep colname when only one non-time-based column. The subset x[, -which.col] would return a vector when the data frame has a time-based column and only one additional column. Do not drop dimensions, so 'x' will still be a data.frame in this case. (#391)

• Treat NA the same as NULL for start or end values. NULL represents an undefined index value. NA represents an unknown or missing index value. xts does not allow NA as index values. Subsetting an xts or zoo object by NA returns a zero-length object. So a NA (unknown) index value is essentially the same as an undefined index value. (#383, #345)

• Warn and remove NA when periodicity() called on date-time with NA. Otherwise the uninformative error below will be thrown. (#289)

  Error in try.xts(x, error = "'x' needs to be timeBased or xtsible") :
    'x' needs to be timeBased or xtsible

• Account for TZ when making names for split.xts(). as.yearmon.POSIXct() always sets tz = "GMT" when calling as.POSIXlt(), regardless of the xts' index tzone. That can cause the as.yearmon() results to be different days for GMT and the index's timezone. Use format.POSIXct() for "months" because it checks for a 'tzone' attribute before converting to POSIXlt and calling format.POSIXlt(). The conversion to POSIXlt is important because it checks and uses the 'tzone' attribute before considering the 'tz' argument. So it effectively ignores the tz = "GMT" setting in as.yearmon(). This is also the reason for calling as.POSIXlt() before calling as.yearqtr(). (#392)

CRAN release 0.13.0

New Features

• Added a xts method for na.fill() to significantly increase performance when 'fill' is a scalar. (#259)

• as.xts() will look for a time-based column in a data.frame if it cannot create an index from the row names. (#381)

• Change print() xts method to only show the first and last 'show.rows' rows if number of rows is > 'max.rows'. (#321)

• Made str() output more descriptive for xts objects. It now differentiates between xts objects that are empty, zero-width, or zero-length, and defines each type of object. It also adds column names to the output. (#168, #378)

• Add startup warning that dplyr::lag() breaks method dispatch, which means calls to lag(my_xts) won't work any more.

• Added open-ended time of day subsetting ranges. This allows users to subset by time of day from the start/end of the day without providing the start/end times (00:00:00.000/23:59:59.999). For example:

  x["/T1800"]  # between the start of the day and 5pm
  x["T0500/"]  # between 5am and the end of the day

  Thanks to Chris Katsulis for the suggestion! (#243)

• Updated to.period() to accept custom 'endpoints' via the 'period' argument. Now you can aggregate on something other than the times that 'endpoints()' supports. Thanks to Ethan B. Smith for the suggestion! (#302)

Fixes

• Fixed typo and expand period.apply() documentation. (#205)

  The original description has:

  • "the data from INDEX[k] to INDEX[k+1]"

  But that's not consistent with the code. It should be:

  • "the data from INDEX[k]+1 to INDEX[k+1]"

• Calls to merge.xts() on zero-width objects now match merge.zoo(). Previously, merge.xts() would return empty xts objects if called on two or more zero-width xts objects. merge.zoo() would return a zero-width object with the correct index. (#227, #379)

• Fixed Ops.xts() so it always returned an object with the same class as the first (left-hand side) argument. It previously returned an xts object even if the first argument was a subclass of xts. (#49)

Other

• Migrated unit tests from RUnit to tinytest. Thanks Mark van der Loo!

• Updated the endpoints() documentation to make it clearer that the result is based on the UNIX epoch (midnight 1970, UTC). Thanks to GitHub user Eluvias for the suggestion! (#299)

• Fixed reclass() to ensure it always adds index attributes from the 'match.to' argument. It was not copying tclass, tzone, or tformat from 'match.to' to the result object. (#43)

• Removed an unnecessary check in na.locf() (which is not user-facing). Thanks to GitHub user @cgiachalis for the suggestion! (#307)

• Updated C entry points so they're not able to accidentally be found via dynamic lookup (i.e. .Call("foo", ...)). This makes each call to the C code a few microseconds faster, which is nice. (#260)

• Made merge.xts() results consistent with merge.zoo() for zero-length xts objects with columns. The result of merge.xts() did not include the columns of any objects that had one or more columns, but zero rows. A join should include all the columns of the joined objects, regardless of the number of rows in the object. This is consistent with merge.zoo(). Thanks to Ethan B. Smith for the report and testing! (#222)

CRAN release 0.12.2

Changes to plotting functionality

• You can now omit the data time range from the upper-right portion of a plot by setting main.timespan = FALSE. (#247)

• Fix addEventLines() when plotted objects have a 'yearmon' index. The ISO-8601 range string was not created correctly. Thanks to @paessens for the report. (#353)

• Make ylim robust against numerical precision issues by replacing == with all.equal(). Thanks to @bollard for the report, PR, and a ton of help debugging intermediate solutions! (#368)

• Series added to a panel now extend the panel's y-axis. Previously the y-axis limits were based on the first series' values and not updated when new series were added. So values of the new series did not appear on the plot if they were outside of the original series' min/max. Thanks to Vitalie Spinu for the report and help debugging and testing! (#360)

• All series added to any panel of a plot now update the x-axis of all panels. So the entire plot's x-axis will include every series' time index values within the original plot's time range. This behavior is consistent with chart_Series(). Thanks to Vitalie Spinu for the report and help debugging and testing! (#360, #216)

• All y-values are now plotted for series that have duplicate index values, but different data values. Thanks to Vitalie Spinu for the report and help debugging and testing! (#360)

• Adding a series can now extend the x-axis before/after the plot's existing time index range, so all of the new series' time index values are included in the plot. This is FALSE by default to maintain backward compatibility. Thanks to Vitalie Spinu for the report and help debugging and testing! (#360)

• plot.xts() now has a yaxis.ticks argument to control the number of y-axis grid lines, instead of always drawing 5 grid lines. Thanks to Fredrik Wartenberg for the feature request and patch! (#374)

Other changes

• Ops.xts() no longer changes column names (via make.names()) when the two objects do not have identical indexes. This makes it consistent with Ops.zoo(). (#114)

• Subsetting a zero-length xts object now returns an object with the same storage type as the input. It previously always returned a 'logical' xts object. (#376)

• tclass() and tzone() now return the correct values for zero-length xts objects, instead of the defaults in the .xts() constructor. Thanks to Andre Mikulec for the report and suggested patch! (#255)

• first() and last() now return a zero-length xts object when n = 0. They previously returned the entire object. This is consistent with the default head() and tail() functions, and data.table's first() and last() functions. Thanks to Ethan B. Smith for the report and patch! (#350)

• Subsetting a zero-width xts now returns an object with the same class, tclass, tzone, and xtsAttributes as the input. Thanks to @shikokuchuo for the report! (#359)

• endpoints() now always returns last observation. Thanks to GitHub user Eluvias for the report. (#300)

• Ensure endpoints() errors for every 'on' value when k < 1. It was not throwing an error for k < 1 for on of "years", "quarters", or "months". Thanks to Eluvias for the report. (#301)

• Fix window() for yearmon and yearqtr indexes. In xts < 0.11-0, window.zoo() was dispatched when window() was called on a xts object because there was no window.xts() method. window.zoo() supports additional types of values for the start argument, and possibly other features. So this fixes a breaking change in xts >= 0.11-0. Thanks to @annaymj for the report. (#312)

• Clarify whether axTicksByTime() returns index timestamps or locations (e.g. 1, 2, 3). Thanks to @ggrothendieck for the suggestion and feedback. (#354)

• Fix merge on complex types when 'fill' is needed. merge() would throw an error because it treated 'fill' as double instead of complex. Thanks to @ggrothendieck for the report. (#346)

• Add a message to tell the user how to disable 'xts_check_TZ' warning. Thanks to Jerzy Pawlowski for the nudge. (#113)

• Update rbind() to handle xts objects without dim attribute. rbind() threw an obscure error if one of the xts objects does not have a dim attribute. We can handle this case even though all xts objects should always have a dim attribute. (#361)

• split.xts() now always return a named list, which makes it consistent with split.zoo(). Thanks to Gabor Grothendieck for the report. (#357)

• xts objects with a zero-length POSIXct index now return a zero-length POSIXct vector instead of a zero-length integer vector. Thanks to Jasper Schelfhout for the report and PR! (#363, #364)

• Add suffixes to output of merge.xts(). The suffixes are consistent with merge.default() and not merge.zoo(), because merge.zoo() automatically uses "." as a separator between column names, but the default method doesn't. Thanks to Pierre Lamarche for the nudge. Better late than never? (#38, #371)

CRAN release 0.12.1

• Various function could change the tclass of xts objects. This would happen in calls to reclass(), period.apply(), and for logical operations on POSIXct indexes. Thanks to Tom Andrews for the report and testing, and to Panagiotis Cheilaris for contributing test cases (#322, #323).
• plot.xts() now supports y-axis labels via 'ylab'. Thanks to Jasen Mackie for the suggestion and PR (#333, #334).
• The API header has been updated to fix the signatures of do_merge_xts() and is_xts, which did not return a SEXP as required of functions callable by .Call(). Thanks to Tomas Kalibera for the report (#317), and Dirk Eddelbuettel for the PR (#337). This is a breaking change, but is required to avoid the potential for a segfault.
• Michael Chirico added an internal isUTC() function to recognize many UTC- equivalent time zones (#319).
• first() now operates correctly on non-xts objects when 'n = -1'. Previously it would always return the last two values. Thanks to GitHub user vxg20 for the report (#325).
• The .xts() constructor would create an xts object with row names if 'x' had row names. This shouldn't happen, because xts objects do not have or support row names (#298).
• Claymore Marshall added many examples of time-of-day subsetting to ?subset.xts. He also fixed a bug in time-of-day subsetting where subsetting by hour only returned wrong results (#304, #326, #328).

CRAN release 0.12-0

• All the index-attributes have been removed from the xts object and are now only attached to the index itself (#245). We took great care to maintain backward compatibility, and throw warnings when deprecated functions are called and when index-attributes are found on the xts object. But there still may be some breaking changes lurking in edge cases.

• @SamoPP found one edge case (#297) where an error was thrown when index() was called on an xts object with an index that had no tclass attribute.

• ...which led Joshua to find that the index setting functions did not always copy index attributes (#305).

• Several binary operations (e.g. +, -, !=, <, etc.) on variations of uncommon xts objects with other xts, matrix, or vector objects, could result in malformed xts objects (#295). Some examples of the types of uncommon xts objects: no dim attribute, zero-width, zero-length.

• Calling as.matrix() on an xts object without a dim attribute no longer throws an error (#294).

• merge.xts() now honors check.names = FALSE (#293).

• The possible values for major.ticks, minor.ticks, and grid.ticks.on in the Details section of ?plot.xts have been corrected. Thanks to Harvey Smith (@harvey131) for the report and patch (#291).

• as.zoo.xts() is now only registered for zoo versions prior to 1.8-5. Methods to convert an object to another class should reside in the package that implements the target class. Thanks to Kurt Hornik for the report (#287).

• .parseISO8601() no longer has a potential length-1 logical error. Thanks to Kurt Hornik for the report (#280).

• endpoints() now honors k > 0 when on = "quarters". Thanks to @alkment for the report (#279).

• Performance for the period.XYZ() functions (sum, prod, min, max) is much faster (#278). Thanks to Harvey Smith (@harvey131) for the report and examples.

• merge.xts() now creates shorter column names when passed unnamed objects. This is now consistent with zoo (#248).

• Time-of-day performance is ~200x faster, thanks to StackOverflow user3226167 (#193).

CRAN release 0.11-2

• The to.period() family of functions now use the index timezone when converting intraday index values to daily values (or lower frequency). Thanks to Garrett See and Gabor Grothendieck for the reports (#53, #277).

• Make column names for merge() results with unnamed objects shorter and more like zoo (#248). This also makes na.fill() much faster (#259). BREAKING: This may break existing code for integer unnamed objects.

• Fix subset when index(x) and i contain duplicates. Thanks to Stack Overflow user 'scs' for the report, and Philippe Verspeelt for debugging (#275).

• Fix if-statement in xts constructor that may use a logical with length > 1. Thanks to @HughParsonage for the report and PR (#270, #272).

• Register shift.time.xts() method. Thanks to Philippe Verspeelt for the report and PR (#268, #273).

• Conditionally register S3 methods for as.timeSeries.xts() and as.fts.xts() when their respective packages are available (as requested by CRAN). Note that this means these two functions are no longer exported. This may break some existing code that calls the methods directly, though 'best practice' is to let method dispatch determine which method to invoke.

CRAN release 0.11-1

• Fix regression in .xts() that caused tclass argument/attribute to be incorrectly set to POSIXct regardless of user-provided value. Thanks to @Eluvias for the report and Tom Andrews for the PR (#249, #250).

• Fix performance regression when xts object is subset by a date-time vector. Thanks to Tom Andrews for the report, and the PR to fix the bug in my patch (#251, #263, #264).

• Restore behavior from 0.10-2 so subsetting an empty xts object by a date- time vector returns an empty xts object instead of throwing an error. Thanks to @alkment for the report (#252).

• Add ability for merge.xts() to handle multiple character or complex xts objects. Thanks to Ken Williams for the report (#44).

• Add ability to use "quarters" to specify tick/grid mark locations on plots. This ran but produced an incorrect result in 0.10-2 and threw an error in 0.11-0. Thanks to Marc Weibel for the report (#256).

• Fix illegal read reported by valgrind. Thanks to Tom Andrews for the report and PR (#236, #264).

CRAN release 0.11-0

• Fix make.index.unique() to always return a unique and sorted index. Thanks to Chris Katsulis for the report and example (#241).

• Add window.xts() method and completely refactor the internal binary search function it depends on. Thanks to Corwin Joy for the PR, tests, review, and patience (#100, #240).

• Better axis tick mark locations for plots. Thanks to Dirk Eddelbuettel for the report (#246). Also incorporate axTicksByTime2() into axTicksByTime() to reduce code duplication from the migration of quantmod::chart_Series() to xts::plot.xts() (#74).

• Add details to plot.xts() parameters that are periodicity, now that RStudio has argument completion. Thanks to Evelyn Mitchell for the PR (#154).

• periodicity() now warns instead of errors if the xts object contains less than 2 observations (#230).

• first() and last() now keep dims when they would otherwise be dropped by a regular row subset. This is consistent with head() and tail(). Thanks to Davis Vaughan for the report (#226).

• Fix subset when ISO8601 string is outside the valid range, so it returns no data instead of all rows (#96).

• Avoid partial name matches from parse.side() (inside .parseISO8601()) results that are passed to firstof() and lastof(). Thanks to @gp2x for the report and the patch (#231).

• na.locf.xts() now loops over columns of multivariate objects in C code, instead of in R. This should improve speed and memory performance. Thanks to Chris Katsulis and Tom Andrews for their reports and patches (#232, #233, #234, #235, #237).

• Change plot.xts() default pch = 0 (rectangles) to pch = 1 (circles) so it looks more like base and zoo plots (#203).