# haven 2.5.4

* Fix for upcoming R-devel change.

# haven 2.5.3

* Fix for upcoming R-devel change.

# haven 2.5.2

* Updated to ReadStat 1.1.9.

  * Fix various SAS catalog file reading bugs (#529, #653, #680, #696, #705).
  * Increase maximum SAS page file size to 16MiB (#697).
  * Ignore invalid SAV timestamp strings (#683).
  * Fix compiler warnings (#707).

* The experimental `write_sas()` function has been deprecated (#224). The
  sas7bdat file format is complex and undocumented, and as such writing SAS
  files is not officially supported by ReadStat. `write_xpt()` should be used
  instead - it produces files in the SAS transport format, which has
  limitations but will be reliably read by SAS.

* `write_*()` functions gain a new `adjust_tz` argument to allow more control
  over time zone conversion for date-time variables (#702). Thanks to @jmobrien
  for the detailed issue and feedback.
  
  Stata, SPSS and SAS do not have a concept of time zone. Since haven 2.4.0
  date-time values in non-UTC time zones are implicitly converted when writing
  to ensure the time displayed in Stata/SPSS/SAS will match the time displayed
  to the user in R (see #555). This is the behaviour when `adjust_tz = TRUE`
  (the default). Although this is in line with general user expectations it can
  cause issues when the time zone is important, for e.g. when looking at
  differences between time points, since the underlying numeric data is changed
  to preserve the displayed time. Use `adjust_tz = FALSE` to write the time as
  the corresponding UTC value, which will appear different to the user but
  preserves the underlying numeric data.

* `write_*()` functions previously returned the data frame with minor
  alterations made to date-time variables. These functions now invisibly return
  the original input data frame unchanged (@jmobrien, #702).

* Fix bug in string variable width calculation that treated `NA` values as width
  2. `NA` values are now treated as blanks for width calculations (#699).

# haven 2.5.1

* All `labelled()` vectors now have left-aligned column headers when printing
  in tibbles for better alignment with labels (#676).
  
* `write_*()` now accept functions as well as strings in the 
  `.name_repair` argument in line with the documentation. Previously they only 
  supported string values (#684).

* `write_sav()` variable name validation no longer treats all non-ASCII 
  characters as invalid (#689).

# haven 2.5.0

## New author

* @gorcha is now a haven author in recognition of his significant and sustained
  contributions.

## File writing improvements

* All `write_` functions can now write custom variable widths by setting the
  `width` attribute (#650).

* When writing files, the minimum width for character variables is now 1. This
  fixes issues with statistical software reading blank character variables with
  width 0 (#650).

* `write_dta()` now uses strL when strings are too long to be stored in an str#
  variable (#437). strL is used when strings are longer than 2045 characters by
  default, which matches Stata's behaviour, but this can be reduced with the
  `strl_threshold` argument.

* `write_xpt()` can now write dataset labels with the `label` argument,  which
  defaults to the `label` attribute of the input data frame, if present (#562).
  
* `write_sav()` now checks for case-insensitive duplicate variable names
  (@juansebastianl, #641) and verifies that variable names are valid SPSS
  variables.

* The `compress` argument for `write_sav()` now supports all 3 SPSS compression
  modes specified as a character string - "byte", "none" and "zsav" (#614).
  `TRUE` and `FALSE` can be used for backwards compatibility, and correspond to
  the "zsav" and "none" options respectively.

* `write_sav()` successfully writes user missing values and ranges for
  `labelled()` integer vectors (#596).

* POSIXct and POSIXlt values with no time component (e.g. "2010-01-01") were
  being converted to `NA` when attempting to convert the output timezone to UTC.
  These now output successfully (#634).

* Fix bug in output timezone conversion that was causing variable labels and
  other variable attributes to disappear (#624).

## Other improvements and fixes

* Updated to ReadStat 1.1.8 RC.

  * Fix bug when writing formats to XPT files (#650).
  * Fix off by one error in indexing for strL variables (#437).

* `labelled()` vectors now throw a warning when combining two vectors with
  conflicting labels (#667).

* `zap_labels()` gains a `user_na` argument to control whether user-defined
  missing values are converted to `NA` or left as is (#638).

* vctrs casting and coercion generics now do less work when working with two
  identical `labelled()` vectors. This significantly improves performance when
  working with `labelled()` vectors in grouped data frames (#658).

* Errors and warnings now use `cli_abort()` and `cli_warning()` (#661).

## Dependency changes

* R 3.4 is now the minimum supported version, in line with [tidyverse  policy](https://www.tidyverse.org/blog/2019/04/r-version-support/).

* cli >= 3.0.0 has been added to Imports to support new error messaging.
  
* lifecycle has been added to Imports, and is now used to manage deprecations.

# haven 2.4.3

* Fix build failure on Solaris.

# haven 2.4.2

* Updated to ReadStat 1.1.7 RC (#620).

* `read_dta()` no longer crashes if it sees StrL variables with missing values
  (@gorcha, #594, #600, #608).

* `write_dta()` now correctly handles "labelled"-class numeric (double) variables 
   that don't have value labels (@jmobrien, #606, #609).

* `write_dta()` now allows variable names up to 32 characters (@sbae, #605).

* Can now correctly combine `labelled_spss()` with identical labels 
  (@gorcha, #599).

# haven 2.4.1

* Fix buglet when combining `labelled()` with identical labels.

# haven 2.4.0

## New features

* `labelled_spss()` gains full vctrs support thanks to the hard work of @gorcha
  (#527, #534, #538, #557). This means that they should now work seamlessly
  in dplyr 1.0.0, tidyr 1.0.0 and other packages that use vctrs. 

* `labelled()` vectors are more permissive when concatenating; output labels 
  will be a combination of the left-hand and the right-hand side, preferring
  values assigned to the left-hand side (#543).

* Date-times are no longer forced to UTC, but instead converted to the 
  equivalent UTC (#555). This should ensure that you see the same date-time 
  in R and in Stata/SPSS/SAS.
   
## Minor improvements and bug fixes

* Updated to ReadStat 1.1.5. Most importantly this includes support for
  SAS binary compression.

* `as_factor(levels = "values")` preserves values of unlabelled elements (#570).

* `labelled_spss()` is a little stricter: it prevents `na_range` and `na_value`
  from containing missing values, and ensures that `na_range` is in the correct
  order (#574).

* `read_spss()` now reads NA values and ranges of character variables (#409).

* `write_dta()` now correctly writes tagged NAs (including tagged NAs in
  labels) (#583) and once again validates length of variables names (#485).

* `write_*()` now validate file and variable metadata with ReadStat. This 
  should prevent many invalid files from being written (#408). Additionally,
  validation failures now provide more details about the source of the problem 
  (e.g. the column name of the problem) (#463).

* `write_sav(compress = FALSE)` now uses SPSS bytecode compression instead of 
  the rarely-used uncompressed mode. `compress = TRUE` continues to use the 
  newer (and not universally supported, but more compact) zlib format 
  (@oliverbock, #544).

# haven 2.3.1

* Add missing methods so `median()`, `quantile()` and `summary()` work
  once more (#520).
  
* Add missing cast methods (#522).

# haven 2.3.0

* `labelled()` gains the necessary support to work seemlessly in dplyr 1.0.0,
  tidyr 1.0.0, and other packages that use vctrs (@mikmart, #496).

* `labelled()` vectors now explicitly inherit from the corresponding base
  types (e.g. integer, double, or character) (#509).

* ReadStat update, including `read_sas()` supports for "any" encoding (#482),
  and fixes for compiler warnings.

# haven 2.2.0

## Partial reading

Thanks to the hard work of @mikmart, all `read_*()` functions gain three new arguments that allow you to read in only part of a large file:

* `col_select`: selects columns to read with a tidyselect interface (#248).
* `skip`: skips rows before reading data (#370).
* `n_max`: limits the number of rows to read.

This also brings with it a deprecation: `cols_only` in `read_sas()` has been deprecated in favour of the new `col_select` argument.

## Minor improvements and bug fixes

* `as_factor()` allows non-unique labels when `levels = "label"`. This fixes 
  a particularly annoying printing bug (#424, @gergness)

* `read_sas()` now supports (IS|E|B)8601(DT|DA|TM) date/time formats (@mikmart).

* All `write_` functions gain a `.name_repair` argument that controls
  what happens when the input dataset has repeated column names (#436).

* All `write_` functions can now write labelled vectors with `NULL` labels 
  (#442).

* `write_dta()` can now write dataset labels with the `label` argument, 
  which defaults to the `label` attribute of the input data frame, if present
  (@gorcha, #449).

* `write_dta()` works better with Stata 15, thanks to updated ReadStat (#461)

# haven 2.1.1

* Fixes for R CMD check

# haven 2.1.0

## Improved labelling

`labelled` objects get pretty printing that shows the labels and NA values when inside of a `tbl_df`. Turn this behaviour off with behavior using `option(haven.show_pillar_labels = FALSE)` (#340, @gergness).

`labelled()` and `labelled_spss()` now allow `NULL` labels. This makes both classes more flexible, allowing you to use them for their other attributes (#219).

`labelled()` tests that value labels are unique (@larmarange, #364)

## Minor improvements and bug fixes

*   `as_factor()`:

    * Is faster when input doesn't contain any missing values (@hughparsonage).
    * Added `labelled` method for backward compatbility (#414).
    * `data.frame` method now correctly passes `...` along (#407, @zkamvar).

*   `write_dta()` now checks that the labelled values are integers, not the 
    values themselves (#401).

*   Updated to latest ReadStat from @evanmiller:

    * `read_por()` can now read files from SPSS 25 (#412)
    * `read_por()` now uses base-30 instead of base-10 for the exponent (#413)
    * `read_sas()` can read zero column file (#420)
    * `read_sav()` reads long strings (#381)
    * `read_sav()` has greater memory limit allowing it to read more labels (#418)
    * `read_spss()` reads long variable labels (#422)
    * `write_sav()` no longer creates incorrect column names when >10k columns (#410)
    * `write_sav()` no longer crashes when writing long label names (#395)
  
# haven 2.0.0

## BREAKING CHANGES

*   `labelled()` and `labelled_spss()` now produce objects with class
    "haven_labelled" and "haven_labelled_spss". Previously, the "labelled"
    class name clashed with the labelled class defined by Hmisc (#329).
    
    Unfortunately I couldn't come up with a way to fix this problem except
    to change the class name; it seems reasonable that haven should be the one 
    to change names given that Hmisc has been around much longer. This
    will require some changes to packages that use haven, but shouldn't
    affect user code.

## Minor improvements

* `labelled()` and `labelled_spss()` now support adding the `label`
  attribute to the resulting object. The `label` is a short,
  human-readable description of the object, and is now also used
  when printing, and can be easily removed using the new `zap_label()`
  function. (#362, @huftis)
  
  Previously, the `label` attribute was supported both when reading
  and writing SPSS files, but it was not possible to actually create
  objects in R having the `label` attribute using the constructors
  `labelled()` or `labelled_spss()`.

# haven 1.1.2

* haven can read and write non-ASCII paths in R 3.5 (#371).

* `labelled_spss` objects preserve their attributes when subsetted 
  (#360, @gergness).

* `read_sav()` gains an `encoding` argument to override the encoding stored in 
  the file (#305). `read_sav()` can now read `.zsav` files (#338). 
  
* `write_*()` functions now invisibly return the input data frame 
  (as documented) (#349, @austensen).

* `write_dta()` allows non-ASCII variable labels for version 14 and above 
  (#383). It also uses a less strict check for integers so that a 
  labelled double containing only integer values can written (#343).

* `write_sav()` produces `.zsav` files when `compress = TRUE` (#338).

* `write_xpt()` can now set the "member" name, which defaults to the file name
  san extension (#328).

* Update to latest readstat.

  * Fixes out of memory error (#342)
  * Now supports reading and writing stata 15 files (#339)
  * Negative integer labelled values were tagged as missing (#367)

* Fix for when `as_factor()` with option `levels="labels"` is used on tagged NAs
  (#340, @gergness)

# haven 1.1.1

* Update to latest readstat. Includes:

  * SPSS: empty charater columns now read as character (#311)
  * SPSS: now write long strings (#266)
  * Stata: reorder labelled vectors on write (#327)
  * State: `encoding` now affects value labels (#325)
  * SAS: can now write wide/long rows (#272, #335).
  * SAS: can now handle Windows Vietnamese character set (#336)

* `read_por()` and `read_xpt()` now correctly preserve attributes if
  output needs to be reallocated (which is typical behaviour) (#313)

* `read_sas()` recognises date/times format with trailing separator and width
  specifications (#324)
  
* `read_sas()` gains a `catalog_encoding` argument so you can independently
  specify encoding of data and catalog (#312)

* `write_*()` correctly measures lengths of non-ASCII labels (#258): this
  fixes the cryptic error "A provided string value was longer than the 
  available storage size of the specified column."

* `write_dta()` now checks for bad labels in all columns, not just the first
  (#326).

* `write_sav()` no longer fails on empty factors or factors with an `NA`
  level (#301) and writes out more metadata for `labelled_spss` vectors 
  (#334).
 
# haven 1.1.0

* Update to latest readstat. Includes:

  * SAS: support Win baltic code page (#231)
  * SAS: better error messages instead of crashes (#234, #270)
  * SAS: fix "unable to read error" (#271)
  * SPSS: support uppercase time stamps (#230)
  * SPSS: fixes for 252-255 byte strings (#226)
  * SPSS: fixes for 0 byte strings (#245)

* Share `as_factor()` with forcats package (#256)

* `read_sav()` once again correctly returns system defined missings 
  as `NA` (rather than `NaN`) (#223). `read_sav()` and `write_sav()` preserve 
  SPSS's display widths (@ecortens).

* `read_sas()` gains experimental `cols_only` argument to only read in 
  specified columns (#248).

* tibbles are created with `tibble::as_tibble()`, rather than by "hand" (#229).

* `write_sav()` checks that factors don't have levels with >120 
  characters (#262)

* `write_dta()` no longer checks that all value labels are at most 32 
  characters (since this is not a restriction of dta files) (#239).

* All write methds now check that you're trying to write a data frame (#287).

* Add support for reading (`read_xpt()`) and writing (`write_xpt()`) SAS 
  transport files.

* `write_*` functions turn ordered factors into labelled vectors (#285)

# haven 1.0.0

* The ReadStat library is stored in a subdirectory of `src` (#209, @krlmlr).

* Import tibble so that tibbles are printed consistently (#154, @krlmlr).

* Update to latest ReadStat (#65). Includes: 

    * Support for binary (aka Ross) compression for SAS (#31).
    * Support extended ASCII encoding for Stata (#71).
    * Support for Stata 14 files (#75, #212).
    * Support for SPSS value labels with more than 8 characters (#157).
    * More likely to get an error when attempting to create an invalid
      output file (#171).

* Added support for reading and writing variable formats. Similarly to 
  to variable labels, formats are stored as an attribute on the vector.
  Use `zap_formats()` if you want to remove these attributes.
  (@gorcha, #119, #123).
  
* Added support for reading file "label" and "notes". These are not currently
  printed, but are stored in the attributes if you need to access them (#186).
  
* Added support for "tagged" missing values (in Stata these are called 
  "extended" and in SAS these are called "special") which carry an extra
  byte of information: a character label from "a" to "z". The downside of
  this change is that all integer columns are now converted to doubles, 
  to support the encoding of the tag in the payload of a NaN.
  
* New `labelled_spss()` is a subclass of `labelled()` that can model
  user missing values from SPSS. These can either be a set of distinct
  values, or for numeric vectors, a range. `zap_labels()` strips labels,
  and replaces user-defined missing values with `NA`. New `zap_missing()`
  just replaces user-defined missing values with `NA`. 
  
    `labelled_spss()` is potentially dangerous to work with in R because
    base functions don't know about `labelled_spss()` functions so will 
    return the wrong result in the presence of user-defined missing values.
    For this reason, they will only be created by `read_spss()` when
    `user_na = TRUE` (normally user-defined missings are converted to
    NA). 

* `as_factor()` no longer drops the `label` attribute (variable label) when
   used (#177, @itsdalmo).

* Using `as_factor()` with `levels = "default` or `levels = "both"` preserves
  unused labels (implicit missing) when converting (#172, @itsdalmo). Labels 
  (and the resulting factor levels) are always sorted by values.

* `as_factor()` gains a new `levels = "default"` mechanism. This uses the
  labels where present, and otherwise uses the labels. This is now the
  default, as it seems to map better to the semantics of labelled values
  in other statistical packages (#81). You can also use `levels = "both"`
  to combine the value and the label into a single string (#82). It also
  gains a method for data frames, so you can easily convert every labelled
  column to a factor in one function call.

* New `vignette("semantics", package = "haven")` discusses the semantics
  of missing values and labelling in SAS, SPSS, and Stata, and how they
  are translated into R.

* Support for `hms()` has been moved into the hms package (#162).
  Time varibles now have class `c("hms", "difftime")` and a `units` attribute
  with value "secs" (#162).

* `labelled()` is less strict with its checks: you can mix double and integer
  value and labels (#86, #110, @lionel-), and `is.labelled()` is now exported 
  (#124). Putting a labelled vector in a data frame now generates the correct
  column name (#193).

* `read_dta()` now recognises "%d" and custom date types (#80, #130).
  It also gains an encoding parameter which you can use to override
  the default encoding. This is particularly useful for Stata 13 and below
  which did not store the encoding used in the file (#163). 

* `read_por()` now actually works (#35).

* `read_sav()` now correctly recognises EDATE and JDATE formats as dates (#72).
  Variables with format DATE, ADATE, EDATE, JDATE or SDATE are imported as 
  `Date` variables instead of `POSIXct`. You can now set `user_na = TRUE` to 
  preserve user defined missing values: they will be given class 
  `labelled_spss`.

* `read_dta()`, `read_sas()`, and `read_sav()` have a better test for missing
  string values (#79). They can all read from connections and compressed files
  (@lionel-, #109)

* `read_sas()` gains an encoding parameter to overide the encoding stored
  in the file if it is incorrect (#176). It gets better argument names (#214).

* Added `type_sum()` method for labelled objects so they print nicely in
  tibbles.

* `write_dta()` now verifies that variable names are valid Stata variables
  (#132), and throws an error if you attempt to save a labelled vector that
  is not an integer (#144). You can choose which `version` of Stata's file 
  format to output (#217).

* New `write_sas()` allows you to write data frames out to `sas7bdat` 
  files. This is still somewhat experimental.

* `write_sav()` writes hms variables to SPSS time variables, and the
  "measure" type is set for each variable (#133).

* `write_dta()` and `write_sav()` support writing date and date/times
  (#25, #139, #145). Labelled values are always converted to UTF-8 before 
  being written out (#87). Infinite values are now converted to missing values 
  since SPSS and Stata don't support them (#149). Both use a better test 
  for missing values (#70).
  
* `zap_labels()` has been completely overhauled. It now works 
  (@markriseley, #69), and only drops label attributes; it no longer replaces
  labelled values with `NA`s. It also gains a data frame method that zaps
  the labels from every column.
  
* `print.labelled()` and `print.labelled_spss()` now display the type.

# haven 0.2.0

* fixed a bug in `as_factor.labelled`, which generated <NA>'s and wrong 
  labels for integer labels.

* `zap_labels()` now leaves unlabelled vectors unchanged, making it easier
  to apply to all columns.

* `write_dta()` and `write_sav()` take more care to always write output as
  UTF-8 (#36)

* `write_dta()` and `write_sav()` won't crash if you give them invalid paths,
  and you can now use `~` to refer to your home directory (#37).

* Byte variables are now correctly read into integers (not strings, #45), 
  and missing values are captured correctly (#43).

* Added `read_stata()` as alias to `read_dta()` (#52).

* `read_spss()` uses extension to automatically choose between `read_sav()`
  and `read_por()` (#53)

* Updates from ReadStat. Including fixes for various parsing bugs, more 
  encodings, and better support for large files.

* hms objects deal better with missings when printing.

* Fixed bug causing labels for numeric variables to be read in as
  integers and associated error: ``Error: `x` and `labels` must be same type``

# haven 0.1.1

* Fixed memory initialisation problems found by valgrind.