This release contains no user-facing changes.
This release contains no user-facing changes.
We embed a development version of libxls (https://github.com/libxls/libxls), which is based on the most recent released version, v1.6.2. The reason for embedding a development version is to ship a version of libxls that incorporates the fix for this CVE (#679):
readxl no longer declares the use of C++11.
readxl should once again compile on Alpine Linux.
Other small readxl-specific patches have been made to the embedded libxls code to comply with CRAN requests, such as avoiding the use of sprintf().
Help files below man/ have been re-generated, so that they give rise to valid HTML5. (This is the impetus for this release, to keep the package safely on CRAN.)
This release is mostly about substantial internal changes that should not be noticeable to most users (but that set the stage for future work):
.xls and .xlsx branchesHowever, there are a few small features / bug fixes:
"Date or Not Date": The classification of number formats as being datetime-ish is more sophisticated and should no longer be so easily fooled by, e.g., colours or currencies. This affects cell and column type guessing, hopefully for the better (#388, #559, @nacnudus, @reviewher).
Cell location is determined more robustly in .xlsx files, guarding against
the idiosyncratic way in which certain 3rd party tools include (or, rather,
do not include) cell location in individual cell nodes (#648, #671).
Warning messages for impossible dates are more specific. Unsupported dates prior to 1900 have their own message now, instead of being lumped in with dates on the non-existent day of February 29, 1900 (#551, #554, @cderv).
readxl is now licensed as MIT (#632).
readxl now states its support for R >= 3.4 explicitly.
Why 3.4? Because the tidyverse policy
is to support the current version, the devel version, and four previous
versions of R.
It was necessary to introduce a minimum R version, in order to state a minimum
version for a package listed in LinkingTo.
readxl embeds libxls v1.6.2 (the previous release embedded v1.5.0). The libxls project is hosted at https://github.com/libxls/libxls and you can learn more about the cumulative changes in its release notes:
readxl has switched from Rcpp to cpp11 and now requires C++11 (#659, @sbearrows).
The minimum version of tibble has been bumped to 2.0.1 (released 2019-01-12), completing the transition to an approach to column name repair used across the tidyverse.
Pragmatic patch release to update some tests in advance of v2.1.0 of the tibble package. That release updates name repair: standard suffix becomes ...j, instead of ..j, partially motivated by user experience in readxl.
readxl embeds libxls v1.5.0. This is the first official release of libxls in several years, although readxl has been tracking the development version in the interim. The libxls project is now officially hosted at https://github.com/libxls/libxls. In particular, libxls v1.5.0 addresses these two CVEs:
readxl exposes the .name_repair argument that is coming to version 2.0.0 of the tibble package. The readxl default is .name_repair = "unique", keeping with the readxl convention to ensure column names are neither missing nor duplicated.
.name_repair is available, defaulting to .name_repair = "unique". Otherwise, the legacy function tibble::repair_names(prefix = "X", sep = "__") is used, replicating the behaviour of readxl v1.1.0.
x."", x, x..1, x..2, x..3X__1, x, x__1read_excel() and friends gain a progress argument that controls a progress spinner (#243, #538).
read_xls() and read_xlsx() pass the trim_ws argument along (#514).
readxl has a new article on reading Excel files with multiple header rows (#486, #492 @apreshill).
xlsx files that do not have a "styles" part can now be read (#505, #506 @jt6)
All paths are passed through normalizePath() (#498, #499, new behaviour for xlsx but not xls) and enc2native() (#370).
readxl is now tested back to R >= 3.1.
Embedded libxls has been updated, using the source in https://github.com/libxls/libxls. readxl's DESCRIPTION now records the SHA associated to the embedded libxls in a Note.
read_excel() and excel_sheets() associate a larger set of file extensions with xlsx and are better able to guess the format of a file with a nonstandard or missing extension. This is about deciding whether to treat a file as xls or xlsx. (#342, #411, #457)
excel_format() is the newly-exported format-guessing function.format_from_ext() is a low-level helper, also exported, that only consults file extension. In addition to the obvious interpretation of .xls and .xlsx, the extensions .xlsm, .xltx, and .xltm are now associated with xlsx.format_from_signature() is a low-level helper, also exported, that consults the file's signature (a.k.a. magic number). It's handy for files that lack an extension.Embedded libxls has been updated to address security vulnerabilitities identified in late 2017 (#441, #442).
xlsx structured as a "minimal conformant SpreadsheetML package" can be read. Most obvious feature of such sheets is the lack of an xl/ directory in the unzipped form. (xlsx, #435, #437)
Reading xls sheet with exactly 65,536 rows no longer enters an infinite loop. (xls, #373, #416, #432 @vkapartzianis)
Doubles, including datetimes, coerced to character from xls now have much higher precision, comparable to the xlsx behaviour. (xls, #430, #431)
Integer-y numbers larger than 2^31 are coerced properly to string (xls, #346)
Shared strings are only compared to NA strings after lookup, never on the basis of their index. (xlsx, #401)
Better checks and messaging around nonexistent files. (#392)
Add $(C_VISIBILITY) to compiler flags to hide internal symbols from the dll. (#385 @jeroen)
Numeric data in a logical column now coerces properly to logical. (xlsx, #385 @nacnudus)
range is a new argument for reading a rectangular range, possibly open. (#314, #8)
n_max is a new argument that limits the number of data rows read. (#306, #281)
Empty cells, rows, columns (xlsx #248 and #240, xls #271): Cells with no content are no longer loaded, even if they appear in the file. Affects cells that have no data but that carry explicit formatting, detectable in Excel as seemingly empty cells with a format other than "General". Such cells may still exist in the returned tibble, with value NA, depending on the sheet geometry.
NA.User-supplied col_names are processed relative to user-supplied col_types, if given. Specifically, col_names is considered valid if it has the same length as col_types, before or after removing skipped columns. (#81, #261)
"list" is a new accepted value for col_types. Loads data as a list of length-1 vectors, that are typed using the logic from col_types = NULL, but on a cell-by-cell basis (#262 @gergness).
"logical" is a new accepted value for col_types. When col_types = NULL, it is the guessed type for cells Excel advertises as Boolean. When a column has no data, it is now filled with logical NA. (#277, #270)
"guess" is a new accepted value for col_types. Allows the user to specify some column types, while allowing others to be guessed (#286)
A user-specified col_types of length one will be replicated to have length equal to the number of columns. (#127, #114, #261)
"blank" has been deprecated in favor of the more descriptive and readr-compatible "skip", which is now the preferred way to request that a column be skipped. (#260, #193, #261)
guess_max is a new argument that lets user adjust the number of rows used to guess column types. (#223, #257 @tklebel and @jennybc)
trim_ws is a new argument to remove leading and trailing whitespace. It defaults to TRUE. (#326, #211)
na can now hold multiple NA values, e.g., read_excel("missing-values.xls", na = c("NA", "1")). (#13, #56, @jmarshallnz)
Coercions and cell data:
NA instead of their integer representation. Throws warning. (#277, #263)read_excel() attempts to coerce the string to numeric and falls back to NA if unsuccessful. Throws warning. (#277, #217, #106)NA (instead of the string "error"). (#277, #62)"Unknown type: 517". (#274, #259)Many 3rd party tools write xls and xlsx that comply with the spec, but that are quite different from files produced by Excel.
Namespace prefixes are now stripped from element names and attributes when parsing XML from xlsx. Workaround for the creative approach taken in some other s/w, coupled with rapidxml's lack of namespace support. (#295, #268, #202, #80)
Excel mixes 0- and 1-indexing in reported row and column dimensions for xls and libxls expects that. Other s/w may index from 0 for both, preventing libxls from reading the last column. Patched to restore access to those cells. (#273, #180, #152, #99)
More robust logic for sheet lookup in xlsx. Improves compatibility with xlsx written by a variety of tools and/or xlsx containing chartsheets. (#233, #104, #200, #168, #116, @jimhester and @jennybc)
The numFmtId attribute is no longer accessed when it does not exist (xlsx written by https://www.epplussoftware.com). (#191, #229)
Location is inferred for cells that do not declare their location (xlsx written by JMP). (#240, #163, #102)
read_xls() and read_xlsx() are now exposed, such that files without an .xls or .xlsx extension can be read. (#85, @jirkalewandowski)
The Lotus 1-2-3 leap year bug is now accounted for, i.e. date-times prior to March 1, 1900 import correctly. Date-times on the non-existent leap day February 29, 1900 import as NA and throw a warning. (#264, #148, #292)
The tibble package is now imported (#175, @krlmlr) and tibble::repair_names() is used to prevent empty, NA, or duplicated column names. (#216, #208, #199 #182, #53, #247).
Default column names for xlsx now start with X__1 instead of X__0. (#98, @zeehio, @krlmlr)
Fix compilation warnings/failures (FreeBSD 10.3 #221, gcc 4.9.3 #124) and/or problems reading xls (CentOS 6.6 #189). (#244, #245, #246 @jeroen)
Unwanted printed output (e.g., DEFINEDNAME: 21 00 00 ...) is suppressed when reading xls that contains a defined range. (#82, #188, @PedramNavid)
_x005F_). (#51, @jmarshallnz)