Functions to Manipulate Connections (Files, URLs, ...)

Description

Functions to create, open and close connections, i.e., “generalized files”, such as possibly compressed files, URLs, pipes, etc.

Usage

file(description = "", open = "", blocking = TRUE,
     encoding = getOption("encoding"), raw = FALSE,
     method = getOption("url.method", "default"))

url(description, open = "", blocking = TRUE,
    encoding = getOption("encoding"),
    method = getOption("url.method", "default"),
    headers = NULL)

gzfile(description, open = "", encoding = getOption("encoding"),
       compression = 6)

bzfile(description, open = "", encoding = getOption("encoding"),
       compression = 9)

xzfile(description, open = "", encoding = getOption("encoding"),
       compression = 6)

zstdfile(description, open = "", encoding = getOption("encoding"),
       compression = 9)

unz(description, filename, open = "", encoding = getOption("encoding"))

pipe(description, open = "", encoding = getOption("encoding"))

fifo(description, open = "", blocking = FALSE,
     encoding = getOption("encoding"))

socketConnection(host = "localhost", port, server = FALSE,
                 blocking = FALSE, open = "a+",
                 encoding = getOption("encoding"),
                 timeout = getOption("timeout"),
                 options = getOption("socketOptions"))

serverSocket(port)

socketAccept(socket, blocking = FALSE, open = "a+",
             encoding = getOption("encoding"),
             timeout = getOption("timeout"),
             options = getOption("socketOptions"))

open(con, ...)
## S3 method for class 'connection'
open(con, open = "r", blocking = TRUE, ...)

close(con, ...)
## S3 method for class 'connection'
close(con, type = "rw", ...)

flush(con)

isOpen(con, rw = "")
isIncomplete(con)

socketTimeout(socket, timeout = -1)

Arguments

Details

The first eleven functions create connections. By default the connection is not opened (except for a socket connection created by socketConnection or socketAccept and for server socket connection created by serverSocket), but may be opened by setting a non-empty value of argument open.

For file the description is a path to the file to be opened (when tilde expansion is done) or a complete URL (when it is the same as calling url), or "" (the default) or "clipboard" (see the ‘Clipboard’ section). Use "stdin" to refer to the C-level ‘standard input’ of the process (which need not be connected to anything in a console or embedded version of R, and is not in RGui on Windows). See also stdin() for the subtly different R-level concept of stdin. See nullfile() for a platform-independent way to get filename of the null device.

For url the description is a complete URL including scheme (such as ‘⁠http://⁠’, ‘⁠https://⁠’, ‘⁠ftp://⁠’ or ‘⁠file://⁠’). Method "internal" is that available since connections were introduced but now mainly defunct. Method "wininet" is only available on Windows (it uses the WinINet functions of that OS) and method "libcurl" (using the library of that name: https://curl.se/libcurl/) is nowadays required but was optional on Windows before R 4.2.0. Method "default" currently uses method "internal" for ‘⁠file://⁠’ URLs and "libcurl" for all others. Which methods support which schemes has varied by R version – currently "internal" supports only ‘⁠file://⁠’; "wininet" supports ‘⁠file://⁠’, ‘⁠http://⁠’ and ‘⁠https://⁠’. Proxies and user credentials can be specified: see download.file.

For gzfile the description is the path to a file compressed by gzip: it can also open for reading uncompressed files and those compressed by bzip2, xz, lzma or (if R was built to support it) zstd.

For bzfile the description is the path to a file compressed by bzip2.

For xzfile the description is the path to a file compressed by xz (https://en.wikipedia.org/wiki/Xz) or (for reading only) lzma (https://en.wikipedia.org/wiki/LZMA).

For zstdfile the description is the path to a file compressed by zstd (https://en.wikipedia.org/wiki/Zstd): support is optional when R is built.

unz reads (only) single files within zip files, in binary mode. The description is the full path to the zip file, with ‘.zip’ extension if required.

For pipe the description is the command line to be piped to or from. This is run in a shell, on Windows that specified by the COMSPEC environment variable.

For fifo the description is the path of the fifo. (Support for fifo connections is optional but they are available on most Unix platforms and on Windows.)

The intention is that file and gzfile can be used generally for text input (from files, ‘⁠http://⁠’ and ‘⁠https://⁠’ URLs) and binary input respectively.

open, close and seek are generic functions: the following applies to the methods relevant to connections.

open opens a connection. In general functions using connections will open them if they are not open, but then close them again, so to leave a connection open call open explicitly.

close closes and destroys a connection. This will happen automatically in due course (with a warning) if there is no longer an R object referring to the connection.

flush flushes the output stream of a connection open for write/append (where implemented, currently for file and clipboard connections, stdout and stderr).

If for a file or (on most platforms) a fifo connection the description is "", the file/fifo is immediately opened (in "w+" mode unless open = "w+b" is specified) and unlinked from the file system. This provides a temporary file/fifo to write to and then read from.

socketConnection(server=TRUE) creates a new temporary server socket listening on the given port. As soon as a new socket connection is accepted on that port, the server socket is automatically closed. serverSocket creates a listening server socket which can be used for accepting multiple socket connections by socketAccept. To stop listening for new connections, a server socket needs to be closed explicitly by close.

socketConnection and socketAccept support setting of socket-specific options. Currently only "no-delay" is implemented which enables the TCP_NODELAY socket option, causing the socket to flush send buffers immediately (instead of waiting to collect all output before sending). This option is useful for protocols that need fast request/response turn-around times.

socketTimeout sets connection timeout of a socket connection. A negative timeout can be given to query the old value.

Value

file, pipe, fifo, url, gzfile, bzfile, xzfile, zstdfile, unz, socketConnection, socketAccept and serverSocket return a connection object which inherits from class "connection" and has a first more specific class.

open and flush return NULL, invisibly.

close returns either NULL or an integer status, invisibly. The status is from when the connection was last closed and is available only for some types of connections (e.g., pipes, files and fifos): typically zero values indicate success. Negative values will result in a warning; if writing, these may indicate write failures and should not be ignored. Connections should be closed explicitly when finished with to avoid wasting resources and to reduce the risk that some buffered data in output connections would be lost (see on.exit() for how to run code also in case of error).

isOpen returns a logical value, whether the connection is currently open.

isIncomplete returns a logical value, whether the last read attempt from a non-blocking connection provided no data (currently no data from a socket or an unterminated line in readLines), or for an output text connection whether there is unflushed output. See example below.

socketTimeout returns the old timeout value of a socket connection.

URLs

url and file support URL schemes ‘⁠file://⁠’, ‘⁠http://⁠’, ‘⁠https://⁠’ and ‘⁠ftp://⁠’.

method = "libcurl" allows more schemes: exactly which schemes is platform-dependent (see libcurlVersion), but all platforms will support ‘⁠https://⁠’ and most platforms will support ‘⁠ftps://⁠’.

Support for the ‘⁠ftp://⁠’ scheme by the "internal" method was deprecated in R 4.1.1 and removed in R 4.2.0.

Most methods do not percent-encode special characters such as spaces in ‘⁠http://⁠’ URLs (see URLencode), but it seems the "wininet" method does.

A note on ‘⁠file://⁠’ URLs (which are handled by the same internal code irrespective of argument method). The most general form (from RFC1738) is ‘⁠file://host/path/to/file⁠’, but R only accepts the form with an empty host field referring to the local machine.

On a Unix-alike, this is then ‘⁠file:///path/to/file⁠’, where ‘⁠path/to/file⁠’ is relative to ‘/’. So although the third slash is strictly part of the specification not part of the path, this can be regarded as a way to specify the file ‘/path/to/file’. It is not possible to specify a relative path using a file URL.

In this form the path is relative to the root of the filesystem, not a Windows concept. The standard form on Windows is ‘⁠file:///d:/R/repos⁠’: for compatibility with earlier versions of R and Unix versions, any other form is parsed as R as ‘⁠file://⁠’ plus path_to_file. Also, backslashes are accepted within the path even though RFC1738 does not allow them.

No attempt is made to decode a percent-encoded ‘⁠file:⁠’ URL: call URLdecode if necessary.

All the methods attempt to follow redirected HTTP and HTTPS URLs.

Server-side cached data is always accepted.

Function download.file and several contributed packages provide more comprehensive facilities to download from URLs.

Modes

Possible values for the argument open are

"r" or "rt"

Open for reading in text mode.

"w" or "wt"

Open for writing in text mode.

"a" or "at"

Open for appending in text mode.

"rb"

Open for reading in binary mode.

"wb"

Open for writing in binary mode.

"ab"

Open for appending in binary mode.

"r+", "r+b"

Open for reading and writing.

"w+", "w+b"

Open for reading and writing, truncating file initially.

"a+", "a+b"

Open for reading and appending.

Not all modes are applicable to all connections: for example URLs can only be opened for reading. Only file and socket connections can be opened for both reading and writing. An unsupported mode is usually silently substituted.

If a file or fifo is created on a Unix-alike, its permissions will be the maximal allowed by the current setting of umask (see Sys.umask).

For many connections there is little or no difference between text and binary modes. For file-like connections on Windows, translation of line endings (between LF and CRLF) is done in text mode only (but text read operations on connections such as readLines, scan and source work for any form of line ending). Various R operations are possible in only one of the modes: for example pushBack is text-oriented and is only allowed on connections open for reading in text mode, and binary operations such as readBin, load and save can only be done on binary-mode connections.

The mode of a connection is determined when actually opened, which is deferred if open = "" is given (the default for all but socket connections). An explicit call to open can specify the mode, but otherwise the mode will be "r". (gzfile, bzfile, xzfile and zstdfile connections are exceptions, as the compressed file always has to be opened in binary mode and no conversion of line-endings is done even on Windows, so the default mode is interpreted as "rb".) Most operations that need write access or text-only or binary-only mode will override the default mode of a non-yet-open connection.

Append modes need to be considered carefully for compressed-file connections. They do not produce a single compressed stream on the file, but rather append a new compressed stream to the file. Readers may or may not read beyond end of the first stream: currently R does so for gzfile, bzfile, xzfile, zstdfile and gzcon connections.

Compression

R supports gzip, bzip2, zstd and xz compression (also read-only support for its precursor, lzma compression).

For reading, the type of compression (if any) can be determined from the first few bytes of the file. Thus for file(raw = FALSE) connections, if open is "", "r" or "rt" the connection can read any of the compressed file types as well as uncompressed files. (Using "rb" will allow compressed files to be read byte-by-byte.) Similarly, gzfile connections can read any of the forms of compression and uncompressed files in any read mode.

(The type of compression is determined when the connection is created if open is unspecified and a file of that name exists. If the intention is to open the connection to write a file with a different form of compression under that name, specify open = "w" when the connection is created or unlink the file before creating the connection.)

For write-mode connections, compress specifies how hard the compressor works to minimize the file size, and higher values need more CPU time and more working memory (up to ca 800Mb for xzfile(compress = 9)). For xzfile negative values of compress correspond to adding the xz argument -e: this takes more time (double?) to compress but may achieve (slightly) better compression. The default (6) has good compression and modest (100Mb memory) usage: but if you are using xz compression you are probably looking for high compression.

For zstd negative numbers correspond to the --fast= compression levels which are faster but yield slightly worse compression.

Choosing the type of compression involves tradeoffs: gzip, bzip2 and xz are successively less widely supported, need more resources for both compression and decompression, and achieve more compression (although individual files may buck the general trend). Typical experience is that bzip2 compression is 15% better on text files than gzip compression, and xz with maximal compression 30% better. The experience with R save files is similar, but on some large ‘.rda’ files xz compression is much better than the other two. With current computers decompression times even with compress = 9 are typically modest and reading compressed files is usually faster than uncompressed ones because of the reduction in disc activity.

The zstd compression support is currently optional and will be only included if the underlying library is present at build time. All related functions will return an error if the support is not present. It typically offers slightly worse compression than xz, but is orders of magnitude faster to compress.

Encoding

The encoding of the input/output stream of a connection can be specified by name in the same way as it would be given to iconv: see that help page for how to find out what encoding names are recognized on your platform. Additionally, "" and "native.enc" both mean the ‘native’ encoding, that is the internal encoding of the current locale and hence no translation is done.

When writing to a text connection, the connections code always assumes its input is in native encoding, so e.g. writeLines has to convert text to native encoding. The native encoding is UTF-8 on most systems (since R 4.2 also on recent Windows) and can represent all characters. writeLines does not do the conversion when useBytes=TRUE (for expert use only, only useful on systems with native encoding other than UTF-8), but the connections code still behaves as if the text was in native encoding, so any attempt to convert encoding (encoding argument other than "" and "native.enc") in connections will produce incorrect results.

When reading from a text connection, the connections code re-encodes the input to native encoding (from the encoding given by the encoding argument). On systems where UTF-8 is not the native encoding, one can read text not representable in the native encoding using readLines and scan by providing them with an unopened connection that has been created with the encoding argument specifying the input encoding. readLines and scan would then instruct the connections code to convert the text to UTF-8 (instead of native encoding) and they will return it marked (aka declared, see Encoding) as "UTF-8". Finally and for expert use only, one may disable re-encoding of input by specifying "" or "native.enc" as encoding for the connection, but then mark the text as being "UTF-8" or "latin1" via the encoding argument of readLines and scan.

Re-encoding only works for connections in text mode: reading from a connection with re-encoding specified in binary mode will read the stream of bytes, but mixing text and binary mode reads (e.g., mixing calls to readLines and readChar) is likely to lead to incorrect results.

The encodings "UCS-2LE" and "UTF-16LE" are treated specially, as they are appropriate values for Windows ‘Unicode’ text files. If the first two bytes are the Byte Order Mark 0xFEFF then these are removed as some implementations of iconv do not accept BOMs. Note that whereas most implementations will handle BOMs using encoding "UCS-2" and choose the appropriate byte order, some (including earlier versions of glibc) will not. There is a subtle distinction between "UTF-16" and "UCS-2" (see https://en.wikipedia.org/wiki/UTF-16): the use of characters in the ‘Supplementary Planes’ which need surrogate pairs is very rare so "UCS-2LE" is an appropriate first choice (as it is more widely implemented).

The encoding "UTF-8-BOM" is accepted for reading and will remove a Byte Order Mark if present (which it often is for files and webpages generated by Microsoft applications). If a BOM is required (it is not recommended) when writing it should be written explicitly, e.g. by writeChar("\ufeff", con, eos = NULL) or writeBin(as.raw(c(0xef, 0xbb, 0xbf)), binary_con)

Encoding names "utf8", "mac" and "macroman" are not portable, and not supported on all current R platforms. "UTF-8" is portable and "macintosh" is the official (and most widely supported) name for ‘Mac Roman’. (R maps "utf8" to "UTF-8" internally.)

Requesting a conversion that is not supported is an error, reported when the connection is opened. Exactly what happens when the requested translation cannot be done for invalid input is in general undocumented. On output the result is likely to be that up to the error, with a warning. On input, it will most likely be all or some of the input up to the error.

It may be possible to deduce the current native encoding from Sys.getlocale("LC_CTYPE"), but not all OSes record it.

Blocking

Whether or not the connection blocks can be specified for file, url (default yes), fifo and socket connections (default not).

In blocking mode, functions using the connection do not return to the R evaluator until the read/write is complete. In non-blocking mode, operations return as soon as possible, so on input they will return with whatever input is available (possibly none) and for output they will return whether or not the write succeeded.

The function readLines behaves differently in respect of incomplete last lines in the two modes: see its help page.

Even when a connection is in blocking mode, attempts are made to ensure that it does not block the event loop and hence the operation of GUI parts of R. These do not always succeed, and the whole R process will be blocked during a DNS lookup on Unix, for example.

Most blocking operations on HTTP/FTP URLs and on sockets are subject to the timeout set by options("timeout"). Note that this is a timeout for no response, not for the whole operation. The timeout is set at the time the connection is opened (more precisely, when the last connection of that type – ‘⁠http:⁠’, ‘⁠ftp:⁠’ or socket – was opened).

Fifos

Fifos default to non-blocking. That follows S version 4 and is probably most natural, but it does have some implications. In particular, opening a non-blocking fifo connection for writing (only) will fail unless some other process is reading on the fifo.

Opening a fifo for both reading and writing (in any mode: one can only append to fifos) connects both sides of the fifo to the R process, and provides an similar facility to file().

Clipboard

file can be used with description = "clipboard" in mode "r" only. This reads the X11 primary selection (see https://specifications.freedesktop.org/clipboard-spec/latest), which can also be specified as "X11_primary" and the secondary selection as "X11_secondary". On most systems the clipboard selection (that used by ‘Copy’ from an ‘Edit’ menu) can be specified as "X11_clipboard".

When a clipboard is opened for reading, the contents are immediately copied to internal storage in the connection.

Unix users wishing to write to one of the X11 selections may be able to do so via xclip (https://github.com/astrand/xclip) or xsel (https://www.vergenet.net/~conrad/software/xsel/), for example by pipe("xclip -i", "w") for the primary selection.

macOS users can use pipe("pbpaste") and pipe("pbcopy", "w") to read from and write to that system's clipboard.

File paths

In most cases these are translated to the native encoding.

The exceptions are file and pipe on Windows, where a description which is marked as being in UTF-8 is passed to Windows as a ‘wide’ character string. This allows files with names not in the native encoding to be opened on file systems which use Unicode file names (such as NTFS but not FAT32).

‘⁠ftp://⁠’ URLs

Most modern browsers do not support such URLs, and ‘⁠https://⁠’ ones are much preferred for use in R.

It is intended that R will continue to allow such URLs for as long as libcurl does, but as they become rarer this is increasingly untested. What ‘protocols’ the version of libcurl being used supports can be seen by calling libcurlVersion().

Number of connections

There is a limit on the number of connections which can be allocated (not necessarily open) at any one time. It is good practice to close connections when finished with, but if necessary garbage-collection will be invoked to close those connections without any R object referring to them.

The default limit is 128 (including the three terminal connections, stdin, stdout and stderr). This can be increased when R is started using the option --max-connections=N, where the maximum allowed value is 4096.

However, many types of connections use other resources which are themselves limited. Notably on Unix, ‘file descriptors’ which by default are per-process limited: this limits the number of connections using files, pipes and fifos. (The default limit is 256 on macOS (and Solaris) but 1024 on Linux. The limit can be raised in the shell used to launch R, for example by ulimit -n.) File descriptors are used for many other purposes including dynamically loading DSO/DLLs (see dyn.load) which may use up to 60% of the limit.

Windows has a default limit of 512 open C file streams: these are used by at least file, gzfile, bzfile, xzfile, zstdfile, pipe, url and unz connections applied to files (rather than URLs).

Package parallel's makeCluster uses socket connections to communicate with the worker processes, one per worker.

Note

R's connections are modelled on those in S version 4 (Chambers 1998). However R goes well beyond the S model, for example in output text connections and URL, compressed and socket connections. The default open mode in R is "r" except for socket connections. This differs from S, where it is the equivalent of "r+", known as "*".

On (historic) platforms where vsnprintf does not return the needed length of output there is a 100,000 byte output limit on the length of a line for text output on fifo, gzfile, bzfile, xzfile and zstdfile connections: longer lines will be truncated with a warning.

References

Chambers J. M. (1998). Programming with Data. Springer, New York. ISBN 978-0-387-98503-9.

Ripley B. D. (2001). “Connections.” R News, 1(1), 16–17. https://journal.r-project.org/articles/RN-2001-005/.

See Also

textConnection, seek, showConnections, pushBack.

Functions making direct use of connections are (text-mode) readLines, writeLines, cat, sink, scan, parse, read.dcf, dput, dump and (binary-mode) readBin, readChar, writeBin, writeChar, load and save.

capabilities to see if fifo connections are supported by this build of R.

gzcon to wrap gzip (de)compression around a connection.

options HTTPUserAgent, internet.info, netrc and timeout are used by some of the methods for URL connections.

memCompress for more ways to (de)compress and references on data compression.

extSoftVersion for the versions of the zlib (for gzfile), bzip2 and xz libraries in use.

To flush output to the Windows and macOS consoles, see flush.console.

Examples

zzfil <- tempfile(fileext=".data")
zz <- file(zzfil, "w")  # open an output file connection
cat("TITLE extra line", "2 3 5 7", "", "11 13 17", file = zz, sep = "\n")
cat("One more line\n", file = zz)
close(zz)
readLines(zzfil)
unlink(zzfil)

zzfil <- tempfile(fileext=".gz")
zz <- gzfile(zzfil, "w")  # compressed file
cat("TITLE extra line", "2 3 5 7", "", "11 13 17", file = zz, sep = "\n")
close(zz)
readLines(zz <- gzfile(zzfil))
close(zz)
unlink(zzfil)
zz # an invalid connection

zzfil <- tempfile(fileext=".bz2")
zz <- bzfile(zzfil, "w")  # bzip2-ed file
cat("TITLE extra line", "2 3 5 7", "", "11 13 17", file = zz, sep = "\n")
close(zz)
zz # print() method: invalid connection
print(readLines(zz <- bzfile(zzfil)))
close(zz)
unlink(zzfil)

## An example of a file open for reading and writing
Tpath <- tempfile("test")
Tfile <- file(Tpath, "w+")
c(isOpen(Tfile, "r"), isOpen(Tfile, "w")) # both TRUE
cat("abc\ndef\n", file = Tfile)
readLines(Tfile)
seek(Tfile, 0, rw = "r") # reset to beginning
readLines(Tfile)
cat("ghi\n", file = Tfile)
readLines(Tfile)

Tfile # -> print() :  "valid" connection
close(Tfile)
Tfile # -> print() :  "invalid" connection
unlink(Tpath)

## We can do the same thing with an anonymous file.
Tfile <- file()
cat("abc\ndef\n", file = Tfile)
readLines(Tfile)
close(Tfile)

## Not run: ## fifo example -- may hang even with OS support for fifos
if(capabilities("fifo")) {
  zzfil <- tempfile(fileext="-fifo")
  zz <- fifo(zzfil, "w+")
  writeLines("abc", zz)
  print(readLines(zz))
  close(zz)
  unlink(zzfil)
}
## End(Not run)

## Unix examples of use of pipes

# read listing of current directory
readLines(pipe("ls -1"))

# remove trailing commas.  Suppose

## Not run: % cat data2_
450, 390, 467, 654,  30, 542, 334, 432, 421,
357, 497, 493, 550, 549, 467, 575, 578, 342,
446, 547, 534, 495, 979, 479
## End(Not run)
# Then read this by
scan(pipe("sed -e s/,$// data2_"), sep = ",")


# convert decimal point to comma in output: see also write.table
# both R strings and (probably) the shell need \ doubled
zzfil <- tempfile("outfile")
zz <- pipe(paste("sed s/\\\\./,/ >", zzfil), "w")
cat(format(round(stats::rnorm(48), 4)), fill = 70, file = zz)
close(zz)
file.show(zzfil, delete.file = TRUE)

## Not run: 
## example for a machine running a finger daemon

con <- socketConnection(port = 79, blocking = TRUE)
writeLines(paste0(system("whoami", intern = TRUE), "\r"), con)
gsub(" *$", "", readLines(con))
close(con)

## End(Not run)

## Not run: 
## Two R processes communicating via non-blocking sockets
# R process 1
con1 <- socketConnection(port = 6011, server = TRUE)
writeLines(LETTERS, con1)
close(con1)

# R process 2
con2 <- socketConnection(Sys.info()["nodename"], port = 6011)
# as non-blocking, may need to loop for input
readLines(con2)
while(isIncomplete(con2)) {
   Sys.sleep(1)
   z <- readLines(con2)
   if(length(z)) print(z)
}
close(con2)

## examples of use of encodings
# write a file in UTF-8
cat(x, file = (con <- file("foo", "w", encoding = "UTF-8"))); close(con)
# read a 'Windows Unicode' file
A <- read.table(con <- file("students", encoding = "UCS-2LE")); close(con)

## End(Not run)