• Important fix in a low-level bug in R_curl_easy_setopt where we were assuming the logical vector isProtected had the same length as values. Thanks to Bruce Hoff for identifying this and providing a patch.
  • Correction for mapping new lines in curlPercentEncode().
  • 1.95-1 &2 & 3

  • Changes for simpler compilation by others on Windows.
    The changes are appreciated.
  • postForm() coerces only atomic arguments for the form, leaving, for example, FileUploadInfo objects as they are.
  • 1.95-0

  • Expose interface to C routine for posting a form, .postForm().
    Request and patch from Hadley Wickham.
  • Field httpHeader added to the error/condition object when an HTTP error is raised.
    Added close() method for CFILE.
  • stop.if.HTTP.error() now raises an error for 4xx and 5xx status codes.
    Thanks to Thomas Themel.
  • 1.94-0

  • Fix to getForm() to allow the URL to contain a ?, e.g. with some parameters specified in the URL
  • 1.93-0

  • getURLContent()'s header parameter can be TRUE or FALSE. TRUE returns the header along with the body as a list containing just these two elements.
  • postForm() coerces its arguments to characters so we can pass values of arbitrary types.
  • postForm() checks the header in the response and throws an error if the request was not successful.
  • Added finalizer for CFILE()
  • Allow writefunction and headerfunction to return TRUE to indicate success.
  • Exported AUTH_* variables, e.g. AUTH_BASIC, AUTH_DIGEST, etc.
  • 1.92-1

  • Set the error buffer when we call reset for a curl object.
    Thanks to Hadley Wickham for bringing this to my attention.
  • 1.92-0

  • Added conversion for downloaded content to map Unicode characters of the form \unnnn to proper UTF encoding in R.
    Code is adapted from Alex Couture-Beil's rjson package. Thanks.
  • 1.91-1

  • Move logo.jpg and a reference to it in an example.
  • 1.91-0

  • Enhanced httpPUT to make it simpler and take care of serializing the content and setting the necessary options.
  • 1.9-5

  • Enable the examples in getURL() help page if the session is interactive so that they won't be run in test/check mode.
  • 1.9-4

  • RCurl doesn't have to be in the search path to find the curlError() function which is called when an error occurs in curlPerform().
  • 1.9-3

    Import methods.


    Documentation updates.


  • Avoid warnings about names slot in an EnumValue by adding an explicit names slot. This was a result of a change in R-2.14.0.
  • 1.9-0

  • Introduced url.exists() function.
  • Introduced curlError() and a collection of error "condition" objects corresponding to CURLCode status errors. Errors can now be caught with tryCatch according to their class.
  • Examples check the URLs are accessible before making the requests.
    The tests do not.
  • 1.8-0

  • getURLContent() attempts to handle encoding more comprehensively.
  • Added encoding argument for dynCurlReader() function.
    This attempts to use this encoding for the content it receives when it is not binary, but text.
  • dynCurlReader() has a more complete reset function and some additional state variables (e.g. requestCount). Contributed by Jeroen Ooms.
  • 1.7-0

  • Added simple wrapper functions httpGET, httpPOST, httpPUT, httpHEAD, httpOPTIONS and httpDELETE corresponding to the HTTP methods.
  • postForm() now correctly handles multiple form parameters with the same name (rather than reusing the first value multiple times).
  • 1.6-10

  • Fix garbage collection for common case of calling getURLContent() that uses the default value of the header parameter.
  • 1.6-9

  • Removed inclusion of curl/types.h which is now not part of libcurl in many installations.
  • 1.6-8

  • Added isHTTP to dynCurlReader so that we can better handle FTP requests.
  • 1.6-7

  • Fixed the case when we omitted the name of the field in httpheaders when there was just one entry which contains a ':', e.g. c(SOAPAction = "http://www.omegahat.org").
  • 1.6-6

  • Fixed a problem when no Content-Encoding is returned.
    Discovered by Philip Laflamme.
  • 1.6-5

  • Modified the FTP example in getURL() to account for the rare case that after we get the list of files in a directory on the server, one or more of them is deleted when we return to retrieve it.
    This appears to be occurring i386-apple-darwin9.8.0 (32-bit) but not on other platforms.
  • 1.6-4

  • Fix for uploading data from memory in PUT or POST with the readfunction option. There was an off by one error that caused an extra byte to be upload in some/most circumstances.
  • Workaround in C code underlying getURIAsynchronous() that fails on Fedora 14 with libcurl 7.21.*
    See http://lists.gnu.org/archive/html/gnash-dev/2010-08/msg00156.html
  • 1.6-3

  • getForm() now correctly processes multiple instances of a form parameter with the same name rather than using the value of the first version, e.g. arg1 = 'bob', arg1 = 'jane'. Found by Richard Cotton.
  • 1.6-2

  • Correction of bug fix in 1.6-1 to ensure results are returned.
  • 1.6-1

  • Changed postForm() to not register a headerfunction if either writefunction or headerfunction is specified by the caller in the options.
    Discovered by Jereon Ooms
  • 1.6-0

  • Respect Content-Encoding for determining if returned content is binary.
  • Pass the encoding to the dynamic content reader.
  • curlPercentEncode has a new post.amp parameter controlling whether the % characters introduce by inserting escape codes are also escaped, i.e. # becomes %23 and then %2523 since %25 is the code for %
  • When posting a form via postForm(), individual parameters can have the class "AsIs" to avoid escaping their contents, e.g. converting &, spaces, etc. to percent-encodings.
  • Added json to the list of text content types.
  • 1.5-0

  • Added .isProtected parameter to curlSetOpt() to allow caller control the preservation of the values from garbage collection to ensure they are not garbage collected while the curl handle is still alive.
    This solves the problem
  • 1.4-3

    Fix for ftpUpload() to close the connection to a local file if that is what is being uploaded (rather than content in memory). Identified by Spencer Graves.


  • Added fields for debugGatherer() for the SSL data-in and data-out fields which was causing an error if SSL was being used with debugging information.
  • 1.4-2

  • New release because there may have been an earlier 1.4-1 on the Web site that got overwritten with this version without the numbers being updated.
  • 1.4-1

  • Handle case where a "blank"/white-space line in the body of an HTTP response can be treated as part of a (new) header.
    Found through the rtracklayer package.
  • 1.4-0

  • Minor but important experimental changes to the way the dynamic reader process "HTTP 100 Continue" responses and gathering the header from the actual call.
    Still needs mor refinement, but this should be an improvement.
  • Added mimeTypeExtensions and guessMIMEType() function.
    These are useful for mapping a file extension to the common MIME type, valuable when uploading/including content in a POST request.
  • 1.3-2

  • Minor changes to parsing of HTTP header for 100 "continue" replies.
  • 1.3-1

  • Added curlPercentEncode() and a parameter in postForm() for specifying the encoding function. This is of use for x-www-form-encoded POST submissions.
  • 1.3-0

  • scp() function for transferring a file from a remote machine to the local one on which the request is made.
  • Improved error handling for requests which are not HTTP requests.
  • Simpler configuration code for Windows at the expense of unnecessary flexibility in the location of the libcurl development files.
  • 1.2-1

  • getForm() did not use the curl handle object it was passed.
  • 1.2-0

  • Recognize more Content-Types as being text and so handle the (HTTP) response as text rather than binary for those situations.
  • 1.1-0

  • Support for using default options for new curl handles taken from options("RCurlOptions").
  • 1.0-0

  • dynCurlReader() (and hence getURLContent()) adds the Content-Type attribute to the resulting R object.
  • getURLContent() allows the caller to specify the reader for the header and body via the header parameter.
  • 0.99-0

  • Added a binary argument to getURLContent(), getForm() and postForm() to allow the caller to control whether the response returned by the Web server is treated as binary or text. If this is not specified, we examine the Content-Type field in the response's header and try to determine from that whether the content is binary or text.
  • The set of application Content-Types recognized as being text is now retrieved from the R options vector via the element text.content.types. This allows users to change this set to recognize "new" types.
  • Fixed a bug in setting the HTTPHEADER when a value contained a ':' and there was only one value being set.
    curlSetOpt() now detects this case better and also allows the caller to explicitly control this via the .forceHeaderNames parameter.
  • getURLContent() was not honoring ... arguments for setting curl options.
  • 0.98-1

  • Added configure- and compile-time checking for different versions and enumerated constants so that this will install for older versions of libcurl. Goes all the way back to 7.14.0, i.e. 2005!
  • 0.98-0

  • Added CFILE class and function to allow us to work with files on disk, e.g. for uploading, without reading them into R, and for speed to avoid using functions.
  • 0.97-3

  • dynCurlReader() discards the original content for the header when it is a 100 status and so is ready to read the subsequent header after the continuation.
    Found by Steffen Durinck in biomaRt.
  • 0.97-2

  • getURL() example involving listing files in an FTP directory now handles newlines of the form \r\n on Windows. Thanks to Brian Ripley.
  • Further tweaks for different versions of libcurl and the generated code mappings.
  • 0.97-1

  • Additional checks in the configuration for existence of specific enumeration constants to handle different versions of libcurl.
  • 0.97-0

  • Added configuration-time tests for the existence of newly added enumerations in the options and corresponding #ifdef's within the code. This handles different versions of libcurl with different options. The code in CodeGeneration/generate.R can deal with different versions and programmatically create the code for the most recent version of libcurl it sees with support for older versions.
  • 0.96-0

  • Updated the number of recognized curl options in curlPerform and higher-level functions.
  • getCurlInfo() can return certificate information and single linked lists that are characters
  • Introduced getCurlOptionTypes() which indicates the expected types of the different options.
  • Added several C-level enumeration collections as R values. These are not exported as there are many of them. They can be accessed with RCurl:::name, e.g. RCurl:::CURLPROXY_HTTP_1_0
    This involves bringing in Enumeration classes from RAutoGenRunTime.
  • Configuration tests for recently added libcurl options in the CURLOPT enumerations and avoid them in compilation if they are not present.
  • 0.95-1

  • parseHTTPHeader() deals with fields in the header that contain multiple ':' characters.
  • dynCurlReader() handles HTTP 100 status messages to Continue better.
  • 1


  • getURLContent() now uses a dynamic reader which processes the header and determines the Content-Type (and the encoding if text) and reads the body of the reply appropriately.
  • getForm() now uses getURLContent() and so handles binary results and different character sets.
  • 0.94-1

  • Silly initialization omission for the readfunction C routine handler which caused the ftpUpload() to produce "infinitely" large files on the server. Thanks to Herve Fulchiron.
  • 0.94-0

  • Added ftpUpload() function and small C routine and R function for READFUNCTION option callbacks.
  • Support for QUOTE, POSTQUOTE, PREQUOTE options for FTP.
  • 0.93-0

  • Added a reset() generic function and a method for resetting the contents of a CURLHandle to the original default values.
  • Fixed bug in terminating the string in calls to the debug callback function.
  • 0.92-0

  • Added getURLContent() as a top-level function which handles binary and text downloads by looking at the Content-Type field of the header.
  • getURLContent() also raises errors for requests with a status in the "400" class (e.g. 404 for Not Found) and the errors have (S3) classes that allow them to be handled via tryCatch().
  • Added getBinaryURL() function as a convenience for downloading binary files. Binary content from POSTing or GETing forms can be done by mimicing the few curl settings illustrated in getBinaryURL().
  • Slightly improved error messages when unnamed options are passed to curlOptions().
  • Fixed a bug in reading binary files when enlarging the "raw" buffer.
  • getForm() and postForm() now issue a warning if a curl option appears in the parameters for the form. This can be legitimate if the form happens to use a parameter with the same name as a curl option (see listCurlOptions()) and the error message can be silenced by explicitly providing a value for the .opts parameter in the call. The value can be list() or some actual curl options; the point is to indicate in the call that you have considered the parameters and the curl options.
  • The 'statusMessage' field of the HTTP header now contains the entire error message rather than the first word. Thanks to Valerie Obenchain for identifying the problem.
  • Errors from HTTP requests now have their own class and use the condition system in R (see tryCatch(), simpleError, etc.). The names of these error classes come from the HTTP specification and are gathered programmatically from the specification document. (See inst/HTTPErrors/). Unknown/non-matching error status values return a GenericHTTPError class. The others return a vector of class names containing a more specific class name (whose name is the status number) and the more general "HTTPError" class name.
  • 0.9-4

  • Fix escaping in the split string used in findHTTPHeaderEncoding which gave rise to warnings about vector lengths in an if() condition in getEncoding().
  • 0.9-3

  • Changes to remove some simple compiler warnings.
  • 0.9-2

  • Fix for base64() computing whether to encode or not incorrectly due to lazy evaluation.
  • Now we use the base64 encode and decode routines from the libcurl distribution with our own names on the routines to avoid synchronization problems.
  • Fix in chunkToLineReader to report number of "characters" read and warning emitted in C code if full number of "characters" not consumed.
  • 0.9-1

    Configuration-time check to see if Curl_base64_encode() in libcurl (unexported) has the new additional parameter and so requires us to use that..


  • Moved the tests/ directory to inst/examples/.
    This has some quite rich/advanced examples that should be available to people and also take some time to run during R CMD check.
  • Allow for an .encoding parameter in getURL()/getURI() for explicitly specifying the encoding of the response from the HTTP.
  • A mechanism that reads the HTTP response header and finds the Content-Type field and attempts to infer the encoding of the response.
    Note that this registers a handler routine for processing the header. This continues to work when an R function is provided as for the headerfunction option, but will not work correctly if a routine is provided for the headerfunction option.
  • Updates to deal with encoding of strings.
    The basic text gathering function (used as the argument for write and the curl option writefunction) returns nchar(, "bytes") which may be more than the number of characters in an encoded string.
  • Replaced internal calls to allocString() with mkCharLenCE() when available (i.e. R >= 2.8.0), or explicitly copy the string given to us from libcurl and create an encoded character string.
  • 0.8-3

  • Support for CURLOPT_POST and using application/x-www-form-urlencoded as the Content-Type in POST'ing a form. This is controlled via the style parameter in postForm.
  • Construct body of a form-urlencoded POST in R with the relevant escaping.
  • 0.8-1

  • Added code to support the new hashing of strings in R-devel (for what will be 2.6.0).
  • Bug fix which caused segmentation faults when external pointer objects were used as options for curl.
  • Allow dynamic symbol lookup in the RCurl DLL/SO to permit the use of routines such as R_curl_write_binary_data.
  • 0.7-0

  • Added support for base64 encoding and decoding using the code from libcurl (copyright
    See the file COPYING.
  • 0.7-0

  • Added mechanism and example for reading binary data from a request.
    This is not connected to R's connection mechanism as the API for that is not public and external extensions (i.e. via packages) are not feasible. (R's internals are not object oriented; one must work through the interpreted language even if working directly in compiled languages.)

    This is useful for handling gzip/bzip2 files, images, etc. To uncompress such in-memory data directly, see the Rcompression package.

  • fileUpload() function and mechanism for specifying details about a file whose contents are to be included in a form submission via postForm.
  • 0.6-4

  • Fix the problem identified by Seth Falcon of releasing/free'ing an R function object in the RCurl memory management.
    Added a type to the RCurlMemory structure so we know when we should R_Release() an object. Probably will also duplicate such functions when preserving them.
  • 0.6-3

    Introduced mechanism for converting options to target types, including enumerated values.
    Works for netrc option and extensible for any of the values.
    Added support for specifying an R function or C routine for the SSL_CTX_FUNCTION option.
    Accept a logical value from the R function used for the write callback. FALSE indicates an error and TRUE means that the function processed the text successfully.
    Fixed bug in memory management and garbage collection of the curl handles.


    Further changes re postForm() with a curl handle provided.
    Problem was NULL for httppost setting.


    Fixed type in postForm() that caused error when a curl handle was specified.
    Name matching in ... of curlSetOpt() caught the cur which should have been curl. Thanks to Jim McDonal for catching this one and using the curl handle mechanism for repeated calls.


  • Allow C routines to be passed as values for the CURL options that can be R functions. Use getNativeSymbolInfo(routine, package)$address to identify the routine.
  • Support for concurrent, asynchronous HTTP requests using the curl_multi interface.
  • See the examples in tests/ or on the Web:
  • 0.5-3

  • Fixed the SSL example in getURL to turn off verification of the peer, i.e.
    getURL('https://sourceforge.net', ssl.verifypeer = FALSE)
  • 0.5-2

    Fix for R-devel and counting of \0's in strings for the length.


    When POSTing a form, only the first string in each element of the list was written to the body of the form. In other words, multi-valued elements such as for checkboxes with several elements selected were mishandled.


    Internal correction for passing numbers as options.
    Manifested itself with progress function not being called.


  • Use the return value from a CURLOPT_WRITEFUNCTION
    This allows the user to raise an error.
  • CURL_CFLAGS and CURL_LIBS computed in configure so as to handle no-GNU versions of make.
  • 0.3-0

    Ensure that parameters for postForm are a list when passed to C code.

    Duncan Temple Lang <duncan@wald.ucdavis.edu>
    Last modified: Tue Mar 5 10:55:53 PST 2013