-
url_parse()
is considerably faster thanks to performance optimisations by and discussion with @DyfanJones (#429). -
req_timeout()
now resets the value ofconnecttimeout
set by curl. This ensures that you can usereq_timeout()
to increase the connection timeout past 10s (#395). -
req_retry()
gives a clearer error ifafter
returns the wrong type of value (#385). -
req_template()
now works when you have a bare:
in a template that uses "uri" style (#389). -
req_perform_stream()
gains around = c("byte", "line")
argument to control how the stream is rounded (#437).
-
local_mock()
andwith_mock()
have been deprecated in favour oflocal_mocked_responses()
andwith_mocked_responses()
(#301). -
multi_req_perform()
is deprecated in favour ofreq_perform_parallel()
.req_stream()
is deprecated in favour ofreq_perform_stream()
(#314). -
oauth_flow_auth_code()
deprecateshost_name
andport
arguments in favour of usingredirect_uri
. It also deprecateshost_ip
since it seems unlikely that changing this is ever useful. -
oauth_flow_auth_code_listen()
now takes a singleredirect_uri
argument instead of separatehost_ip
andport
arguments. This is a breaking change but I don't expect anyone to call this function directly (which was confirmed by a GitHub search) so I made the change without deprecation. -
req_body_form()
andreq_body_multipart()
now require data...
; they no longer accept a single list for compatibility with the 0.1.0 API.
-
New
req_perform_sequential()
performs a known set of requests sequentially. It has an interface similar toreq_perform_parallel()
but with no limitations, and the cost of being slower (#361). -
New
req_perform_iterative()
performs multiple requests, where each request is derived from the previous response (@mgirlich, #8). -
req_perform_parallel()
replacesmulti_req_perform()
to match the new naming scheme (#314). It gains aprogress
argument. -
req_perform_iterative()
,req_perform_parallel()
, andreq_perform_sequential()
share a new error handling strategy. By default, errors will be bubbled up, but you can choose an alternative strategy with theon_error
argument (#372). -
A new family of functions
resps_successes()
,resps_failures()
,resps_requests()
andresps_data()
make it easier to work with lists of responses (#357). Behind the scenes, these work because the request is now stored in the response (or error) object (#357). -
resp_body_json()
andresp_body_xml()
now cache the parsed values so that you can use them repeatedly without worrying about the performance cost. This supports the designreq_perform_iterative()
by avoiding the need to carefully pass around a parsed object.
-
A new
vignette("oauth")
gives many more details about how OAuth works and how to use it with httr2 (#234), and the OAuth docs have been overhauled to make it more clear that you should usereq_oauth_*()
, notoauth_*()
(#330). -
If you are using an OAuth token with a refresh token, and that refresh token has expired, then httr2 will now re-run the entire flow to get you a new token (#349).
-
New
oauth_cache_path()
returns the path that httr2 uses for caching OAuth tokens. Additionally, you can now change the cache location by setting theHTTR2_OAUTH_CACHE
env var. This is now more obvious to the user, because httr2 now informs the user whenever a token is cached. -
oauth_flow_auth_code()
gains aredirect_uri
argument rather than deriving this URL automatically from thehost_name
andport
(#248). It uses this argument to automatically choose which strategy to use to get the
auth code, either launching a temporary web server or, new, allowing you to manually enter the details with the help of a custom JS/HTML page hosted elsewhere, or by copying and pasting the URL you're redirected to (@fh-mthomson, #326). The temporary web server now also respects the path component ofredirect_uri
, if the API needs a specific path (#149). -
New
oauth_token_cached()
allows you to get an OAuth token while still taking advantage of httr2's caching and auto-renewal features. For expert use only (#328).
-
@mgirlich is now a httr2 contributor in recognition of his many contributions.
-
req_cache()
gainsmax_n
,max_size
, andmax_age
arguments to automatically prune the cache. By default, the cache will stay under 1 GB (#207). -
New
req_body_json_modify()
allows you to iteratively modify a JSON body of a request. -
New
req_cookie_preserve()
lets you use a file to share cookies across requests (#223). -
New
req_progress()
adds a progress bar to long downloads or uploads (#20). -
New
resp_check_content_type()
to check response content types (#190).resp_body_json()
and friends give better errors if noContent-Type
header is present in the response (#284). -
New
resp_has_body()
returns aTRUE
orFALSE
depending on whether or not the response has a body (#205). -
New
resp_url()
,resp_url_path()
,resp_url_queries()
andresp_url_query()
to extract various part of the response url (#57). -
req_url_query()
gains a.multi
parameter that controls what happens when you supply multiple values in a vector. The default will continue to error but you can use.multi = "comma"
to separate with commas,"pipe"
to separate with|
, and"explode"
to generate one parameter for each value (e.g.?a=1&a=2
) (#350). -
New
secret_encrypt_file()
andsecret_decrypt_file()
for encrypting and decrypting files (#237).
-
The httr2 examples now only run on R 4.2 and later so that we can use the base pipe and lambda syntax (#345).
-
OAuth errors containing a url now correctly display that URL (instead of the string "uri").
-
curl_translate()
now uses the base pipe, and produces escapes with single quotes or raw strings in case double quotes can't be used (@mgirlich, #264). It gains the argumentsimplify_headers
that removes some common but unimportant headers, likeSec-Fetch-Dest
orsec-ch-ua-mobile
(@mgirlich, #256). It also parses the query components of the url (@mgirlich, #259) and works with multiline commands from the clipboard (@mgirlich, #254). -
local_mocked_responses()
andwith_mocked_responses()
now accept a list of responses which will be returned in sequence. They also now correctly trigger errors when the mocked response represents an HTTP failure (#252). -
oauth_flow_refresh()
now only warns, not errors, if therefresh_token
changes, making it a little easier to use in manual workflows (#186). -
obfuscated()
values now display their original call when printed. -
req_body_json()
gains custom contenttype
argument and respects custom content-type set in header (@mgirlich, #189). -
req_cache()
now combine the headers of the new response with the headers of the cached response. In particular, this fixesresp_body_json/xml/html()
on cached responses (@mgirlich, #277). -
req_perform()
now throws error with classhttr2_failure/httr2_error
if the request fails, and that error now captures the curl error as the parent. If the request succeeds, but the response is an HTTP error, that error now also has super classhttr2_error
. This means that all errors thrown by httr2 now inherit from thehttr2_error
class. See new docs in?req_error()
for more details. -
req_perform()
's progress bar can be suppressed by settingoptions(httr2_progress = FALSE)
(#251). Progress bars displayed while waiting for some time to pass now tell you why they're waiting (#206). -
req_oauth_bearer_jwt()
now includes the claim in the cache key (#192). -
req_oauth_device()
now takes aauth_url
parameter making it usable (#331, @taerwin). -
req_url_query()
gains a.multi
parameter that controls what happens when you supply multiple values in a vector. The default will continue to error but you can use.multi = "comma"
to separate with commas,"pipe"
to separate with|
, and"explode"
to generate one parameter for each value (e.g.?a=1&a=2
) (#350).
-
New
example_url()
to launch a local server, making tests and examples more robust. -
New
throttle_status()
to make it a little easier to verify what's happening with throttling. -
req_oauth_refresh()
now respects therefresh_token
for caching (@mgirlich, #178). -
req_perform()
now always sleeps before a request, rather than after it. It also gains anerror_call
argument and communicates more clearly where the error occurred (@mgirlich, #187). -
req_url_path()
andreq_url_path_append()
can now handleNULL
or empty...
and the elements of...
can also have length > 1 (@mgirlich, #177). -
sys_sleep()
(used byreq_retry()
andreq_throttle()
) gains a progress bar (#202).
-
curl_translate()
can now handle curl copied from Chrome developer tools (@mgirlich, #161). -
req_oauth_*()
can now refresh OAuth tokens. One, two, or even more times! (@jennybc, #166) -
req_oauth_device()
can now work in non-interactive environments, as intendend (@flahn, #170) -
req_oauth_refresh()
andoauth_flow_refresh()
now use the envvarHTTR2_REFRESH_TOKEN
, notHTTR_REFRESH_TOKEN
(@jennybc, #169). -
req_proxy()
now uses the appropriate authentication option (@jl5000). -
req_url_query()
can now opt out of escaping withI()
(@boshek, #152). -
Can now print responses where content type is the empty string (@mgirlich, #163).
-
"Wrapping APIs" is now an article, not a vignette.
-
req_template()
now appends the path instead of replacing it (@jchrom, #133)
-
req_body_form()
,req_body_multipart()
, andreq_url_query()
now support multiple arguments with the same name (#97, #107). -
req_body_form()
,req_body_multipart()
, now match the interface ofreq_url_query()
, taking name-value pairs in...
. Supplying a singlelist()
is now deprecated and will be removed in a future version. -
req_body_json()
now overrides the existing JSON body, rather than attempting to merge with the previous value (#95, #115). -
Implement
req_proxy()
(owenjonesuob, #77).
-
httr_path
class renamed tohttr2_path
to correctly match package name (#99). -
oauth_flow_device()
gains PKCE support (@flahn, #92), and the interactive flow is a little more user friendly. -
req_error()
can now correct force successful HTTP statuses to fail (#98). -
req_headers()
will now overrideContent-Type
set byreq_body_*()
(#116). -
req_throttle()
correctly sets throttle rate (@jchrom, #101). -
req_url_query()
never uses scientific notation for queries (#93). -
req_perform()
now respectshttr::with_verbose()
(#85). -
response()
now defaultsbody
toraw(0)
for consistency with real responses (#100). -
req_stream()
no longer throws an error for non 200 http status codes (@DMerch, #137)
-
Fix R CMD check failures on CRAN
-
Added a
NEWS.md
file to track changes to the package.