gh() now signals a classed gh_interrupt interrupt when a paginated
call is interrupted (e.g. via Ctrl+C / Escape). The condition object
carries the records fetched so far on its $gh_result field. If the
interrupt is not caught, then you can also access it via
rlang::last_error(), i.e. the partial results are in
rlang::last_error()$gh_result (#98).
GitHub PAT format validation now issues a warning by default instead of
throwing an error, so a PAT in an unrecognized (e.g. newly introduced)
format is still used. Set the gh_validate_tokens option or the
GH_VALIDATE_TOKENS environment to "off", "warn" or "error"
to configure this.
Token validation now recognizes newer GitHub App installation tokens
(ghs_ prefix) (#231, @jharmon-gilead).
gh() no longer returns empty results when httr2's HTTP cache
revalidates a stored response with 304 Not Modified.
gh() no longer errors on a 304 Not Modified response. This makes
manual conditional requests (e.g. passing If-None-Match via
.send_headers) usable: a 304 returns an empty gh_response with the
response headers (including the ETag) still attached (#219).
New fake_github_app(), a webfakes app that implements a small subset
of the GitHub REST API. This app is now used in the gh test suite, and
it can be used by upstream package authors as well.
Token validation now recognizes newer GitHub App installation tokens
(ghs_ prefix) (#231, @jharmon-gilead).
gh() no longer crashes when reporting a GitHub API error whose
errors field is a plain string rather than the documented array of
objects, as happens on some 422 responses (#229).
gh(), and other functions that use it, now do not save the request
headers in the returned object. Consequently, if you use the gh_next(),
gh_prev(), gh_first() or gh_last() functions and passed .token
and/or .send_headers explicitly to the original gh() (or similar)
call, then you'll also need to pass the same .token and/or
.send_headers to gh_next(), gh_prev(), gh_first() or gh_last().
New gh_token_exists() tells you if a valid GH token has been set.
gh() now uses a cache provided by httr2. This cache lives in
tools::R_user_dir("gh", "cache"), maxes out at 100 MB, and can be
disabled by setting options(gh_cache = FALSE) (#203).
gh_token() can now pick up on the viewer's GitHub credentials (if any)
when running on Posit Connect (@atheriel, #217).
gh_next(), gh_prev(), gh_first() and gh_last()
now work correctly again (#181).
When the user sets .destfile to write the response to disk, gh now
writes the output to a temporary file, which is then renamed to
.destfile after performing the request, or deleted on error (#178).
gh() gains a new .max_rate parameter that sets the maximum number of
requests per second.
gh is now powered by httr2. This should generally have little impact on normal
operation but if a request fails, you can use httr2::last_response() and
httr2::last_request() to debug.
gh() gains a new .max_wait argument which gives the maximum number of
minutes to wait if you are rate limited (#67).
New gh_rate_limits() function reports on all rate limits for the active
user.
gh can now validate GitHub fine-grained personal access tokens (@jvstein, #171).
gh now accepts lower-case methods i.e. both gh::gh("get /users/hadley/repos") and gh::gh("GET /users/hadley/repos") work (@maelle, #167).
Response headers ("response_headers") and response content
("response_content") are now returned in error conditions so that error
handlers can use information, such as the rate limit reset header, when
handling github_errors (@gadenbuie, #117).
gh now shows the correct number of records in its progress bar when paginating (#147).
New .params argument in gh() to make it easier to pass parameters to
it programmatically (#140).
gh_gql() now passes all arguments to gh() (#124).
gh now handles responses from pagination better, and tries to properly merge them (#136, @rundel).
gh can retrieve a PAT from the Git credential store, where the lookup is based on the targeted API URL. This now uses the gitcreds package. The environment variables consulted for URL-specific GitHub PATs have changed.
GITHUB_PAT_GITHUB_COM now, instead of
GITHUB_PAT_API_GITHUB_COMGITHUB_PAT_GITHUB_ACME_COM now,
instead of GITHUB_PAT_GITHUB_ACME_COM_API_V3See the documentation of the gitcreds package for details.
The keyring package is no longer used, in favor of the Git credential store.
The documentation for the GitHub REST API has moved to https://docs.github.com/rest and endpoints are now documented using the URI template style of RFC 6570:
GET /repos/:owner/:repo/issuesGET /repos/{owner}/{repo}/issuesgh accepts and prioritizes the new style. However, it still does parameter substitution for the old style.
Fixed an error that occurred when calling gh() with .progress = FALSE
(@gadenbuie, #115).
gh() accepts named NA parameters that are destined for the request
body (#139).
Raw responses from GitHub are now returned as raw vector.
Responses may be written to disk by providing a path in the .destfile
argument.
gh now sets .Last.error to the error object after an uncaught error,
and .Last.error.trace to the stack trace of the error.
gh() now silently drops named NULL parameters, and throws an
error for named NA parameters (#21, #84).
gh() now returns better values for empty responses, typically empty
lists or dictionaries (#66).
gh() now has an .accept argument to make it easier to set the
Accept HTTP header (#91).
New gh_gql() function to make it easier to work with the GitHub
GraphQL API.
gh now supports separate personal access tokens for GitHub Enterprise
sites. See ?gh_token for details.
gh now supports storing your GitHub personal access tokens (PAT) in the
system keyring, via the keyring package. See ?gh_token for details.
gh() can now POST raw data, which allows adding assets to releases (#56).
First public release.