gargle provides common infrastructure for use with Google APIs. This vignette describes one possible design for using gargle to deal with auth, in a client package that provides a high-level wrapper for a specific API.
There are frequent references to googledrive, which uses the design described here, along with bigrquery (v1.2.0 and higher), gmailr (v1.0.0 and higher), and googlesheets4 (the successor to googlesheets).
Getting a token requires several pieces of information and there are stark differences in how much users (need to) know or control about this process. Let’s review them, with an eye towards identifying the responsibilities of the package author versus the user.
In googledrive, the main user-facing auth function is
googledrive::drive_auth()
. Here is its definition (at least
approximately, remember this is static code):
# googledrive::
drive_auth <- function(email = gargle::gargle_oauth_email(),
path = NULL,
scopes = "https://www.googleapis.com/auth/drive",
cache = gargle::gargle_oauth_cache(),
use_oob = gargle::gargle_oob_default(),
token = NULL) {
# this catches a common error, where the user passes JSON for an OAuth client
# to the `path` argument, which only expects a service account token
gargle::check_is_service_account(path, hint = "drive_auth_configure")
cred <- gargle::token_fetch(
scopes = scopes,
client = drive_oauth_client() %||% <BUILT_IN_DEFAULT_CLIENT>,
email = email,
path = path,
package = "googledrive",
cache = cache,
use_oob = use_oob,
token = token
)
if (!inherits(cred, "Token2.0")) {
# throw an informative error here
}
.auth$set_cred(cred)
.auth$set_auth_active(TRUE)
invisible()
}
drive_auth()
is called automatically upon the first need
of a token and that can lead to user interaction, but does not
necessarily do so. drive_auth()
can be called explicitly by
the user, but usually that is not necessary. token_fetch()
is described in the vignette("how-gargle-gets-tokens")
. The
internal .auth
object maintains googledrive’s auth state
and is explained next.
A client package can use an internal object of class
gargle::AuthState
to hold the auth state. In googledrive,
the main auth file defines a placeholder .auth
object:
The actual initialization happens in .onLoad()
:
.onLoad <- function(libname, pkgname) {
utils::assignInMyNamespace(
".auth",
gargle::init_AuthState(package = "googledrive", auth_active = TRUE)
)
# other stuff
}
The initialization of .auth
is done this way to ensure
that we get an instance of the AuthState
class using the
current, installed version of gargle (vs. the ambient version from
whenever gargle was built, perhaps by CRAN).
An AuthState
instance has other fields which, in this
googledrive example, are not set at this point. The OAuth
client
and api_key
are configurable by the
user and, when NULL
, downstream functions can fall back to
internal credentials. The cred
field is populated by the
first call to drive_auth()
(direct or indirectly via
drive_token()
).
Most users should present OAuth user credentials to Google APIs. However, most users would love to be spared the fiddly details surrounding this. The OAuth client is one example. (Historically, following the lead of the httr package, we have used the term OAuth app, but we now use the term OAuth client.) The client is a component that most users do not even know about and they are content to use the same client for all work through a wrapper package: possibly, the client built into the package.
There is a field in the .auth
auth state to hold the
OAuth client
. Exported auth helpers,
drive_oauth_client()
and
drive_auth_configure()
, retrieve and modify the current
client to support users who want to (or must) take that level of
control.
library(googledrive)
# first: download the OAuth client as a JSON file
drive_auth_configure(
path = "/path/to/the/JSON/that/was/downloaded/from/gcp/console.json"
)
drive_oauth_client()
#> <gargle_oauth_client>
#> name: acme-corp-google-client
#> id: 123456789.apps.googleusercontent.com
#> secret: <REDACTED>
#> type: installed
#> redirect_uris: http://localhost
Do not “borrow” an OAuth client ID and secret from gargle or any other package; always use credentials associated with your package or provided by your user. Per the Google User Data Policy https://developers.google.com/terms/api-services-user-data-policy, your application must accurately represent itself when authenticating to Google API services.
Some APIs and scopes are considered so sensitive that is essentially impossible for a package to provide a built-in OAuth client. Users must get and configure their own client. Among the packages mentioned as examples, this is true of gmailr.
Some Google APIs can be used in an unauthenticated state, if and only if requests include an API key. For example, this is a great way to read a Google Sheet that is world-readable or readable by “anyone with a link” from a Shiny app, thereby designing away the need to manage user credentials on the server.
The user can provide their own API key via
drive_auth_configure(api_key =)
and retrieve that value
with drive_api_key()
, just as with the OAuth client. The
API key is stored in the api_key
field of the
.auth
auth state.
Many users aren’t motivated to take this level of control and appreciate when a package provides a built-in default API key. As with the client, packages should obtain their own API key and not borrow the gargle or tidyverse key.
Some APIs are not usable without a token, in which case a wrapper package may not even expose functionality for managing an API key. Among the packages mentioned as examples, this is true of bigrquery.
In contrast to the OAuth client and API key, every user must express which identity they wish to present to the API. This is a familiar concept and users expect to specify this. Since users may have more than one Google account, it’s quite likely that they will want to switch between accounts, even within a single R session, or that they might want to explicitly declare the identity to be used in a specific script or app.
That explains why drive_auth()
has the optional
email
argument that lets users proactively specify their
identity. drive_auth()
is usually called indirectly upon
first need, but a user can also call it proactively in order to specify
their target email
:
If email
is not given, gargle also checks for an option
named "gargle_oauth_email"
. The email
is used
to look up tokens in the cache and, if no suitable token is found, it is
used to pre-configure the OAuth chooser in the browser. Read more in the
help for gargle::gargle_oauth_email()
.
Most users have no concept of scopes. They just know they want to work with, e.g., Google Drive or Google Sheets. A client package can usually pick sensible default scopes, that will support what most users want to do.
Here’s a reminder of the signature of
googledrive::drive_auth()
:
# googledrive::
drive_auth <- function(email = gargle::gargle_oauth_email(),
path = NULL,
scopes = "https://www.googleapis.com/auth/drive",
cache = gargle::gargle_oauth_cache(),
use_oob = gargle::gargle_oob_default(),
token = NULL) { ... }
googledrive ships with a default scope, but a motivated user could
call drive_auth()
preemptively at the start of the session
and request different scopes. For example, if they intend to only read
data and want to guard against inadvertent file modification, they might
opt for the drive.readonly
scope.
The location of the token cache and whether to prefer out-of-bound
auth are two aspects of OAuth where most users are content to go along
with sensible default behavior. For those who want to exert control,
that can be done in direct calls to drive_auth()
or by
configuring an option. Read the help for
gargle::gargle_oauth_cache()
and
gargle::gargle_oob_default()
and
vignette("auth-from-web")
for more.
Here’s a concrete outline of how one could set up a client package to get its auth functionality from gargle.
Imports
.R/YOURPKG_auth.R
.gargle::AuthClass
object to hold
auth state. Follow the googledrive example above.R/YOURPKG_auth.R
. Examples: tidyverse/googledrive/R/drive_auth.R
and r-dbi/bigrquery/R/bq_auth.R
.YOURPKG_token()
and
YOURPKG_api_key()
(defined in the standard auth interface)
to insert a token or API key in your package’s requests.I focus on early use, by the naive user, with the OAuth flow. When
the user first calls a high-level googledrive function such as
drive_find()
, a Drive request is ultimately generated with
a call to googledrive::request_generate()
. Here is its
definition, at least approximately:
# googledrive::
request_generate <- function(endpoint = character(),
params = list(),
key = NULL,
token = drive_token()) {
ept <- drive_endpoint(endpoint)
if (is.null(ept)) {
# throw error about unrecognized endpoint
}
## modifications specific to googledrive package
params$key <- key %||% params$key %||%
drive_api_key() %||% <BUILT_IN_DEFAULT_API_KEY>
if (!is.null(ept$parameters$supportsAllDrives)) {
params$supportsAllDrives <- TRUE
}
req <- gargle::request_develop(endpoint = ept, params = params)
gargle::request_build(
path = req$path,
method = req$method,
params = req$params,
body = req$body,
token = token
)
}
googledrive::request_generate()
is a thin wrapper around
gargle::request_develop()
and
gargle::request_build()
that only implements details
specific to googledrive, before delegating to more general functions in
gargle. The vignette("request-helper-functions")
documents
these gargle functions.
googledrive::request_generate()
gets a token with
drive_token()
, which is defined like so:
# googledrive::
drive_token <- function() {
if (isFALSE(.auth$auth_active)) {
return(NULL)
}
if (!drive_has_token()) {
drive_auth()
}
httr::config(token = .auth$cred)
}
where drive_has_token()
in a helper defined as:
By default, auth is active, and, for a fresh start, we won’t have a
token stashed in .auth
yet. So this will result in a call
to drive_auth()
to obtain a credential, which is then
cached in .auth$cred
for the remainder of the session. All
subsequent calls to drive_token()
will just spit back this
token.
Above, we discussed scenarios where an advanced user might call
drive_auth()
proactively, with non-default arguments,
possibly even loading a service token or using alternative flows, like
an external account. Any token loaded in that way is stashed in
.auth$cred
and will be returned by subsequent calls to
drive_token()
.
Multiple gargle-using packages can use a shared token by obtaining a suitably scoped token with one package, then registering that token with the other packages. For example, the default scope requested by googledrive is also sufficient for operations available in googlesheets4. You could use a shared token like so:
library(googledrive)
library(googlesheets4)
drive_auth(email = "jane_doe@example.com") # gets a suitably scoped token
# and stashes for googledrive use
gs4_auth(token = drive_token()) # registers token with googlesheets4
# now work with both packages freely ...
It is important to make sure that the token-requesting package (googledrive, above) is using an OAuth client for which all the necessary APIs and scopes are enabled.
The exported functions like drive_auth()
,
drive_token()
, etc. constitute the auth interface between
googledrive and gargle and are centralized in tidyverse/googledrive/R/drive_auth.R
.
That is a good template for how to use gargle to manage auth in a client
package. In addition, the docs for these gargle-backed functions are
generated automatically from standard information maintained in the
gargle package.
drive_token()
retrieves the current credential, in a
form that is ready for inclusion in HTTP requests. If
auth_active
is TRUE
and cred
is
NULL
, drive_auth()
is called to obtain a
credential. If auth_active
is FALSE
,
NULL
is returned; client packages should be designed to
fall back to including an API key in affected HTTP requests, if sensible
for the API.drive_auth()
ensures we are dealing with an
authenticated user and have a credential on hand with which to place
authorized requests. Sets auth_active
to TRUE
.
Can be called directly, but drive_token()
will also call it
as needed.drive_deauth()
clears the current token. It might also
toggle auth_active
, depending on the features of the target
API. See below.drive_oauth_client()
returns
.auth$client
.drive_api_key()
returns
.auth$api_key
.drive_auth_configure()
can be used to configure auth.
This is how an advanced user would enter their own OAuth client and API
key into the auth config, in order to affect all subsequent
requests.drive_user()
reports some information about the user
associated with the current token. The Drive API offers an actual
endpoint for this, which is not true for most Google APIs. Therefore the
analogous function in bigrquery, bq_user()
is a better
general reference.APIs split into two classes: those that can be used, at least partially, without a token and those that cannot. If an API is usable without a token – which is true for the Drive API – such requests must include an API key. Therefore, the auth design for a client package is different for these two types of APIs.
For an API that can be used without a token:
drive_deauth()
can be used at any time to enter a
de-authorized state. It sets auth_active
to
FALSE
and .auth$cred
to NULL
. In
this state, requests are sent out with an API key and no token. This is
a great way to eliminate any friction re: auth if there’s no need for
it, i.e. if all requests are for resources that are world readable or
available to anyone who knows how to ask for it, such as files shared
via “Anyone with the link”. The de-authorized state is especially useful
in non-interactive settings or where user interaction is indirect, such
as via Shiny.
For an API that cannot be used without a token: BigQuery is an
example of such an API. bq_deauth()
just clears the current
token, so that the auth flow starts over the next time a token is
needed.
Advanced users can use their own OAuth client and API key.
drive_auth_configure()
lives in R/drive_auth.R
and it provides the ability to modify the current client
and api_key
. Recall that drive_oauth_client()
and drive_api_key()
also exist for targeted, read-only
access.
The vignette("get-api-credentials")
describes how to get
an API key and OAuth client.
Packages that always send a token will omit the API key functionality here.
One reason for a user to call drive_auth()
directly and
proactively is to switch from one Google identity to another or to make
sure they are presenting themselves with a specific identity.
drive_auth()
accepts an email
argument, which
is honored when gargle determines if there is already a suitable token
on hand. Here is a sketch of how a user could switch identities during a
session, possibly non-interactive:
library(googledrive)
drive_auth(email = "janedoe_work@gmail.com")
# do stuff with Google Drive here, with Jane Doe's "work" account
drive_auth(email = "janedoe_personal@gmail.com")
# do other stuff with Google Drive here, with Jane Doe's "personal" account
drive_auth(path = "/path/to/a/service-account.json")
# do other stuff with Google Drive here, using a service account