If not in an interactive session, can gargle_oauth_email() default to TRUE?

[MarkEdmondson1234]

To help with non-interactive sessions, in the context of the advice on the website here:

Could this be made even smoother by detecting if in an interactive session or not (e.g. if(interactive()), and if FALSE defaulting the gargle_oauth_email = TRUE option?

It will cover 95% of the use cases I've seen (where a user has one cache token on a server for scheduled tasks) and prevent a potential hangup when the option is not set and the interactive menu comes up. I favour the JSON service email option instead, but that usually involves adding the service email to accounts that may not be possible.

[jennybc]

Hmm, I'll have to think about it. I totally agree it would remove a lot of failure. The question is: should we? The original plan was to be exceedingly careful -- maybe even, pedantic -- about getting the user's consent to use a token that we find.

[MarkEdmondson1234]

If not, perhaps the error returned could give a hint on what has happened. "Do you need to set option(gargle_oauth_email) ?" I'll try it on a non-interactive script to see what happens now, for httr it was a cryptic error about not connecting to a server IIRC.

[MarkEdmondson1234]

The most pertinent example of this is when attempting auth in an RMarkdown doc - if not set correctly then the rendering just hangs, with no error, which is mega-confusing for folks not intimate with the auth process. For googleAuthR I've put in the below thats tested before the call to token_fetch() but I reckon something similar would be a good addition in gargle too:

  # to aid non-interactive scripts
  if(is.null(email) 
     && !interactive() 
     && is.null(getOption("gargle_oauth_email"))){
    stop("Non-interactive session and no authentication email selected.
         \nSetup JSON service email auth or specify email in gar_auth(email='me@preauthenticated.com'", call.=FALSE)
  }

[EricGoldsmith]

+1 for this request. The current behavior is a breaking change from httr OAuth cache handling.

[jennybc]

Yeah, maybe we could check if interactive or for an env var that suggests we're knitting and act as if email = TRUE, but warn that the user needs to specify this clearly in code to suppress the warning.

[jennybc]

Also, just to spread the knowledge, this situation is addressed specifically in the Non-interactive auth vignette: https://gargle.r-lib.org/articles/non-interactive-auth.html#i-just-want-my--rmd-to-render

[EricGoldsmith]

Yeah, that vingette was how I found the workaround/solution.

In addition to the use-case that Mark mentioned, I have several anomaly detection and reporting scripts that run from cron/launchd that rely on cached credentials. Since upgrading recently to the packages that use gargle, I've had to make code changes (adding options(gargle_oauth_email = TRUE)) to restore the prior (transparent) credential caching behavior.

So, I'm very much in favor of making email = TRUE the default for non-interactive sessions. But it's also a convenience for interactive sessions, to mimic the prior behavior.

To net this out, I'd suggest that if there's only one credential in the cache, use it by default. Only prompt the user if there's more than one to choose from (but still allow this to be overridden via options(gargle_oauth_email = "jenny@example.org")).

Perhaps I'm being too prescriptive here, based on my use-cases. With all the work you've done on various packages, I'm sure you have a more informed view of the landscape. So, I'll just stick with relaying my desire for backward compatibility with httr's cached credential use behavior.

[jennybc]

Our increased commitment to getting permission to use user creds is due to an overall awareness of the need to be really transparent when accessing and using sensitive data. I think jobs running on remote servers should probably be using service account tokens, for example. OAuth2 user tokens are really designed for use by actual users, hence our design choices. I think the level of magic in httr is partially responsible for lots of people pushing OAuth2 creds to GitHub, because they have so little awareness of them. I will not link to this 😱 search, but let's just say there are several hundreds of these.

So I agree we could switch to a "use and warn" strategy. But I'd really like these remote-running scripts to contain code that indicates clear permission to use auto-discovered user credentials.

[MarkEdmondson1234]

For me, the warning or even a stop() would solve the problem of sitting there wondering why the script isn't working. It is a small change once you know what it is, the main issue I think is diagnosing what the problem is.

That way individual packages can decide how much they want to support OAuth2 tokens vs nudging to JSON creds. For the reasons mentioned above I removed OAuth2 support for compute engine/cloud storage as it wasn't worth it, but for the more public APIs there is less critical stuff at risk (e.g.not stuff you can be charged for)

[jennybc]

Quick note to self: succeed but warn

Update: I can't literally warn, because of tryCatch() in token_fetch(), but I can emit some messaging.

[jennybc]

@MarkEdmondson1234 @EricGoldsmith Please try this out and let me know if you have any feedback.

[EricGoldsmith]

I appreciate the compromise. Thanks.

[EricGoldsmith]

Follow-up question ...

With gargle v0.4.0, I get the following warning:

Using an auto-discovered, cached token.
To suppress this message, modify your code or options to clearly consent to the use of a cached token.
See gargle's "Non-interactive auth" vignette for more details:
https://gargle.r-lib.org/articles/non-interactive-auth.html
The gmailr package is using a cached token for <my email address>.

But, the first line of my code is:

options(gargle_oauth_email = TRUE)

which is cited as a solution to the warning in the vignette referenced in the message.

I want to make sure I'm not misunderstanding the warning or the docs before opening a new issue for this.

[jennybc]

Yeah, at the moment, I only suppress the warning message when the user gives a specific email address, continuing to be very conservative. I could be persuaded that the generic TRUE is sufficient. Do you have any specific reason to use TRUE instead of an actual email here?

[m-macaskill]

I've just run into the same issue as @EricGoldsmith, of failing to suppress this warning despite having taken the suggested route of setting the email argument to TRUE.

Do you have any specific reason to use TRUE instead of an actual email here?

In our case, we have R Markdown reports that a number of users can access and run locally, either via their local version of a shared Google folder, or by downloading from Github. Therefore we can't specify an individual e-mail address in the code, as that would fail for all but one user. Conversely, setting email=TRUE would work for all, provided that they have each authenticated beforehand.

[jennybc]

If they have multiple tokens on hand, we take the "first" one, whatever that means. Can you be sure that is always correct? This is the general background on why I'm hesitant to be completely silent when email = TRUE.

I tend to think that your workflow @m-macaskill suggests agreeing on a convention for what you want to call the correct Google identity and then consult an option or env var for it.

Very related to #118

[MarkEdmondson1234]

Yes my current workflow for what Michael describes is to set option(gargle_email = Sys.getenv(“GARGLE_EMAIL”)) then each user can set it within their session. It’s a bit rough if the env arg has forgot to be set, (so being “”) perhaps an error if not TRUE or a valid email would be nice too.

…

________________________________ From: Jennifer (Jenny) Bryan <notifications@github.com> Sent: Monday, February 24, 2020 6:21 PM To: r-lib/gargle Cc: Mark; Mention Subject: Re: [r-lib/gargle] If not in an interactive session, can gargle_oauth_email() default to TRUE? (#92) If they have multiple tokens on hand, we take the "first" one, whatever that means. Can you be sure that is always correct? This is the general background on why I'm hesitant to be completely silent when email = TRUE. I tend to think that your workflow @m-macaskill<https://github.com/m-macaskill> suggests agreeing on a convention for what you want to call the correct Google identity and then consult an option or env var for it. Very related to #118<#118> — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#92?email_source=notifications&email_token=AAYCPLEUUJC5AZVM6WMOZM3REP6Y7A5CNFSM4H6VSOIKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEMYYLCA#issuecomment-590448008>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAYCPLA67PNQ5KR5KFR6QZTREP6Y7ANCNFSM4H6VSOIA>.

[jennybc]

It’s a bit rough if the env arg has forgot to be set, (so being “”) perhaps an error if not TRUE or a valid email would be nice too.

Care to open an issue for that? That is easy to fix.