Permalink
Find file
Fetching contributors…
Cannot retrieve contributors at this time
98 lines (54 sloc) 12.3 KB

Portier and Other Projects

Portier considers itself a spiritual successor to Mozilla Persona. Though elements of Portier's design appear in other projects, the specific combination of starting with an email address, attempting dynamic discovery, and then failing over to traditional email confirmation appears to be unique to Portier and Persona.

Email-First, Connected Systems

Poetica had an excellent email-first, connected login flow prior to its acquisition by Condé Nast: Users would enter an email address and then see a filtered menu of compatible authentication options. For example, Google Sign-In would appear for people using Gmail, but not for those using Yahoo! Mail.

Portier follows this user experience, but like Persona, automatically selects the most specific authentication method available, rather than forcing the user to choose (and remember) a strategy.

Magic Links

Many contemporary websites support login-via-email workflows:

These workflows are all email-only. In contrast, Portier first attempts to use web-based, federated protocols before falling back to a magic link.

Confirmation Codes

Many services send information to one device in order to verify interactions on another: Two-Factor Authentication (2FA) solutions often work by texting a short code that they must transcribe before users are allowed to sign in. Digits, by Twitter, offers this style of SMS-based authentication as a service.

Portier uses SMTP, but the principle and user experience are comparable: a user can prove control of an email address via a code or link sent only to that address. The ability to manually transcribe codes is important, since it eases authentication for users with smartphones: glance at the phone, copy the code, done.

Nevertheless, web-based authentication with a federated protocol is still preferable when available, since it eliminates context switches for the user.

Enterprise-Ready Solutions

CoreOS's open source Dex and Red Hat's Keycloak (sold as "Red Hat Single Sign-On") offer more extensive, enterprise-level feature sets.

These projects may serve as inspiration for future work on Portier.

Portier Compared to Persona

Note: Initial interest in Portier will likely come from individuals familiar with the Persona project, hence the detailed retrospective below. In the following section, "we" refers to the Persona core team.

Though imbued with Persona's spirit, Portier is decidedly less ambitious and more narrowly scoped.

Where Persona Failed

Persona made choices that, in retrospect, were overly ambitious or even contradictory to its goal of solving authentication on the Web:

  • Persona had a user-visible brand. We encouraged buttons labeled "Sign-in with Persona," which confused users. In practice, Persona was just infrastructure, and should have been (non-)branded as such. "Sign-in with Email" better described Persona's function and requirements.
  • Persona had conflicting missions. While Persona was initially designed to solve authentication for the Web, the team was later tasked with additionally solving authentication for Firefox OS handsets and Mozilla's corporate infrastructure using the same code. This required adding opt-in centralization and mandatory passwords to a decentralized, password-optional protocol. Other features of centralized systems, including support for unverified email addresses and forced invalidation of federated credentials, were also implemented. This added complexity slowed Persona's development.
  • Persona tried to protect users from their own email providers. While laudable, meaningful privacy from a user's own email provider is an impossible goal that unnecessarily constrained Persona's design: whenever a website sends an email, that privacy is violated. Instead, users who are concerned about such privacy are empowered to choose providers that they trust, or run their own provider, without any degradation in user experience.
  • Persona subverted websites' session management. If a user's session expired on a website, Persona would forcibly log back in. While well intentioned, this expanded Persona's scope and intrusively overrode the site's own session management. Instead of serving developers, Persona forced websites to conform to it. Session management also had poor failure modes which could trap users in an infinite loop of perpetually failing login attempts.

In addition to high-level scope creep, we made tactical decisions that did not succeed:

  • Persona had its own, broken account system. While designed to reduce friction, Persona's account system introduced a point of centralization in an otherwise decentralized system and created complex state interactions that proved to be more of a hindrance than convenience.

    For example, each email address could only belong to a single Persona account. This made using shared addresses extremely difficult, and it was even possible to accidentally split your own addresses between two accounts with no clear way to merge them.

  • Persona had its own passwords. Because email confirmation loops were considered onerous, we asked users to complete a traditional confirmation loop once per address and establish a Persona-specific password for future interactions. This centralized password meant that users could still authenticate after losing access to an email address, breaking Persona's promise of verified identities.

    This also created user experience challenges: if an email provider added support for the BrowserID protocol, Persona would use that instead of passwords. Later, if they removed support for BrowserID, Persona would suddenly resume asking users for a password they had not used in months.

  • Persona had a long memory. Persona remembered every time it saw an email provider that supported the BrowserID protocol. If that provider later removed support for BrowserID, Persona would wait a week before falling back to its own email confirmation system, under the presumption that the domain was having a temporary technical problem. This ensured that Persona never assumed undue authority over a domain, but also meant that users could be locked out of websites for a week.

    This also complicated planning for providers wanting to add support for BrowserID: Persona's long-lived state meant that reverting a bad deployment could inadvertently trigger a week-long lockout.

  • The BrowserID protocol was all-or-nothing. Because the protocol worked at the domain level, it wasn't possible for providers to turn on support for a subset of users before rolling it out to everyone at the domain. This made deployments and upgrades riskier than necessary.

  • Persona used pop-ups for its UI. This had the benefit of preserving page state during log-in. Unfortunately, pop-ups are easily lost when juggling windows, hard to associate with a specific tab, and many users have a reflexive aversion to pop-ups, closing the Persona window before it could load. There were also technical issues with pop-up blockers and mainstream browsers like Chrome for iOS that did not support the window.opener property for communicating back to the parent tab.

  • Persona relied on unfinished specifications. Persona was an early adopter of the JSON Web Token (JWT) standard and associated JOSE specifications. Unfortunately, the specifications were not yet stable, and made frequent, backwards incompatible changes. As a result, we had to maintain Persona-specific libraries, rather than using third party libraries that tracked the current versions of the specifications.

  • Persona required third party cookies. In the absence of native browser support, interactions with Persona happened via invisible iframes rooted at external origins. Privacy-conscious users often manually disable browser support for third party cookies, or use browser add-ons that do the same. This configuration rendered Persona inoperable.

  • Persona overemphasized native browser support. There are simply too many platforms and too much bureaucratic inertia to effectively pursue native implementation in a majority of browsers. The Web platform has no such limitations. Once we realized this, we pivoted to a Web-first strategy, treating browser support as an optional enhancement.

  • Persona was impossible to self-host or fork. We inadvertently designed intractable points of centralization into Persona, some of which we only fully recognized when we tried to map out a future for Persona independent of Mozilla.

    For example, in the absence of native browser support, identity providers and websites using Persona had to use a third party service to relay messages across origins. Both parties had to select the same relay ahead of time, without ever communicating, otherwise the protocol would stall and fail. We "solved" this by expecting everyone to use the same, Mozilla-operated relay. However, reliance on this service meant that without native browser support, Persona could not be decentralized: it would fracture into mutually incompatible subnetworks.

The Fallacy of the Three-Way Cold Start

We often cite a three-way cold start between browser vendors, websites, and identity providers for Persona's failure. In retrospect, this incorrectly evaluates Persona as a consumer product that needed growth, visibility, and network effects to be considered successful, rather than as infrastructure that simply needed to be reliable and independently useful.

When websites rejected Persona because it didn't have enough users or name recognition, we should have realized that we had mis-positioned Persona as a social authentication button. We were pursuing the wrong goals.

Similarly, the emphasis on native browser support in the hypothetical three-way cold start is misplaced: Persona targeted the standard Web platform. It worked well in all browsers, and for all email addresses. Why, then, would adoption be contingent on the marginal UX improvements associated with native support? Not to mention that Mozilla could have independently shipped support in Firefox and broken this assumed deadlock. That we consciously chose to focus on other tasks indicates that we did not truly believe native support was critical to Persona's success.

What Persona Got Right

Persona did get many things right, which Portier preserves:

  • Email as Identity. Email is universal, decentralized, and pseudonymous. Most sites already collect it during registration, it is commonly used as a natural key for accounts, and it allows for out-of-band communication with users.
  • Protocol Bridges. Developing bespoke integrations with providers like Gmail offers an immediately improved experience to hundreds of millions of people. Simultaneously, it serves as a direct replacement for existing social authentication buttons on websites, but without the UI clutter, API keys, or multiple protocols.
  • Decentralization. To maintain its sovereignty, the Web needs a modern authentication system that isn't controlled by or reliant on a single, for-profit, American corporation like Facebook, Google, Twitter, or GitHub.
  • Self-Certification. Similarly, to safeguard user choice on the Web, domains must be able to certify their own users.
  • Web First. The Web is the only universal platform. Solutions must, first and foremost, work well on the Web.
  • Hosted Verifier. Persona offered a hosted service that would handle the more complex steps of verifying identity certificates. This may not be necessary now that the JWT spec has stabilized, but the principle of delegating computation to an optional, hosted service significantly lowered Persona's barrier of adoption.