[RFC] A global mailmap service

I came up with the idea of creating a global mailmap service earlier this year and, given a recent discussion on maintaining .mailmap, I decided to bring it up here. While only marginally related to Git development, I hope that it is relevant enough to not be considered spam.

The basic idea of the service I imagine is simple:

1. You register a primary email address and specify a password. You
   receive a verification email to confirm that the address is yours.

2. At any time, you can add additional email addresses and link them to
   your primary email address, using your previously specified password.
   You can also update your primary email address. Any new addresses
   obtain verification emails such that you cannot steal somebody else's
   identity.

3. Anybody can use a public lookup interface to obtain the current
   primary email address corresponding to any registered email address
   they enter.

According to the principle of data economy, the full list of email addresses is kept private. An email address is only returned if the user performing a lookup already owns a (possibly outdated) email address of the same user.

A batch query to the service can be used to automatically generate a .mailmap file without having to maintain it on a per-project basis and without having to be careful and confirm every entry manually.

I created a PoC here [1]. You can run

    git log --pretty='%ae' | sort -u | curl -Ftopic=git -Femails='<-' https://mailmap.org/

from the Git source tree to auto-generate a .mailmap file. Of course, this idea only works if people agree that it is useful and the majority of developers register their email addresses to the service.

I am aware that some users use different email addresses for different projects. The current approach is allowing the user to add primary "topic email addresses". A request can optionally specify a topic (such as "git" in the request above) and the lookup may potentially return an email address different from the generic primary address. Of course, these topics need to be standardized in some way. There might be better, more sophisticated ways to tackle this and I'd be happy to discuss them if there is interest in the general approach.

I am also aware that this may not cover all potential uses of .mailmap files but, according to my understanding and experience, it should cover a fairly large amount. The remaining adjustments can still be made on top on a per-project basis.

Best regards, Lukas

[1] https://mailmap.org/

Lukas Fleischer <lfleischer@lfos.de> writes:

I would do so with my current, reachable address, I'd presume.

With this, I won't be able to add my ancient identities that appear in our history. I would imagine that one of the common reasons people use different identities in a project is that people changed e-mail providers or jobs.

On Thu, 13 Dec 2018 at 03:17:21, Junio C Hamano wrote:

Correct.

Well, this is only a temporary issue. It holds for your current ancient identities but won't hold for your ancient identities in the far future because ancient times have always been present at some point in time. If we agreed that most people register their current email addresses from now on, we'd limit the issue to email addresses which were abandoned before 2018-12-13. As projects grow, this will become a small fraction. Projects started in the far future won't be affected at all.

As a short-term solution, we could keep the current mappings as complementary mappings in the local .mailmap files.

That being said, I also had the idea of importing old mappings to the service. There are two approaches that crossed my mind:

1. Link email addresses according to the .mailmap files of popular
   trusted projects, such as Git or the Linux kernel. One of the issues
   with this approach is that "topic email addresses", i.e. project
   specific preferences, may not be mapped correctly. Maybe it doesn't
   matter too much.

2. Add the option to register a "dead" email address to the service.
   Instead of sending a verification email, the service sends a warning
   stating that the email address will be linked if no action is taken.
   It contains a link where the user can cancel the request and block
   the email address from further requests. Another such warning is sent
   after a couple of days. After a grace period with no action taken,
   the address is linked. Not optimal but with additional measures such
   as rate limits to prevent from abuse, this might be good enough.

Best regards, Lukas