patch, 2 partssend-email: add --mailmap support

I recently sent a series to enable mailmap support in format patch. The discussion led me to realize that the true problem we wanted solved is to map addresses at send time, so that we do not accidentally include a dead mail address when sending an old change.

Instead of worrying about what the formatted patch has, this series implements support for mailmap at the send-email, which will translate all addresses, and not just the author/commit addresses for a patch, but also the email for any trailers.o

I considered also that it may be useful to make send-email read a mailmap file from the identity config blocks, but have not figured out how to implement this yet.

I ended up needing to extend git check-mailmap to handle addresses without the angle brackets as well.

I think this is closer to solving the actual problem we have, which is wanting to avoid adding dead email addresses for coworkers who have moved on, but without completely removing their name from the works.

Link to previous: https://lore.kernel.org/r/20240813-jk-support-mailmap-git-format-patch-v1-1-1aea690ea5dd@gmail.com

Signed-off-by: Jacob Keller <jacob.keller@gmail.com>
---
Jacob Keller (2):
      check-mailmap: add --no-brackets mode
      send-email: add support for --mailmap

 builtin/check-mailmap.c             | 27 ++++++++++++-----
 Documentation/git-check-mailmap.txt |  8 ++++-
 git-send-email.perl                 | 14 +++++++++
 t/t4203-mailmap.sh                  | 60 +++++++++++++++++++++++++++++++++++++
 t/t9001-send-email.sh               | 49 ++++++++++++++++++++++++++++++
 5 files changed, 149 insertions(+), 9 deletions(-)
---
base-commit: 87a1768b93a67d0420255a43d9e07387b2e805ad
change-id: 20240816-jk-send-email-mailmap-support-1a9e86867c72

Best regards,

From: Jacob Keller <jacob.keller@gmail.com>

The git check-mailmap command can be used to convert identities to their canonical real name and email address. Currently, if a simple email address is provided without surrounding angle brackets, git check-mailmap will fail:

  $ git check-mailmap test@example.com
  fatal: unable to parse contact: test@example.com

This is generally fine since identifies are expected to be of the form "name <email@domain>". However, requiring brackets around simple email addresses can make it difficult to support mailmap operation in other environments where angle brackets may be missing.

Specifically, attempting to support the mailmap within git send-email is tricky, since angle brackets are not always provided for addresses.

Teach check-mailmap a new '--no-brackets' mode. In this mode, any contact line which cannot be interpreted by split_ident_line is treated as a simple address without a name. In addition, when any contact does not have a name, output the mail address without the angle brackets. Note that angle brackets are accepted if they are present, however the output will strip them.

This mode will be useful for git send-email in a following feature implementation to enable mapping any email addresses to their canonical value.

Signed-off-by: Jacob Keller <jacob.keller@gmail.com>
---
 builtin/check-mailmap.c             | 27 ++++++++++++-----
 Documentation/git-check-mailmap.txt |  8 ++++-
 t/t4203-mailmap.sh                  | 60 +++++++++++++++++++++++++++++++++++++
 3 files changed, 86 insertions(+), 9 deletions(-)

On Fri, Aug 16, 2024 at 7:06 PM Jacob Keller <jacob.e.keller@intel.com> wrote:

s/identifies/identities/

What is not explained by the commit message is why we need this new option as opposed to merely making git-check-mailmap accept such addresses by default. Are there difficulties or downsides to making this the default behavior? Do other things break if this new behavior becomes the default as opposed to being an explicit opt-in?

I'm a bit skeptical about how this new flag also changes the output format to suppress the angle brackets. It seems like that's something the caller could do easily enough if desired. For instance:

    $addr =~ s/^<(.*)>$/\1/;

Or, is there some deeper reason for doing this?

On 8/16/2024 4:22 PM, Eric Sunshine wrote:

Mostly I did it this way out of conservative caution to avoid breaking existing users. It could be that nothing breaks and loosening the restriction on what we pass it would be sufficient.

For one, the documentation of git check-mailmap specifies that it passes other values through "as-is", but this mode would convert "user@example.com" to "<user@example.com>"

I suppose thats not a big deal, and it was more a matter of not wanting to bother the caller to force it to strip any brackets if it didn't want them. I think its probably fine to require that.

On Fri, Aug 16, 2024 at 7:42 PM Jacob Keller <jacob.e.keller@intel.com> wrote:

I figured that was probably the reason for making this an opt-in.

I think my somewhat negative reaction to this new option -- and the reason I asked the above question -- is that it feels very special-case, thus it lacks generality (or at least the way the commit message sells it, it makes it difficult to see it as anything other than a bandaid added just to support mailmap in git-send-email).

From: Jacob Keller <jacob.keller@gmail.com>

In certain cases, a user may be generating a patch for an old commit which now has an out-of-date author or other identity. For example, consider a team member who contributes to an internal fork of a project, and then later leaves the company.

It may be desired to submit this change upstream, but the author identity now points to an invalid email address which will bounce. This is likely to annoy users who respond to the email on the public mailing list.

This can be manually corrected, but requires a bit of effort, as it may require --suppress-cc or otherwise formatting a patch separately and manually removing any unintended email addresses.

Git already has support for the mailmap, which allows mapping addresses for old commits to new canonical names and addresses.

Teach git send-email the --mailmap option. When supplied, use git check-mailmap (with the --no-brackets mode) as a final stage when processing address lists. This will convert all addresses to their canonical name and email according to the mailmap file.

A mailmap file can then be configured to point the invalid addresses either to their current canonical email (if they still participate in the open source project), or possibly to new owner within the company.

This enables the sender to avoid accidentally listing an invalid address when sending such a change.

Signed-off-by: Jacob Keller <jacob.keller@gmail.com>
---
 git-send-email.perl   | 14 ++++++++++++++
 t/t9001-send-email.sh | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 63 insertions(+)

On Fri, Aug 16, 2024 at 7:06 PM Jacob Keller <jacob.e.keller@intel.com> wrote:

Nit: The final two paragraphs appear to repeat what was already stated
or implied earlier, thus don't seem to add any value to the commit
message.

Nit aside, similar to the question I asked about [1/2], are there downsides to merely enabling this new behavior by default? It seems like it would be generally desirable to have this translation happen by default, so making everyone opt-in may be a disservice. On the other hand, starting out with it disabled by default is understandable as a cautious first step, though it might be nice to explain that in the commit message. Similarly, one can imagine a world in which people want to enable this and forget about it, thus would like it to be controlled by configuration (though that can, of course, be left for a future change).

For some reason, I found this logic more difficult to follow than expected, possibly because it doesn't feel quite Perlish, or possibly because in this codebase, we often take care of the easy cases first and return early. Thus, I may have been expecting the above to be written more along the lines of:

    sub mailmap_address_list {
        return @_ unless @_ && $mailmap;
        return Git::command('check-mailmap', '--no-brackets', @_);
    }

Of course, it's highly subjective and not at all worth a reroll.

On 8/16/2024 4:41 PM, Eric Sunshine wrote:

Sure, I think I got a bit verbose here.

I definitely did it mostly out of conservative: "don't change the default behavior".

For a general mailmap I think enabling it by default, with a config option to disable it. I think it might also make sense to have per-identity configuration, so that different identities could point at a different mailmap.

For example, one of the use cases we have is to have a mailmap file that takes the now-invalid addresses and points them all to the current maintainer. This way, the maintainer (who is sending these patches) would have all the old addresses automatically point to him instead of generating the bounced messages as currently happens on accident. This type of mailmap file likely does not make sense as a general-purpose file.

I think for per-identity configurations, we would either need to add an option to git check-mailmap to pass the mailmap file, or I would need to figure out how to set config when calling Git::command() to force a specific mailmap.

Ah, yea that makes more sense. I had been trying to figure that out but I am not as used to the unless syntax.