Re: [PATCH] git-send-email: add sendmailCommand option

[Gregory Anders <greg@gpanders.com>]

On Wed, 12 May 2021 11:04 +0200, Ævar Arnfjörð Bjarmason wrote:
>
>On Tue, May 11 2021, Gregory Anders wrote:
>
>> diff --git a/Documentation/git-send-email.txt b/Documentation/git-send-email.txt
>> index 93708aefea..d9fe8cb7c0 100644
>> --- a/Documentation/git-send-email.txt
>> +++ b/Documentation/git-send-email.txt
>> @@ -159,13 +159,23 @@ Sending
>>  ~~~~~~~
>>
>>  --envelope-sender=<address>::
>> -	Specify the envelope sender used to send the emails.
>> -	This is useful if your default address is not the address that is
>> -	subscribed to a list. In order to use the 'From' address, set the
>> -	value to "auto". If you use the sendmail binary, you must have
>> -	suitable privileges for the -f parameter.  Default is the value of the
>> -	`sendemail.envelopeSender` configuration variable; if that is
>> -	unspecified, choosing the envelope sender is left to your MTA.
>> +	Specify the envelope sender used to send the emails.  This is
>> +	useful if your default address is not the address that is
>> +	subscribed to a list. In order to use the 'From' address, set
>> +	the value to "auto". If you use the sendmail binary, you must
>> +	have suitable privileges for the -f parameter.  Default is the
>> +	value of the `sendemail.envelopeSender` configuration variable;
>> +	if that is unspecified, choosing the envelope sender is left to
>> +	your MTA.
>
>Please don't include word-wrapping for unrelated changes in the main
>patch.

My mistake, this has been pointed out to me multiple times now. I'll 
remove it in the next revision and I'll be sure to avoid this in the 
future.

>
>> -	$smtp_server ||= 'localhost'; # could be 127.0.0.1, too... *shrug*
>> +
>> +	if (!defined $sendmail_command) {
>> +		$smtp_server = 'localhost'; # could be 127.0.0.1, too... *shrug*
>> +	}
>>  }
>
>This "let's not accept a 0" change seems unrelated & should be split
>into a prep cleanup / refactoring patch. On the one hand it's sensible,
>on the other nobody cares about having a command named "0" in their path
>(or a hostname), so I think it's fine to have the ||= Perl idiom leak
>out here.
>
>But also, this just seems like confusing logic. Per your docs "your
>sendmailCommand has precedence over smtpServer.".
>
>Why not make this "if not $sendmail_command" part of the top-level check
>here (the if this one is nested under), which is only done if
>$smtp_sever is not defined, if $sendmail_command is defined we don't
>care about $smtp_server later on, no?

I mostly left this the way it is to minimize the diff, as this is the 
style the code was already written in. I agree that explicitly checking 
whether sendmail_command is undefined is probably clearer to the reader.

>
>>  if ($compose && $compose > 0) {
>> @@ -1490,14 +1497,15 @@ sub send_message {
>>
>>  	unshift (@sendmail_parameters, @smtp_server_options);
>>
>> +	if (file_name_is_absolute($smtp_server)) {
>> +		# Preserved for backward compatibility
>> +		$sendmail_command ||= $smtp_server;
>> +	}
>> +
>>  	if ($dry_run) {
>>  		# We don't want to send the email.
>> -	} elsif (file_name_is_absolute($smtp_server)) {
>> -		my $pid = open my $sm, '|-';
>> -		defined $pid or die $!;
>> -		if (!$pid) {
>> -			exec($smtp_server, @sendmail_parameters) or die $!;
>> -		}
>> +	} elsif (defined $sendmail_command) {
>> +		open my $sm, '|-', "$sendmail_command @sendmail_parameters";
>
>Can we really not avoid moving from exec-as-list so Perl quotes
>everything, to doing our own interpolation here? It looks like the tests
>don't check arguments with whitespace (which should fail with this
>change).
>

Shell interpolation in this case is considered a feature, not a bug, 
i.e. we want to provide users the ability to use arbitrary shell 
expressions (as they do in e.g. aliases) or pass arguments. I also 
modeled this after the implementation for --to-cmd and --cc-cmd, which 
also run their respective commands in the shell just like this.

Also, this *did* cause the tests to fail since the tests write output to 
a path with a space in it. You'll notice that in the diff for the tests 
I had to wrap '$(pwd)/fake.sendmail' in additional quotes to resolve 
this.

>> @@ -1592,14 +1600,14 @@ sub send_message {
>>  		printf($dry_run ? __("Dry-Sent %s\n") : __("Sent %s\n"), $subject);
>>  	} else {
>>  		print($dry_run ? __("Dry-OK. Log says:\n") : __("OK. Log says:\n"));
>> -		if (!file_name_is_absolute($smtp_server)) {
>> +		if (defined $sendmail_command) {
>> +			print "Sendmail: $sendmail_command ".join(' ',@sendmail_parameters)."\n";
>> +		} else {
>>  			print "Server: $smtp_server\n";
>>  			print "MAIL FROM:<$raw_from>\n";
>>  			foreach my $entry (@recipients) {
>>  			    print "RCPT TO:<$entry>\n";
>>  			}
>> -		} else {
>> -			print "Sendmail: $smtp_server ".join(' ',@sendmail_parameters)."\n";
>>  		}
>>  		print $header, "\n";
>>  		if ($smtp) {
>
>Minor nit: Let's just continue to use "if (!" here to keep the diff
>minimal or split up such refactoring into another change...

Sure, I can do that.

Thanks for your feedback.