Gregory Anders <greg@xxxxxxxxxxxx> writes:
> Use a block scoped variable to print the sendmail invocation at the end
> of the 'send_message' subroutine. Assigning directly to $sendmail_cmd
> (as in the v2 patch) causes some bizarre problems; namely, it seems to
> affect the value of $sendmail_cmd that is read at earlier points in the
> same subroutine, which causes test invocations of the form
>
> git send-email --smtp-server="$(pwd)/fake.sendmail"
>
> to fail. The value passed to --smtp-server was assigned to $sendmail_cmd
> at the end of the 'send_message' subprocedure, but somehow this caused
> the 'if (defined $sendmail_cmd)' condition earlier in the subproc to
> evaluate to true.
Are you talking about the use of $sm that is local to the debug
output? I think leaving $sendmail_cmd intact by using a separate
variable is the right choice. Isn't the problem you observed a
consequence of send_message() getting called once for each message,
so assigning to $sendmail_cmd in the function for the first
invocation of the function would change its value for the second
invocation?
Also, if we have been using
--smtp-server=$(pwd)/fake.sendmail
we cannot expect to use the same value like this:
--sendmail-cmd=$(pwd)/fake.sendmail
because we deliberately add a space in the $(pwd) by choosing the
name of the test directory to be "trash directory.something". We'd
need to do something like
--sendmail-cmd='$(pwd)/fake.sendmail'
so that the shell sees '$(pwd)/fake.sendmail' literally and runs pwd
to find out what the path to the program is, I would think.
> diff --git a/t/t9001-send-email.sh b/t/t9001-send-email.sh
> index 65b3035371..583fbba410 100755
> --- a/t/t9001-send-email.sh
> +++ b/t/t9001-send-email.sh
> @@ -2148,6 +2148,37 @@ test_expect_success $PREREQ 'leading and trailing whitespaces are removed' '
> test_cmp expected-list actual-list
> '
>
> +test_expect_success $PREREQ 'test using command name with --sendmail-cmd' '
> + clean_fake_sendmail &&
> + PATH="$(pwd):$PATH" \
> + git send-email \
> + --from="Example <nobody@xxxxxxxxxxx>" \
> + --to=nobody@xxxxxxxxxxx \
> + --sendmail-cmd="fake.sendmail" \
> + HEAD^ &&
> + test_path_is_file commandline1
> +'
Nice demonstration of the "we no longer need an absolute pathname"
feature.
> +test_expect_success $PREREQ 'test using arguments with --sendmail-cmd' '
> + clean_fake_sendmail &&
> + git send-email \
> + --from="Example <nobody@xxxxxxxxxxx>" \
> + --to=nobody@xxxxxxxxxxx \
> + --sendmail-cmd="\"$(pwd)/fake.sendmail\" -f nobody@xxxxxxxxxxx" \
> + HEAD^ &&
> + test_path_is_file commandline1
> +'
Hmph, if $(pwd) has a double quote character in it, this may not
work as expected, as the shell that is expanding the command line
arguments for "git send-email" would see $(pwd), expand it and our
program will see
"/path/with/d"quote/git/t/trash directory.9001/fake.sendmail" -f nobody@e.c
as the value of --sendmail-cmd, which would not interpolate well,
no?
We want the shell that eats the command line of 'git send-email' to see
--sendmail-cmd='$(pwd)/fake.sendmail'\" -f nobody@xxxxxxxxxxx"
and because this is inside a sq pair, it would become
--sendmail-cmd='\''$(pwd)/fake.sendmail'\''\" -f nobody@xxxxxxxxxxx"
after we replace each sq with '\'', or something like that, perhaps?
> +test_expect_success $PREREQ 'test shell expression with --sendmail-cmd' '
> + clean_fake_sendmail &&
> + git send-email \
> + --from="Example <nobody@xxxxxxxxxxx>" \
> + --to=nobody@xxxxxxxxxxx \
> + --sendmail-cmd="f() { \"$(pwd)/fake.sendmail\" \"\$@\"; };f" \
> + HEAD^ &&
> + test_path_is_file commandline1
> +'
Nice demonstration of how a bit of scripting can be used.
> test_expect_success $PREREQ 'invoke hook' '
> mkdir -p .git/hooks &&
Thanks.
