Re: [PATCH v2 4/6] imap-send: make --curl no-optional

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Thu, Feb 02 2023, Junio C Hamano wrote:

> Ævar Arnfjörð Bjarmason  <avarab@gmail.com> writes:
>
>> In the preceding commit the old "USE_CURL_FOR_IMAP_SEND" define became
>> always true, as we now require libcurl for git-imap-send.
>>
>> But as we require OpenSSL for the "tunnel" mode we still need to keep
>> the OpenSSL codepath around (ee [1] for an attempt to remove it). But
>
> "(ee" -> ???

Should be "e.g.", will fix.

>> we don't need to keep supporting "--no-curl" to bypass the curl
>> codepath for the non-tunnel mode.
>
> We do not need to because...?

We don't have that code anymore, will clarify.

>> @@ -1519,12 +1519,8 @@ int cmd_main(int argc, const char **argv)
>>  	if (argc)
>>  		usage_with_options(imap_send_usage, imap_send_options);
>>  
>> -#if defined(NO_OPENSSL)
>> -	if (!use_curl) {
>> -		warning("--no-curl not supported in this build");
>> -		use_curl = 1;
>> -	}
>> -#endif
>> +	if (!use_curl)
>> +		die(_("the --no-curl option to imap-send has been deprecated"));
>
> We used to force use of cURL when there is no other way to make the
> program work (i.e. there is no direct OpenSSL codepath available),
> instead of refusing to work (and forcing user to say --curl or to
> stop saying --no-curl, which is one unnecessary roadblock for the
> user).  Why do we want to change the error handling strategy that
> has been in place?

I can change this to a soft error, but it seemed more sensible to rip
the band-aid off an option that's never going to do anything now,
whereas before it would do something based on how you compiled git.

> I think I made the same comment in some other thread, but the
> principle is the same.  If there is no other choice the user can
> take, do we force users to stop and be explicit to choose that only
> available choice, or do we let the program choose the only available
> option for the user while clearly telling the user that is what we
> did?  Here, changing the behaviour sounds like a disservice to the
> users.

At best we can make --no-curl use curl anyway with a warning, would that
be better?