Re: [PATCH v9 0/2] send-email: expose header information to git-send-email's sendemail-validate hook

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

On Mon, Jan 23 2023, Michael Strawbridge wrote:

> On 2023-01-23 08:51, Ævar Arnfjörð Bjarmason wrote:
  * Aside from that, shouldn't we have a new "validate-headers" or
>>    whatever hook, instead of assuming that we can add another argument
>>    to existing users?...
>
> While it's true we could (and I don't have a super strong opinion here),
> I suppose I was foreseeing the potential that a user may want to have
> logic that requires both the email headers and contents.  For example,
> only checking contents for a specific mailing list.  If we split the
> hooks, a user would then need to figure out how to have them coordinate.

...

>>
>>  * ...except can we do it safely? Now, it seems to me like you have
>>    potential correctness issues here. We call format_2822_time() to make
>>    the headers, but that's based on "$time", which we save away earlier.
>>
>>    But for the rest (e.g. "Message-Id" are we sure that we're giving the
>>    hook the same headers as the one we actually end up sending?
>>
>>    But regardless of that, something that would bypass this entire
>>    stdin/potential correctness etc. problem is if we just pass an offset
>>    to the the, i.e. currently we have a "validate" which gets the
>>    contents, if we had a "validate-raw" or whatever we could just pass:
>
> I think there might be a part missing here: "problem is if we just pass
> an offset to the ___."  So there's a chance I may not fully grasp your
> suggestion.

Sorry, a byte offset into the file to indicate the boundary between the
headers and the content.

>
>> 	<headers>
>> 	\n\n
>> 	<content>
>>
>>    Where the current "validate" just gets "content", no? We could then
>>    either pass the offset to the "\n\n", or just trust that such a hook
>>    knows to find the "\n\n".
>>
>>    I also think that would be more generally usable, as the tiny
>>    addition of some exit code interpretation would allow us to say "I
>>    got this, and consider this sent", which would also satisfy some who
>>    have wanted e.g. a way to intrecept it before it invokes "sendmail"
>>    (I remember a recent thread about that in relation to using "mutt" to
>>    send it directly)
>>
>>    
>
> Are you suggesting to simply add the header to the current
> sendemail-validate hook?

No, I'm saying that we currently don't pass them at all, and your patch
adds another argument to a file with the headers.

That *may* break some existing users if they're only expecting the
current argument(s) (although that's probably unlikely), more
importantly we're now doing extra work for all existing hook users, for
the benefit of only some new users.

So I'm suggesting having some opt-in mechanism for the new semantics,
both to preserve the existing semantics for existing users, and for
current and new users avoid writing out the file etc. when we don't need
to.

Which we could do with a config variable,
e.g. hooks."sendemail-validate".includeHeaders=true, or just by having a
new "sendemail-validate-raw" (or whatever we'd call it).

I think it's fine to enforce that if such a new hook exists we'd take it
over the "sendemail-validate" (if any), i.e. we wouldn't need to support
both.