Re: [PATCH 2/2] imap_fetch: Add a command to continuously fetch from an imap mailbox

[ebiederm@xmission.com (Eric W. Biederman)]

Eric Wong <e@yhbt.net> writes:

> "Eric W. Biederman" <ebiederm@xmission.com> wrote:
>> Eric Wong <e@yhbt.net> writes:
>> > "Eric W. Biederman" <ebiederm@xmission.com> wrote:
>> >> > The email messages are placed without modification into the public
>> >> > inbox repository so minimize changes of corruption or of loosing
>> >> > valuable information.  I use the command imap_fetch for all of my
>> >> > email and not just a mailling list mirror so I don't want automation
>> >> > to accidentally cause something important to be lost.
>> >
>> > Btw, Email::MIME usage is gone from 1.5.0 due to nasty
>> > performance problems and replaced by PublicInbox::Eml.  Eml
>> > should be completely non-destructive unless somebody sends an
>> > abusive message which exceeds the new safety limits; in which
>> > case it won't OOM or burn CPU like E::M did.
>> >
>> > That said, {-public_inbox_raw} still works and Eml looks
>> > like a drop-in replacement as far as imap_fetch is concerned.
>> 
>> I almost did that. But I looked and saw PublicInbox::MIME still present
>> and a number of other references to Email::MIME so I wasn't certain
>> exactly how that was being handled.  But since Email::MIME still
>> worked I didn't mess with that.
>
> I think the Import .pod documentation is the only place aside
> from some random comments and maintainer tests in xt/*, right?

I am looking at 1.5.0 so you may have made a bit more progress
but Import.pm still uses Email::MIME,
and PublicInbox::MIME still uses Email::MIME as a base class.

> I'll definitely keep the import + indexing parts compatible with
> Email::MIME.
>
> I'm not sure how much of a Perl API I want to expose in case
> this gets reimplemented/migrated to another language someday,
> but maybe exposing the most heavily-used parts is OK.
>
> I also don't think another language is likely, at this point.


>> >> > No email messages are deleted from the server instead IMAPTracker
>> >> > is used to remember which messages were downloaded.
>> >
>> > Yup.  I've integrated IMAPTracker into a local branch, already.
>> > I think it could be reused for tracking NNTP fetches, too,
>> > since UIDs and NNTP article numbers seem interchangeable.
>> 
>> True.  I guess it is possible to make the equivalent of my imap_fetch
>> for NNTP queries as well.  Does NNTP have the equivalent of IDLE where
>> it is possible to get timely updates of new messages?
>
> I haven't seen anything in NNTP like IDLE.  Maybe there's
> something in the draft stages, but AFAIK nothing widely
> implemented.
>
>> I suppose I should see if the git protocol does as well.  I am trying to
>> figure out what the best way to track other public inbox git repositories.
>
> I'll likely implement a custom HTTP endpoint and usable with
> (curl -XIDLE ...); but the timeout will need to vary depending
> on firewalls.
>
>> Currently I have a script that every N minutes does git remote update.
>> Which is ok.  But I think something like imap_fetch might be nicer.
>
> Definitely.
>
>> >> Bah.  I sent this a little too soon.  The patch needs this small
>> >> incremental fix to actually work.
>> >
>> > No worries.
>> >
>> > Btw, any reason you create the SSLSocket yourself instead of
>> > passing (Ssl => \@SSL_Socket_options) to IMAPClient->new?
>> 
>> When I read the documentation it looked like that was the way to do
>> things.  Even now when I reread the documentation that looks like the
>> way to go.  Especially if I wanted to be certain the connection was
>> encrypted.
>
> There seems more than one way to do it, but `Starttls' and `Ssl'
> are just as documented from what I tell (in v3.38).
> Socket/RawSocket seem useful for using an external command to
> connect/launch an IMAP tunnel or server; so it'll be used to
> mimic the `imap.tunnel' support of git-imap-send.

Now that you point it out I can see it.  Commands like starttls
are a bit dangerous as they are subject to man in the middle attacks.

But I think that is the difference of just tossing something together
for yourself versus making something that works with everyone's setup.

The one challenge I ran into was getting ssl verification to work on
RHEL7.  Apparently IO::Socket::SSL::default_ca() does not exist in
the old version of perl that comes with RHEL7.  Which is why I have
the %ca and the eval.

Eric