[haiku-development] Re: [GSoC proposal] IMAP FS - An outsider's thoughts

  • From: Ingo Weinhold <ingo_weinhold@xxxxxx>
  • To: haiku-development@xxxxxxxxxxxxx
  • Date: Sat, 16 Apr 2011 12:35:42 +0200

On 2011-04-16 at 01:37:12 [+0200], Donn Cave <donn@xxxxxxxxxxx> wrote:
> It's an appealing idea.  RFC822 headers, MIME, base64 decoding,
> even multi-line headers and CRLF line separators, it's a giant
> nuisance.

On the other hand the existing mail clients do support all that already and 
have to continue to support it for POP3 (and in reverse for SMTP).

> I wouldn't be worried about the effect on queries,
> because I expect they're broken in that way already. (Whether or
> not the first implementation supports on-demand partial message
> download, eventually we will not have all parts of the message
> on disk available for queries.)

ATM queries don't work on content anyway. Only the header is needed to 
provide the same functionality. Working query support is a must-have.

> But in practice, would you like it?

Nope. IMO it would just break the existing mail applications for no 
practical gain I can see.

> That's the head scratcher
> for me.  If you could write an application and use it to transform
> all your mail messages into your decoded sets ... the majority
> of my email has no MIME structure at all, and of the ones that do,
> most of it isn't very interesting (html alternative, etc.)
> So whatever you do has to work well with the plain emails.
> 
> Maybe with a custom filesystem, it's possible to blur the lines
> between a directory and a file, and instead of returning EISDIR
> on a normal read, you could support read on the message node itself
> the way we've been talking about, and also support directory
> operations on it.  Wouldn't guess so, but it would make the whole
> idea more interesting (if only in the sense where `May you live
> in interesting times!' is a curse.)

It would be possible, but depending on what node time stat() reports 
unsuspecting programs would expect only either the file or the directory 
functionality to work (there is no "dile" node type :-)).

CU, Ingo

Other related posts: