Re: [Evolution-hackers] redesign of camel's mime-part content objects

[Not Zed <notzed ximian com>]

On Tue, 2003-07-15 at 04:56, Jeffrey Stedfast wrote:
> On Thu, 2003-07-10 at 16:28, Jeffrey Stedfast wrote:
> > - CamelDataWrapper gets changed so that it has a CamelEncodingType
> >   enum rather than int rawtext member (since rawtext will now always be
> >   TRUE).

Note that this will mean that camelmimepart has a duplicate of the
content_type and encoding parameters.

Should:
 - they be kept, and used as sort of 'intended' vs 'actual' values
 - they be removed, since 'intended' vs 'actual' is already covered by
the fact you have a contentobject with its own values.

> > - CamelDataWrapper::write_to_stream() will qp/base64 decode for us,
> >   but will not do any charset conversion. Perhaps have a flag as to
> >   whether to decode? this way if we are re-writing out the raw message
> >   someplace, no need to have write_to_stream() decode, cuz then we'd
> >   have to re-encode which would break the whole point of this change.
> > 
> >   Hmmm, could we maybe have camel-mime-part.c grab the dw's stream and
> >   write that out raw instead? Will this work in the offline case?
> >   Problem is that if we make the write_to_stream() interface take a
> >   flag as to whether or not to decode, we have to pass extra garbage
> >   (which wouldn't even make sense) when writing out CamelMimeMessages
> >   or CamelMimePart objects.
> > 
> > - camel-mime-part-utils.c will no longer qp/base64 decode the content,
> >   rather it will just read it raw into a memory stream and then set
> >   that on a CamelDataWrapper and also set which decoder to use.

And on the inverse, when creating parts, we just attach them as binary
parts and set the content-transfer-encoding on the content object as
'binary', but the mimepart's content-transfer-encoding to what we want
it set as.

> hacked up most of the above, but it turns out I missed some things that
> are gonna be affected by this change.

Sorry i didn't get back to you on all this earlier ...

> 1. charset conversions in camel-mime-part.c:write_to_stream() - we can
> no longer depend on text parts to be in UTF-8 and stuff. This may not be
> so bad, however, since if the CamelContentType struct on the content
> data-wrapper should contain a charset param... we can then use that
> charset as the 'from_charset' argument and use the charset we want for
> output as the 'to_charset' like we currently do.

Yeah this crossed my mind too.

I think relying on the data-wrapper's content-type for the actual
content type of the physical content (as above) sounds good.

> 2. the composer will have to set the content's charset to UTF-8 if we
> aren't gonna do the actual charset conversion there (we currently
> don't).

Yeah this sounds good.

And then we can set the charset we want to use to send it out with on
the container part?

> 3. camel-mime-message.c:camel_mime_message_set_best_encoding() will need
> to be fixed wrt charsets since it can no longer rely on the content
> being in UTF-8

It's really only needed when creating a new message whose content IS
already utf8, so it might not need to be changed much.


BTW as to your followup, should charset stuff just be removed - i would
tend to say not, because it'll be needed somewhere anyway, and it at
least centralises the code behind a simple api, rather than every user
of camel having to deal with the issue separately.