Re: [Evolution-hackers] CamelFolder updates

> I'd like to fix up camel-folder. Specifically:
> 1) Fix the glib precondition stuff, fix the gtk-doc comments, etc.
> 2) Remove deprecated methods:
>    - async open/close mehthods
>    - list_subfolders and get_uid_list because they explode on Very
>      Large folders, whereas get_subfolder_info / get_message_info
>      don't.

I dont think the list of subfolders will ever be unusably big.

The summary will always have to be in-memory for any performance
at all (vis the totally unusable mailer of a month ago, and
that was partially from going through get_message_info, and it
was all STILL in memory!).

Personally I think get_uid_list and get_messgae_info should
both go, and you should just have an accessor method to get a
pointer to the summary array or summary object.

It could be done without this, but if it is, then almost
every operation to do with folders just got a whole lot more
complicated (searching, adding, indexing, etc).

>    - copy_message_to because it's a useless optimization
> 3) Make all of the capability checks use the same interface
>    (overridable method whose default implementation checks a
>    gboolean CamelFolder member)

Hmm, so implement it as a bitmask?

> 4) Message numbers
>    Message numbers are not thread safe (or, in the case of IMAP, safe
>    against multiple concurrent accessors to the same folder).
>    We've talked about removing the by_number interfaces, but some
>    stores have no concept of UID, so if they're forced to implement
>    "pretend" UIDs, they'll have nothing but message numbers to work
>    with.

I see no problem with that, personally.

>    Also, message numbers are used by camel_folder_get_message_info,
>    which is used to get summary information which is needed to use the
>    uid routines. The old solution of having the routine return summary
>    info for the entire folder is potentially bad, because folders
>    could be very very very very large.

Well, read above.  This will basically break everything, and/or
make it all unusably slow even on modest message sizes.

So, what would be 'very very very' large?  I thought news would
be the biggest message base, and that isn't 'very very very' large.
(when taken group at a time).

>    One possibility is to require that the message numbers in a
>    CamelFolder can never change while the folder is open. This would
>    mean that some Folders would need to keep a real to virtual message
>    number map. Probably get_message_info would just ignore the holes
>    in the numbering, such that asking for summary info for 1-5 when 3
>    had been deleted would return 4 summary entries.
>    That seems like it could be messy. Anyway, we need to do something
>    here.
> 5) Get rid of the concept of non-existent folders. Move the folder
>    creation code into CamelStore, and have camel_store_get_folder
>    return an exception if you ask for a folder that doesn't exist.
>    This gets rid of a lot of code that we're not taking any advantage
>    of.
>    Also move delete_folder into CamelStore.
> 6) Consider getting rid of the concept of closed folders. Currently
>    "open", "create", "delete", and "delete_messages" are the only
>    methods that require closed folders, but there's lots of checking
>    of folder state that goes on that could be gotten rid of if we made
>    it so camel_store_get_folder returns the folder in an "open" state,
>    and had "close" happen automatically when all references to the
>    folder went away.
> -- Dan
> _______________________________________________
> evolution-hackers maillist  -  evolution-hackers helixcode com

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]