Complete.Org: Mailing Lists: Archives: offlineimap: March 2008:
Re: [OfflineIMAP] #20: Memory leak on Mac OS X ?
Home

Re: [OfflineIMAP] #20: Memory leak on Mac OS X ?

[Top] [All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index] [Thread Index]
To: Boris Gordon <boris.gordon@xxxxxxxxx>
Cc: offlineimap@xxxxxxxxxxxx
Subject: Re: [OfflineIMAP] #20: Memory leak on Mac OS X ?
From: Vincent Beffara <vbeffara+ml@xxxxxxxxx>
Date: Mon, 31 Mar 2008 23:38:10 +0100

Hi,

Thanks for your feedback.

> > All I can say is that it fixed the crash on the one reproducible
> > test-case I had (one single 5MB e-mail on a local IMAP server on my
> > laptop). I essentially slice the download into 1MB chunks, but maybe
> > this is too large already ... especially with parallel connections.
>=20
> Oh oh.
> Has anyone had any success on Mac OS X over the net rather than local?

Well, from the point of view of offlineimap, this should not make a=20
difference, it is still accessing the server via the same method (no=20
unix socket or anything like that). Not 100% sure about that though.

If you provide me with a reliable way to make offlineimap crash on such=20
a gmail connexion (other than giving me your password, obviously), I can=20
try to investigate. I believe that what needs to be done is to replace=20
the appropriate read() method at the right level, and from the trace=20
what I did is not quite deep enough yet. Or simply the size of the chunk=20
I set (1MB) it too large, but I fear performance problems if I set it=20
too low ...

> > BTW that's only for plain old IMAP, not this fancy schmancy IMAPS whi=
ch
> > I can't test at the moment.
>=20
> ... or with IMAPS ?
> I don't think Gmail offers anything other than IMAP ssl.

Ah. At least you're not the only one with the problem ... though=20
apparently the victims of that particular crash seem to be appearing at=20
an alarming rate these days. Maybe something changed in Leopard and made=20
the crash more likely, who knows.

> > I guess the morale of the story is, hopefully Python itself (if not
> > Darwin ?) will be patched soon.
>=20
> I'm assuming the python people know about this at least ...

They know, and seem to consider that the bug is in Darwin.=20

Anyway, hopefully we find a fix eventually!

  /v

--=20
Vincent Beffara
UMPA - ENS Lyon
46 All=E9e d'Italie
69364 LYON cedex 07
Tel: 04 72 72 85 25
Fax: 04 72 72 84 80



[Prev in Thread] Current Thread [Next in Thread]