Comment 19 for bug 98836
Revision history for this message
| Michael B. Trausch (mtrausch) wrote Re: [Bug 98836] Re: [MASTER] "OS locks must die" - dirstate file write locks exclude readers and limit portability | #19 |
On Fri, 2010-08-06 at 00:46 +0000, Martin Pool wrote:
> If someone puts up a patch that solves this without introducing worse
> bugs I will very happily take it. I guess many people may feel it's a
> reasonable tradeoff but just taking chances on losing their data seems
> like a poor risk.
>
> (On the other hand, since I think we do have a lockdir lock, perhaps
> doing without the OS lock won't risk more than a transient error.
> I'll have a look at it.)
I should think that there would be a way for bzr to do all of its
locking internally to the repository, instead of using operating system
locks which are more-or-less different on every operating system. Sure,
they're abstracted a bit, but the exact semantics are different
depending on where you're running and where your data is. I think that
probably one of the major problems with NFS is just that.
Perhaps bzr could learn how to fall back to an internal lock format,
such that even in the face of a broken operating system or a broken
filesystem, bzr can continue operations safely, if maybe more slowly.
--- Mike