Re: [Evolution] Evo 0.8 & NFS
- From: Berend De Schouwer <bds jhb ucs co za>
- To: John Affleck <jaffleck ping ne mediaone net>
- Cc: Jeffrey Stedfast <fejj ximian com>, evolution ximian com
- Subject: Re: [Evolution] Evo 0.8 & NFS
- Date: Fri, 12 Jan 2001 16:34:36 +0200
On Fri, 12 Jan 2001 16:17:01 John Affleck wrote:
| On Thu, Jan 11, 2001 at 05:45:31PM -0500, Jeffrey Stedfast wrote:
| > "Downgrading" lock types is a bad idea I think. If fcntl doesn't work,
| > then you don't know if it's because the system doesn't support it or if
| > there's something else wrong...so switching to flock (which is broken
| > on 90% of the systems anyway) is not a good idea.
| > Having a user-setting isn't exactly a good idea either because the
| > average user will not know the difference.
| > This is why it's compile-time meant for the system admin to decide.
| Disclaimer: I know nothing about file locking, software development
| or anything else. And it's way to early to be writing this.
| So please feel free to drop this in the bit bucket, but...
| OK. So flock is broken, that I didn't know. But I came up with two
| scenarios where having an application-global file lock type seems less
| than optimal:
| 1 - A mix of locally mounted and NFS mounted home directories. It
| seems that one would like to use fcntl for the locals and perhaps
| flock or something else for the NFS.
No. flock() always returns SUCCESS over NFS, even when it
shouldn't. Over NFS you _must_ use fcntl(). If you use flock()
you don't actually lock the file.
| 2 - A change from an NFS mounted directory to a local one. Do you
| really want to recompile Evolution to support this ?
You don't need to. Evolution shouldn't have to care (or even know)
if its NFS.
| I'm not saying that downgrading is the right thing to do. I don't
| know. I'm just making suggestions. But I do think that Evolution
| will need some kind of runtime configuration of locks. But maybe
| no-one else does it, so Evo doesn't either.
Why? Most users wouldn't be able to give the correct information.
The system administrator should install it correctly. Its not
up to the users to know that flock() is broken, and fcntl() works.
| I do think that behavior on lock failure needs to be looked at a
| little bit. Getting an error message about 'no locks available'
| doesn't seem terribly clear. And the behavior from that point on is
| (or was) a little inconsistent.
Thats a different issue. That is getting error messages, and their
translations correct. But yes, the errors should be clear.
| And just as a general note, (and please don't take this as a flame),
| if Evolution is targeted at the desktop or the 'kinder, gentler
| Linux', is having something as 'user' specific as locking a compile
| time only option the right thing to do ?
Its not user-specific. Its system-specific. NFS is broken or
working system-wide, not user-specific.
| I _really_ like Evolution. I like the attitude behind the
| development. I want it to succeed. I'm not saying that I've proven a
| point, I'm just voicing a different opinion. I'm perfectly content to
| recompile with --file-locking=flock.
| John A.
| evolution maillist - evolution helixcode com
Berend De Schouwer, +27-11-712-1435, UCS
] [Thread Prev