Hi

Seems like the ext4 file system has a bug or some issue with files updated often :

https://bugs.edge.launchpad.net/ubuntu/+source/linux/+bug/317781

I use XML files heavily. It is great, XML units save me a lot of work ! My programs update de xml files quite often.

I am concerned about this particular ext4 bug. Could you say if freepascal's XML units are affected by this bug ? ext4 developers recomend this :

https://bugs.edge.launchpad.net/ubuntu/+source/linux/+bug/317781/comments/54

Thanks in advance

Gustavo

ps: excuse my english

I can't comment on pascal's XML handling, but perhaps I can shed some light on the situation. I belive this new filesystem thing will bite you if you do something like this:

assign (my_file, '/some/file');
rewrite (my_file);
// ... write out the contents of the file ...
close (my_file);

reset (my_file); <-- right here is where you get screwed.

ext4 and other modern filesystems have not committed your previous rewrite to disk yet! So when you try to reset and read the file, you get an EMPTY file from the OS. If you then read in the (empty) data, make a few changes and write it out again, you're going to end up with a patchy file.

Also, if some kind of crash occurs before the file has been sync'd then
the file would be completely gone.

When I write a config file out, I usually do so by writing out program variables by name, hardcoded in my program -- rather than by retrieving data from a config file and then writing out the same data that I just read.

If you must do it by reading and then writing, then call (the pascal equivalent of) fsync after close, keeping in mind that you will have to wait for the (very slow) disk whenever that sync gets called.

Disclaimer: I don't REALLY know what I'm talking about: I am really going out on a limb with some guesswork here. But I think I understood Ts'o's message, and I doubt fpc calls sync on close ;)

Thanks, by the way, for including that message from Theodore Ts'o, which I found to be quite enlightening!

~David.

https://bugs.edge.launchpad.net/ubuntu/+source/linux/+bug/317781/comments/54

This is only a problem when your machine crashes!

Syncing and renaming is not done by the XML units. This is something the programmer has to do.

Well, there is no 'bug', there is a misunderstanding of how things work. That said, XML algorithms use method 1.

You can do method 3 yourself (it's just renaming files and syncing), but I wouldn't recommend it as standard practice, as it uses 3 times more diskspace and is quite slow.

And if methods 1 & 2 are "unsafe", then I think that there is a fundamental bug in the filesystem (I never believed "journaling" filesystems can work 100% correctly in the first place), despite what Theodore T'so says.

The filesystem should just work. The standard application programmer should never have to take into account the eventualities described there.

Michael.

And if methods 1 & 2 are "unsafe", then I think that there is a fundamental bug in the filesystem (I never believed "journaling" filesystems can work 100% correctly in the first place),

Journaling means only: the file system is always in an internally consistent state It does not mean that it doesn't lose data!

despite what Theodore T'so says. The filesystem should just work. The standard application programmer should never have to take into account the eventualities described there.

The filesystem works. It failes only if power is turned off or the system crashes. But those are eventualities the programmer should not take into account ;)

In this definition, yes. But I find that rather limited :-)

That is exactly what I meant...

Michael.

Michael Van Canneyt wrote on do, 12 mrt 2009:

Well, that is the only definition of file system journaling I know of.

But I find that rather limited :-)

It is much better than the previous situation where losing power could mean that your entire partition would become unreadable (due to a corruption of the inode indexes/file allocation tables/b-trees/...).

It's of course not a substitute for backups or versioned file systems.

Jonas

---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.

Ok, thanks for all your replys !

I will take more care with my xml files.

Gustavo

2009/3/12 Jonas Maebe :