>>>>> On Mon, 3 Apr 2000 23:37:41 -0400 (EDT), [EMAIL PROTECTED] said:
linas> It's been rumoured that Rob Walker said:
>>
>>
Christopher> d) Store *text* backups, checked in to RCS :-).
>>
Christopher> Option d) [RCS] is a *prime* reason why I would argue for
Christopher> using a text-based data format.
linas> Well, I like this, it does have a certain aesthetic appeal.
linas> -- I beleive it should be quite easy to create an xml version of the
linas> data file. Just copy FileIO.c and replace all occurances of
linas> write( fd, &numAcc, sizeof(int) );
linas> with
linas> fprintf (fh, "<gnc:numAcc>%d</gnc:numAcc>\n", numAcc);
Interesting. I tried to do it with sed, but my sed-fu is weak.
linas> -- I have battle scars from five years ago, when the VRML
linas> mailing list used to list '#1 top priority - binary file
linas> format' and I naively suggested using gzip on the then-current
linas> ascii file format. Despite considerable back & forth, I lost;
linas> the binary file format was invented and was still much better
linas> than gzip, and faster too.
linas> Although cpus are faster today, and will continue to be, if
linas> save file is 500K binary, how big would the gzipp'ed version
linas> be? How long would it take to zip/unzip?
why do we want to compress it? if the wrong part of the file gets
corrupted, no one can uncompress it. at all. ever.
>> Agreed completely. However, I think that Larry McVoy is on to
>> something when he mentions the lack of data integrity being built
>> into RCS via checksums or some other method.
linas> Hmm. This does make me nervous.
linas> -- Single byte corruptions, e.g. someone vi'ed the rcs file,
linas> intending to only view some version number, and accidentally
linas> deleted/added one byte. Oops! *all* versions potentially
linas> corrupted.
linas> -- Multi-byte corruptions, e.g. incomplete writes due to lack
linas> of disk space.
Is there a reason some sort of checksums couldn't be created (per
file, per line, per chunk of output, I don't know) for data integrity,
and after someone edited the file, there would have to be some way to
re-check for integrity, and then re-create the checksums so that
future invocations of the file could read the data and trust it.
linas> Ugh, I have more than once run out of file system space. The
linas> current backup scheme is designed to always leave you with a
linas> usable (if older) copy no matter what. I don't want to loose
linas> that.
Why would we have to lose that? Another nice thing about RCS is that
it doesn't need to have the entire file saved each time.
rob
--
Gnucash Developer's List
To unsubscribe send empty email to: [EMAIL PROTECTED]
