Yes, I agree, but as most of the customer base I target uses the O/S that cannot be named ;-) , file names could become a problem just as 'ln -s' is out of the question.
Yet, this might be the best trade-off. Regards, Philippe Oren Tirosh wrote: > Philippe C. Martin wrote: >> Hi, >> >> I am looking for a stand-alone (not client/server) database solution for >> Python. >> >> 1) speed is not an issue >> 2) I wish to store less than 5000 records >> 3) each record should not be larger than 16K > > How about using the filesystem as a database? For the number of records > you describe it may work surprisingly well. A bonus is that the > database is easy to manage manually. One tricky point is updating: you > probably want to create a temporary file and then use os.rename to > replace a record in one atomic operation. > > For very short keys and record (e.g. email addresses) you can use > symbolic links instead of files. The advantage is that you have a > single system call (readlink) to retrieve the contents of a link. No > need to open, read and close. > > This works only on posix systems, of course. The actual performance > depends on your filesystem but on linux and BSDs I find that > performance easily rivals that of berkeleydb and initialization time is > much faster. This "database" also supports reliable concurrent access > by multiple threads or processes. > > See http://www.tothink.com/python/linkdb > > Oren -- http://mail.python.org/mailman/listinfo/python-list