At 1/21/99 11:54 AM -0700, Christopher Hart wrote:
>Another Option.
>We also wanted unique file names.

Another solution I've seen used is to build a message file and populate it with
a reasonable number (based on expected need with generous padding) of possible
names of temporary files.  When a temp file was needed, a standard routine was
called to simply read the message file to get a valid name.  When the temp file
was no longer needed and was purged, the file name was written back into the
same message file, thus becoming available again.  For one reason or another,
sometimes a temp file name would be lost from the queue, but since there were
extra entries in the circular queue this wasn't a problem in the short term.
Occasionally, at convenient times when the application was otherwise down, a
program was run which would delete any temp files left laying about for
whatever reason, rebuild the message file and repopulate it with the desired
number of entries.  This was a very quick task and provided substantially
better performance for acquiring a temporarily unique file name than the
previous solution we had used, which was to increment a counter in an Image
dataset and construct the file name accordingly.