Re: Filetype identification in gmc

[mawarkus t-online de (Matthias Warkus)]

+++ Wed, Aug 04, 1999 at 03:20:01PM +0100 +++
Mark R. Bowyer e-mails me. Film at 11. Reply right now, after the break.
> 
> >From: mawarkus@t-online.de (Matthias Warkus)
> 
> >The problem here is that it doesn't scale, if you've got a directory
> >with 10,000 files, you need to do 10,000 calls to file, which spawns
> >10,000 processes one after another and does 10,000 open() calls.
> >
> >Over a network, this could take hours.
> 
 
> This isn't a very good arguement.  Why can't it just do a single
> "file *" in each directory, and then parse the output?  Make sure
> you don't over-run any buffers (in your 10,000 file case) but you
> get 1 call to file, spawning 1 process, which returns quite quick,
> even over a network, scaled by the number of files - not hugely
> longer than what we have now.  And it uses a system resource, rather
> than reinventing the wheel, and making it triangular, as we have
> now.
> 
> Note that, at least on Solaris, you have options to (a) use your own
> magic file (so users can extend the list without needing root
> access) and (b) give it a list of files, for file selections and in
> case the "*" option isn't quite right.

Whatever. So you save the overhead for spawning file processes.
Instead of 40,000 open() calls, you just need to do 10,000.
Nevertheless, I *don't* want to see 30 users accessing an NFS-mounted
directory with a file manager that does this. Wee! Get over a quarter
million file accesses for free!

mawa
-- 
Multiple exclamation marks are sign of a diseased mind.
                                                    -- Terry Pratchett