[GRASSLIST:4586] Re: .tmp files

[Christoph Hoegl]

Hi all,

why not implement a rather easy method involving a lock file?

grass tasks could create 
first a lock file called $tempname.lock containing the pid of the creating 
program,
second the $temp_dir temporary directory
after work is done the $temp_dir and afterwards the $tempname.lock file
are unlinked.

If manage_a_lock.sh dies then there will be no program using the pid resulting
in deletion of the $tempname.lock as well as the $temp_dir, or better
have it live a grace period that is checked by a cronjob

(this applies only if the pid wraparound time is long enough)

The actual deletion of the $temp_dir and lock files as well as the creation of 
them could be managed by a shell wrapper (pretty easy but a bit of agreement 
on the side of core developers would be nice):

example pseudo code:

before:

v.r.sth args # process creating a tempfile

after
manage_a_lock.sh v.r.sth args 

with manage_a_lock.sh looking like this (pseudo again):

create_temp_dir
create_temp_dir.lock
run_grass_task in $temp_dir
if status of job == error keep files and send a message
else delete $temp_dir $temp_dir.lock
send message of job completion

this could of course be implemented in each command creating a tempfile
(and is by design the better solution but while testing the idea a shell
wrapper is doing enough)

There are some caveats though:

Are all commands writing to a tempdirectory which may be given
as a variable (like $GRASS_TEMP)?

What is the best strategy to implement it into library code
so transition/portability is provided for module authors?


regards,
Christoph

-- 
Christoph Hoegl / C.Hoegl at gmx.net
=> OOOOOOOOOOOOOO = Now in the "TUNNEL IN NO TIME"-team = OOOOOOOOOOOOOOO =>