The shell reads scripts in as it runs, and not even nicely line-by-line, so when you modify a running script, weird things can happen as the shell gets inconsistent versions. Usually this just results in harmless errors like “/home/jlxmodmap not found”, but if you’re like me you worry about “rm tmpfile; process importantfile” turning into “rm importantfile”. The usual technique for the paranoid is to “mv script script.old; cp script.old script”; the shell sees the old script, which you keep unmodified, while you edit the copy. But it’d be nice if you didn’t have to do this everytime…

So, someone told me the trick of having all the work a script does happen in a shell function, which you call at the end of your script. I’m using this for my .xsession (canonical example of a script modified while it’s running), which looks like:

main() {
  do stuff
  do more stuff
  do even more stuff
  exit
}
 
main

voilà. It’s already read in all of main() by the time it runs it, so you can modify it freely, except when it’s just starting up and parsing things.