> More evidence that relying on shell functions is correct: ever since autoconf > 2.62, autotest has been using shell functions without explicitly requiring > them > (it was only suggesting them).
Indeed, though how many people run something as slow as autotest on old, and hence probably not-so-fast, systems? :-) > If we don't apply Paolo's patch 5, then > we should really check in something like this, to make autotest's use of > shell > functions explicit (rather than relying on the luck of suggestions being > met). > However, Paolo's patch makes AS_PREPARE require rather than suggest shell > functions, at which point this patch is nothing more than documentation. So > does anyone still think this is worth pushing? I'm personally thinking of > dropping it, with just this mail as an archive of why. Yes, indeed, unless for some weird reason we cannot apply patch 5 (which would be sad). In fact depending on available free time I'd like to follow up with *autoconf* usage of shell functions. > On the other hand, I've noticed that we don't do any sanity checking if > CONFIG_SHELL is set. Would it make sense to require that CONFIG_SHELL can > pass > all of the required tests, and if not, complain to the user and abort rather > than the current behavior of trying to limp along until the use of a required > feature finally trips up the script? Yes, that would probably be better. Paolo
