Re: Bug#822728: RFS: libtcod/1.6.0~pre1+dfsg-1 [ITP] -- graphics and utility library for roguelike developers

[Jakub Wilk <jwilk@debian.org>]

* Adam Borowski <kilobyte@angband.pl>, 2016-04-27, 18:13:
> > For pre-squeeze, there's "getconf _NPROCESSORS_ONLN".
>
> At least one Debian derivative, Dyson, has:
>
> draconis:[~]$ dpkg --print-architecture
> illumos-amd64
> draconis:[~]$ getconf _NPROCESSORS_ONLN
> getconf: Invalid argument (_NPROCESSORS_ONLN)

That's odd.

> so let's not advertise getconf, as -j`something with no stdout output` 
> results in a fork bomb.

I was always baffled with the semantics of make's argumentless -j. Why 
would anyone want this behavior? :-\

Anyway, I'd expect any decent DEB_BUILD_OPTIONS parser to reject the 
bogus "parallel=".

For example, debhelper does it right:

	if ($opt =~ /^parallel=(-?\d+)$/ && $wanted eq 'parallel') {
		return $1;
	}

[Optional minus sign is to support CPUs made of antimatter.]

The snippet I've been using in my d/rules is also immune to accidental 
lone -j:

parallel = -j$(or $(patsubst parallel=%,%,$(filter parallel=%,$(DEB_BUILD_OPTIONS))),1)
...
	$(MAKE) $(parallel)

Unfortunately, the snippet Policy §4.9.1 recommends is forkbomb-prone:

ifneq (,$(filter parallel=%,$(DEB_BUILD_OPTIONS)))
   NUMJOBS = $(patsubst parallel=%,%,$(filter parallel=%,$(DEB_BUILD_OPTIONS)))
   MAKEFLAGS += -j$(NUMJOBS)
endif

--
Jakub Wilk