Re: race condition with reap_children when $(shell) used

[Grant Taylor]

>>>>> "Howard Chu" <address@hidden> writes:

> Interesting, but the original idea was to use tokens instead of load
> average to control job spawning. If you're using load average, then
> there should be no token processing at all.

Well, if the intent is that you use -l xor -j#, then there should be
code to prevent uers from doing -l and -j#.  I would not expect make
to leak tokens in any token case; any token leak seems like a clear
bug to me.

On Linux, at least, you're not likely to get useful behavior from
something like -j -l4; Linux updates the load once every several
seconds, so in our environment, the thing will instantly spawn 1500
jobs and the machine will keel over.  That doesn't seem useful.

IMHO it's entirely reasonable to specify something like -j8 -l4;
especially when used with things like distcc that offload part of each
job's work to some other system.  With my patch this works well, make
runs some variable number of jobs up to 8 and the load hovers right
around 4.

--
Grant Taylor - address@hidden - http://www.picante.com/~gtaylor/
  Linux Printing Website and HOWTO:  http://www.linuxprinting.org/