Re: Issues with larger numbers with -j

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Wed, Jun 14, 2017 at 11:20:16AM +0200, Marc Espie wrote:
> make isn't really an appropriate build tool for larger machines with
> lots of cpu.

That's bullshit. make scales quite well to a lot of CPU.

> best-of-the-breed has been ninja for years. I wish it were in more common
> use, as some big projects that support it (chrome) tend to not always use
> it, so they are missing dependencies as well.

Again, while ninja is nice, the primary advantage is for incremental
builds, which are not really relevant for packaging systems.

Missing dependencies have almost always been an issue of lazy
developers, not an issue of build systems. Any form of make has had
support for expressing them.


> Anyway, ninja, is really good at representing complex dependency graphs
> over a whole project, totally avoiding any kind of "recursive make"
> behavior while eschewing extra complexity from some of its competitors.

There is little wrong with "recursive make", if you start with a decent
implementation. The majority of the original paper is a prime lesson in
the problems of GNU make, since it can't express temporary dependencies
for the recursion and therefore has to serialize a lot more than
necessary.

Joerg