Re: [Qemu-devel] [PULL 04/16] Makefile: Rules for docker testing

[Eduardo Habkost]

On Mon, Jun 06, 2016 at 04:32:50PM +0200, Paolo Bonzini wrote:
> On 06/06/2016 16:15, Eduardo Habkost wrote:
> >> > No, the second time you'll get the expansion of
> >> > 
> >> > DOCKER_SRC_COPY := docker-src.$(CUR_TIME)
> > And why it isn't enough to simply write the above line in the
> > Makefile without any eval trick?
> 
> Because 1) initially the idea was to delay the $(mkdir) too 2) Too many
> $(shell) end up slowing down make.

Now I see: you wanted to delay the shell command, but evaluate it
only once.

But I don't see why it makes sense here, because the variable is
being expanded immediately after it is defined (in the
"$(DOCKER_SRC_COPY):" line), and CUR_TIME is already defined
using ":=".

> 
> >>> > > Why did you use this trick instead of just relying on ":=" to
> >>> > > expand the shell command only once?
> >> > 
> >> > I think initially the $(mkdir) was in DOCKER_SRC_COPY too.  Also, too
> >> > many $(shell) end up slowing down make.
> > And how exactly does the eval trick helps avoiding slowing down
> > make, that wouldn't happen if using just:
> >   DOCKER_SRC_COPY = docker-src.$(CUR_TIME)
> > ?

Oops, I mean:
  DOCKER_SRC_COPY := docker-src.$(CUR_TIME)

> 
> If you define CUR_TIME with "=", the DOCKER_SRC_COPY value changes every
> second.  If you define CUR_TIME with ":=", again you have a useless
> $(shell).

Why a useless shell?

> 
> In this particular case I agree that it's probably premature optimization.

In this particular case it doesn't seem to be an optimization at
all.

-- 
Eduardo