diff between defining a "no-op" task and using [noexec]?

diff between defining a "no-op" task and using [noexec]?

[Robert P. J. Day]

  using a concrete example, there's this from oe core's
linux-dummy.bb:

do_configure() {
        :
}

do_compile () {
        :
}

do_install() {
        :
}

  what is the difference between the above and writing:

do_configure[noexec] = "1"
do_compile[noexec] = "1"
do_install[noexec] = "1"

  *is* there a difference? if so, does it have to do with the
processing of other flags for that task?

  i can see that, technically, the first variation still defines a
task to be run, while the second specifies that the task *not* be run,
so i can appreciate that those two operations have different results.

  thanks.

rday

-- 

========================================================================
Robert P. J. Day                                 Ottawa, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: diff between defining a "no-op" task and using [noexec]?

[Richard Purdie]

On Thu, 2013-10-17 at 08:07 -0400, Robert P. J. Day wrote:
>   using a concrete example, there's this from oe core's
> linux-dummy.bb:
> 
> do_configure() {
>         :
> }
> 
> do_compile () {
>         :
> }
> 
> do_install() {
>         :
> }
> 
>   what is the difference between the above and writing:
> 
> do_configure[noexec] = "1"
> do_compile[noexec] = "1"
> do_install[noexec] = "1"
> 
>   *is* there a difference? if so, does it have to do with the
> processing of other flags for that task?
> 
>   i can see that, technically, the first variation still defines a
> task to be run, while the second specifies that the task *not* be run,
> so i can appreciate that those two operations have different results.

The difference is just that, with the first case bitbake will run the
task, it will do nothing. In the second case, bitbake will not bother
running it at all.

Cheers,

Richard

Re: diff between defining a "no-op" task and using [noexec]?

[Robert P. J. Day]

On Fri, 18 Oct 2013, Richard Purdie wrote:

> On Thu, 2013-10-17 at 08:07 -0400, Robert P. J. Day wrote:
> >   using a concrete example, there's this from oe core's
> > linux-dummy.bb:
> >
> > do_configure() {
> >         :
> > }
> >
> > do_compile () {
> >         :
> > }
> >
> > do_install() {
> >         :
> > }
> >
> >   what is the difference between the above and writing:
> >
> > do_configure[noexec] = "1"
> > do_compile[noexec] = "1"
> > do_install[noexec] = "1"
> >
> >   *is* there a difference? if so, does it have to do with the
> > processing of other flags for that task?
> >
> >   i can see that, technically, the first variation still defines a
> > task to be run, while the second specifies that the task *not* be run,
> > so i can appreciate that those two operations have different results.
>
> The difference is just that, with the first case bitbake will run
> the task, it will do nothing. In the second case, bitbake will not
> bother running it at all.

  i figured as much, but is there any important (pragmatic)
distinction between the two if one simply wants to nullify an existing
task?

  if there's a defined task dependency of some kind, i can certainly
see how having a defined task consisting of a no-op would behave
differently from specifying to not run that task at all, in terms of
subsequent dependencies. does that distinction come into play
anywhere? is there an example in the OE source that recognizes this?

  and WRT to how developers should code, is one of these approaches
considered better than the other? i've seen both approaches used in
recipe files, and i'm wondering simply about coding style. thanks.

rday

-- 

========================================================================
Robert P. J. Day                                 Ottawa, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================

Re: diff between defining a "no-op" task and using [noexec]?

[Richard Purdie]

On Fri, 2013-10-18 at 09:13 -0400, Robert P. J. Day wrote:
> On Fri, 18 Oct 2013, Richard Purdie wrote:
> 
> > On Thu, 2013-10-17 at 08:07 -0400, Robert P. J. Day wrote:
> > >   using a concrete example, there's this from oe core's
> > > linux-dummy.bb:
> > >
> > > do_configure() {
> > >         :
> > > }
> > >
> > > do_compile () {
> > >         :
> > > }
> > >
> > > do_install() {
> > >         :
> > > }
> > >
> > >   what is the difference between the above and writing:
> > >
> > > do_configure[noexec] = "1"
> > > do_compile[noexec] = "1"
> > > do_install[noexec] = "1"
> > >
> > >   *is* there a difference? if so, does it have to do with the
> > > processing of other flags for that task?
> > >
> > >   i can see that, technically, the first variation still defines a
> > > task to be run, while the second specifies that the task *not* be run,
> > > so i can appreciate that those two operations have different results.
> >
> > The difference is just that, with the first case bitbake will run
> > the task, it will do nothing. In the second case, bitbake will not
> > bother running it at all.
> 
>   i figured as much, but is there any important (pragmatic)
> distinction between the two if one simply wants to nullify an existing
> task?

One is slightly more efficient since bitbake doesn't run it at all.

>   if there's a defined task dependency of some kind, i can certainly
> see how having a defined task consisting of a no-op would behave
> differently from specifying to not run that task at all, in terms of
> subsequent dependencies. does that distinction come into play
> anywhere? is there an example in the OE source that recognizes this?

No, the way we use noexec tasks, we still have to ensure dependencies
are recognised and built before the noexec task is skipped.

>   and WRT to how developers should code, is one of these approaches
> considered better than the other? i've seen both approaches used in
> recipe files, and i'm wondering simply about coding style. thanks.

noexec is newer, slightly more efficient and not used everywhere. The
":" notation is slightly more readable in some circumstances.

You're looking for a "use X" type answer but I'm not giving one since
sometimes one can make slightly more sense that the other and I don't
want patches doing wholesale conversion from one to the other. I
appreciate that is frustrating but like any things, hard and fast rules
don't make sense, you should think about it.

Cheers,

Richard

Re: diff between defining a "no-op" task and using [noexec]?

[Robert P. J. Day]

On Fri, 18 Oct 2013, Richard Purdie wrote:

> On Fri, 2013-10-18 at 09:13 -0400, Robert P. J. Day wrote:

... snip ...

> >   i figured as much, but is there any important (pragmatic)
> > distinction between the two if one simply wants to nullify an
> > existing task?
>
> One is slightly more efficient since bitbake doesn't run it at all.
>
> >   if there's a defined task dependency of some kind, i can
> > certainly see how having a defined task consisting of a no-op
> > would behave differently from specifying to not run that task at
> > all, in terms of subsequent dependencies. does that distinction
> > come into play anywhere? is there an example in the OE source that
> > recognizes this?
>
> No, the way we use noexec tasks, we still have to ensure
> dependencies are recognised and built before the noexec task is
> skipped.

  ah, that answers that question.

> >   and WRT to how developers should code, is one of these
> > approaches considered better than the other? i've seen both
> > approaches used in recipe files, and i'm wondering simply about
> > coding style. thanks.
>
> noexec is newer, slightly more efficient and not used everywhere.
> The ":" notation is slightly more readable in some circumstances.
>
> You're looking for a "use X" type answer but I'm not giving one
> since sometimes one can make slightly more sense that the other and
> I don't want patches doing wholesale conversion from one to the
> other. I appreciate that is frustrating but like any things, hard
> and fast rules don't make sense, you should think about it.

  i'm good with all that, thanks.

rday
-- 

========================================================================
Robert P. J. Day                                 Ottawa, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================