Re: parallelism in the module-level, generated GNUmakefile's

[Laszlo Ersek <lersek@redhat.com>]

On 01/23/19 17:36, Gao, Liming wrote:
> Laszlo: By design, BaseTools generates GNUMakefile with the complete
> dependency. So, the generated GNUMakefile should support parallelism
> run. But, I don't verify this functionality. Have you found any
> limitation to forbid parallelism run in module GNUMakefile?

No, I haven't, and that is what surprised me :) I *expected* problems,
but didn't find any.

Given that BaseTools / "build" does not offer any way at all to build a
single generated  GNUMakefile with "make -j", such parallel builds of
said GNUMakefiles have never been tested.

Therefore, I expected them to break, under "make -j".

When they didn't, I was surprised. :)

The question is if we can *rely* on BaseTools to generate the
GNUMakefiles with complete dependencies.

In other words, if we now start building them with "make -j" (via the
outer make), and run into an error due to missing dependencies, can we
report a TianoCore BZ about that? Will it be in scope? Is "complete
dependencies" an explicit goal?

If the answer is "yes", I will happily take that answer.

If the answer is "no", then BaseTools should use .NOTPARALLEL in the
GNUMakefiles (or else "build" should invoke the inner "make" with an
explicit "-j1" flag -- as I've learned today).

Thank you!
Laszlo