Re: running with -jN on Windows

[Eli Zaretskii]

> Date: Mon, 16 Jan 2006 01:36:02 -0500
> Cc: address@hidden
> From: "Paul D. Smith" <address@hidden>
> 
> I don't understand: I thought the entire point was to support the
> jobserver on Windows.  If jobserver isn't supported, just what does -jN
> _do_ during recursion?

Sorry, I'm confused: are you saying that -jN is useless without the
jobserver?  I doubt that you are saying that.

The point of what I did was to allow -jN when sh.exe is nowhere in
sight, because that limitation is probably due to bugs in creating and
using temporary batch files which are then submitted to cmd.exe.  With
the current code that handles the temporary batch files, this problem
is gone, so the limitation can be lifted; this is what my changes did.
I think this is progress of sorts, no? ;-)

The other issue is jobserver and recursion.  Do I understand you
correctly that, without jobserver support, recursive sub-Makes are
_supposed_ to be invoked with -j1, unless the Makefile explicitly
tells otherwise?  If so, then I understand why that single test from
parallelism behaved like it did.  Perhaps parallelism should be
amended to expect the non-jobserver result as well (and announce
loudly that the system doesn't seem to support the jobserver).

Adding jobserver support on Windows requires more work.  Reading the
tests in configure, I understand that jobserver requires the following
features:

  . SIGCHLD
  . sigaction with SA_RESTART
  . waitpid or wait3
  . pipe

Is that true?  If so, then, with the exception of pipes, everything
else needs to be emulated on Windows.  This is a non-trivial job that
will have to wait for another rainy day.