Re: Using GNU make jobserver from GCC

[Jan Hubicka]

> On Mon, 2014-05-19 at 07:09 +0200, Jan Hubicka wrote:
> > > http://make.mad-scientist.net/jobserver.html
> > > 
> > > Of course, another option is to have the waitpid() done in another
> > > thread; that would also solve the problem.
> > 
> > Yeah, I read the article. I assume real world implementation will be 
> > slightly trickier,
> > since one needs to take care i.e. of cases where GCC crashes or return 
> > early, right?
> 
> Whatever takes a token out of the jobserver pipe must be certain to put
> it back.  It's not valid to crash or return early without putting the
> token back.  That's just a guarantee you have to make if you want to
> participate.  If you don't then the number of jobs is permanently
> reduced.  It will never go to zero, though, because the make program
> that invoked GCC takes a token for GCC itself and whenever GCC exits,
> for any reason, make will put that token back.  So, the build will
> always progress but will lose parallelism if GCC's handling of jobserver
> is improperly implemented.

Most Makefiles will just quit if GCC fails, but I guess we want to have this
implemented correctly. This is why I would preffer to separate the code into
a small library rather than trying to get across all details independelty
of the GNU make implementation.

I looked at the GNU make implementation very briefly couple months back
and it indeed was not very stand alone, but it should not be that much work,
right?

> 
> > Yeah, do we have some options except for named pipes on hosts where named 
> > pipes
> > are supported?
> 
> Sorry, I don't understand the question.

I think the "+" requirement to connect to jobserver is really a problem here;
while this is documented in GCC manual, most users don't get it right and it 
requires
more changes to makes than really needed.  So if we can't find another 
alternative,
I think it would make sense to go for named pipes in tmp directory. What do you 
think?

Thanks a lot quick reaction!
Honza