Fix parallel building of Singular

[jdemeyer]

See ​http://sage.math.washington.edu/home/palmieri/misc/singular-3-1-3-3.p1.log from #12016.

New spkg: ​http://boxen.math.washington.edu/home/jdemeyer/spkg/singular-3-1-3-3.p2.spkg

Reported upstream: ​http://www.singular.uni-kl.de:8002/trac/ticket/395

[jdemeyer]

Fixed one issue, hit another. It seems Singular's Makefiles are very broken. Looks like a "$MAKE -j1" is in order.

[malb]

At some Sage/Singular? days there was a push to make Singular compile in parallel and it worked. Perhaps it got broken recently, but the Singular developers do fix these bugs. Hence, they should be reported upstream and fixed,not worked around by "make -j1".

[jdemeyer]

Replying to malb:

At some Sage/Singular? days there was a push to make Singular compile in parallel and it worked. Perhaps it got broken recently, but the Singular developers do fix these bugs. Hence, they should be reported upstream and fixed,not worked around by "make -j1".

Fair enough. I found two problems and fixed them. I am now continuously building Singular in a loop, so far it seems to work. So needs_review.

[jdemeyer]

diff for the new Singular spkg, for review only

[vbraun]

Is there a particular reason why you are adding the + to some commands in the makefile?

-       $(recurse) 
+       +$(recurse) 

I'm fine with it as it helps debugging, but it shouldn't change whether or not the Singular spkg builds.

[jdemeyer]

Replying to vbraun:

Is there a particular reason why you are adding the + to some commands in the makefile?

To fix the warning

make[1]: warning: jobserver unavailable: using -j1.  Add `+' to parent make rule.

The problem is that make doesn't recognize it as a recursive rule because the command is the result of a variable expansion.

I'm fine with it as it helps debugging

Debugging, how?

but it shouldn't change whether or not the Singular spkg builds.

True. But it does fit with the theme of fixing parallel building.

[vbraun]

I guess the usual use for + rules is to execute the command even if you run make --just-print. I hadn't thought about the make job server, this is indeed modified as well.

[jhpalmieri]

This also works for me. I could force the old version to fail when building in parallel (on sage.math: MAKEFLAGS='j -l8' ./sage -f ..., and similar on OS X). With the new spkg, it has built every time on sage.math (using both -l30 and -l2). On OS X, I had one failure with the new spkg using -l8, although I've had 4 successful builds that way. The machine only has 2 cores, so -l8 is probably a bad setting anyway. With -l2 on that machine, no problems.