libSingular slower than Singular-via-pexpect

[SimonKing]

Basic arithmetic in libSingular seems to be slower than in Singular (used via pexpect):

  sage: P.<x,y,z> = QQ[]
  sage: def test1(x,y,z):
  ....:     for i in range(100):
  ....:         p = (x+y+z)^i
  ....:
  sage: def test2(x,y,z):
  ....:     x = singular(x)
  ....:     y = singular(y)
  ....:     z = singular(z)
  ....:     for i in range(100):
  ....:         p = (x+y+z)^i
  ....:
  sage: %time test1(x,y,z)
  CPU times: user 2.66 s, sys: 0.00 s, total: 2.66 s
  Wall time: 2.67 s
  sage: %time test2(x,y,z)
  CPU times: user 0.15 s, sys: 0.08 s, total: 0.23 s
  Wall time: 1.94 s 

These timings were with sage-4.6.2, but they are similar with sage-4.7.2.alpha_whatever. ​Martin confirms these timings, and finds that similar things also occur with multiplication.

Citing Martin:

However, mod p it seems libSingular wins. Hence my guess:
we are using different memory allocation functions for GMP
and these functions might be slower (for Singular's use case)? 

[jdemeyer]

Milestone sage-4.7.3 deleted