Unnecessary imports cause slower sage startup

[jasongrout]

This ticket takes care of some unnecessary imports that were done at startup time instead of inside of functions. It takes care of some of the worst offenders when sage -startuptime is run.

CC: @robertwb @mwhansen

Component: performance

Keywords: import startuptime

Author: Jason Grout

Reviewer: Jeroen Demeyer

Merged: sage-4.6.1.alpha3

Issue created by migration from https://trac.sagemath.org/ticket/10220

[jasongrout]

Attachment: trac-10220-speedup.patch.gz

[jasongrout]

comment:1

I haven't run full doctests yet, so I guess this is needs work right now. I think it's good to go, though, modulo running doctests.

[mwhansen]

comment:2

What is the actual speedup?

[jdemeyer]

Reviewer: Jeroen Demeyer

[jdemeyer]

comment:3

Works for me.

[mwhansen]

comment:4

What is the actual speedup?

[jdemeyer]

comment:5

Replying to @mwhansen:

What is the actual speedup?

Statistically insignificant...

[robertwb]

comment:6

To really speed stuff up you have to eliminate all imports of a module. Easier said than done.

[mwhansen]

comment:7

Replying to @robertwb:

To really speed stuff up you have to eliminate all imports of a module. Easier said than done.

For example, #6494

[jdemeyer]

comment:8

Should this be merged if it doesn't change the startuptime?

[jasongrout]

comment:9

I agree with both robertwb and jdemeyer. When I test it, startup doesn't seem to go down significantly, though the top one or two entries from sage -startup are eliminated.

I think the patch is helpful, but maybe doesn't go far enough to eliminate all imports of a package.

This is interesting, since this patch eliminates the imports with a - next to them in my list of slowest imports in sage -startup. However, the overall startup time does not seem to diminish appreciably. I suppose other modules take the "-" places below so that startuptime doesn't really go down? Or is it that the modules are loaded anyway somewhere else, and loading a module a second time (even in a different module) is insignificant?

sage.all: 1.608 (None)


1.608 sage.all (None)
0.387 sage.schemes.all (sage.all)
0.279 elliptic_curves.all (sage.schemes.all)
0.275 ell_rational_field (elliptic_curves.all)
0.245 ell_number_field (ell_rational_field)
0.244 sage.misc.all (sage.all)
0.243 ell_field (ell_number_field)
0.242 ell_curve_isogeny (ell_field)
0.190 sage.modular.all (sage.all)
0.151 twisted.persisted.styles (sage.all)
0.146 abvar.all (sage.modular.all)
0.145 abvar_newform (abvar.all)
0.141 abvar (abvar_newform)
0.135 sage.groups.all (abvar)
0.124 perm_gps.all (sage.groups.all)
-0.122 cubegroup (perm_gps.all)
0.117 sage.rings.all (sage.all)
-0.113 sage.plot.all (cubegroup)
0.111 sage.symbolic.pynac (sage.all)
0.106 hyperelliptic_curves.all (sage.schemes.all)
0.105 sage_input (sage.misc.all)
-0.104 sage.misc.functional (sage_input)
-0.098 complex_plot (sage.plot.all)
-0.097 numpy (complex_plot)
0.096 sage.schemes.generic.all (hyperelliptic_curves.all)
0.090 fano_toric_variety (sage.schemes.generic.all)
0.088 sage.geometry.all (fano_toric_variety)
0.084 cone (sage.geometry.all)
0.084 zope.interface (twisted.persisted.styles)
0.081 sage_timeit_class (sage.misc.all)
0.081 sage.rings.complex_double (sage.misc.functional)
-0.081 sage_timeit (sage_timeit_class)
0.080 add_newdocs (numpy)
-0.079 interpreter (sage_timeit)
-0.078 IPython.ipapi (interpreter)
0.076 pkg_resources (zope.interface)
0.073 lib (add_newdocs)
0.073 sage.combinat.all (sage.all)
0.063 type_check (lib)
0.063 numpy.core.numeric (type_check)
0.063 sage.algebras.all (sage.all)
0.046 numeric (numpy.core.numeric)
0.046 misc (sage.misc.all)
0.045 sage.combinat.posets.posets (cone)
0.042 sage.graphs.all (sage.combinat.posets.posets)
0.041 Shell (IPython.ipapi)
0.039 sage.server.all (sage.all)
0.037 notebook.all (sage.server.all)
0.037 sagenb.notebook.all (notebook.all)
0.037 notebook_object (sagenb.notebook.all)

[robertwb]

comment:10

I'm in favor of any patches that remove unnecessary imports, as it will make the import order (which is now a mess) easier to understand and actual speedups easier to achieve.

[jdemeyer]

Merged: sage-4.6.1.alpha3

[jdemeyer]

Changed keywords from none to import startuptime

[jdemeyer]

comment:11

Replying to @robertwb:

I'm in favor of any patches that remove unnecessary imports, as it will make the import order (which is now a mess) easier to understand and actual speedups easier to achieve.

As you wish :-)