pb_ds tests are a bottleneck in the testsuite

[Jonathan Wakely]

I noticed that the ext/pb_ds/regression/trie_set_rand.cc test was the
last one running when all other tests (started with -j40) had finished
some time ago.

That test takes around 30 seconds, even on a fast machine. The 21
tests in the ext/pb_ds/regression/ directory take 3-4 minutes, which
is an incredibly long time for 21 tests when you consider we can run
14000 tests in under ten minutes!

We need to do something about this.

I've attached the times for those tests run on a moderately powerful
laptop. They're all too slow, but some are ridiculous.



-------------- next part --------------

associative_containers.cc
real	0m11.787s
user	0m10.898s
sys	0m0.881s

hash_map_rand_debug.cc
real	0m9.986s
user	0m9.139s
sys	0m0.833s

hash_set_rand_debug.cc
real	0m9.226s
user	0m8.378s
sys	0m0.837s

list_update_map_rand_debug.cc
real	0m9.127s
user	0m8.284s
sys	0m0.837s

list_update_set_rand_debug.cc
real	0m8.826s
user	0m8.110s
sys	0m0.720s

priority_queue_rand.cc
real	0m24.784s
user	0m23.789s
sys	0m0.973s

priority_queues.cc
real	0m9.891s
user	0m9.123s
sys	0m0.765s

tree_map_rand_debug.cc
real	0m9.020s
user	0m8.192s
sys	0m0.813s

tree_set_rand_debug.cc
real	0m8.748s
user	0m7.950s
sys	0m0.792s

trie_map_rand_debug.cc
real	0m8.881s
user	0m8.107s
sys	0m0.779s

trie_set_rand_debug.cc
real	0m9.123s
user	0m8.319s
sys	0m0.797s

hash_map_rand.cc
real	0m35.104s
user	0m33.939s
sys	0m1.085s

hash_set_rand.cc
real	0m32.647s
user	0m31.603s
sys	0m0.975s

list_update_map_rand.cc
real	0m19.273s
user	0m18.326s
sys	0m0.901s

list_update_set_rand.cc
real	0m17.870s
user	0m16.924s
sys	0m0.921s

priority_queue_binary_heap-62045.cc
real	0m9.542s
user	0m8.776s
sys	0m0.763s

priority_queue_rand_debug.cc
real	0m9.099s
user	0m8.346s
sys	0m0.753s

tree_map_rand.cc
real	0m49.500s
user	0m48.211s
sys	0m1.166s

tree_set_rand.cc
real	0m47.879s
user	0m46.695s
sys	0m1.103s

trie_map_rand.cc
real	0m37.150s
user	0m35.901s
sys	0m1.151s

trie_set_rand.cc
real	0m36.262s
user	0m35.205s
sys	0m0.981s