Here's all I could get out of a instrumented `osrm-partition`; debug build,
with all the bells and whistles I could think of to make it more verbose:
```
==17928==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 1560 byte(s) in 3 object(s) allocated from:
#0 0x7f4244185b30 in operator new[](unsigned long) ../../../../libsanitizer/asan/asan_new_delete.cc:62
#1 0x7f4242a788b3 (/usr/lib/libtbb.so.2+0x208b3)
SUMMARY: AddressSanitizer: 1560 byte(s) leaked in 3 allocation(s).<Paste>
``
Symbolizing the address results in
```
echo "/usr/lib/libtbb.so 0x7f4242a788b3" | llvm-symbolizer
_fini
```
Looks like a crt finalizer => static global dtor "leaking" from tbb.
Which turned out to be a missing `tbb::task_scheduler_init` on our end:
> Using task_scheduler_init is optional in Intel® Threading Building
> Blocks (Intel® TBB) 2.2. By default, Intel TBB 2.2 automatically creates
> a task scheduler the first time that a thread uses task scheduling
> services and destroys it when the last such thread exits.
https://www.threadingbuildingblocks.org/docs/help/hh_goto.htm?index.htm#reference/task_scheduler/task_scheduler_init_cls.html
Without an explicit instanz the first call to a tbb algorithm seem to initialize
a global scheduler singleton which then "leaks" until the program exits.
Phew.
Implements parallel recursion for the partitioner
Fixes osrm-extract's -dump-partition-graph: accept no further tokens
References:
- http://www.boost.org/doc/libs/1_55_0/doc/html/boost/program_options/bool_switch.html
Pulls parameters through to make them configurable from the outside
Defaults are equivalent to:
./osrm-partition \
berlin-latest.osrm \
--max-cell-size 4096 \
--balance 1.2 \
--boundary 0.25 \
--optimizing-cuts 10
Fixes parallel_do call for Intel TBB 4.2 (Trusty): no range-based overload
Extends explanation for recursive bisection ids
Cleans up Bisection State
Removes license boilerplate from partitioner config
Sorts Spatially and picks Sources and Sinks
Uses sets for sources and sinks for now; see how large they will get
Runs n cuts in parallel changing the slope and uses the best
Clarifies balance <-> ratio naming
* Implements Random Access Iterator Facade for EdgeIDIterator
* Makes StaticGraph Node and Edge requirements explicit
* Cleans up Bisection Graph, Node and Edge
* Cleans up GraphView
- moves collapse into a dedicated set of functions / files
- make collapse scenarios distinct (slight performance cost)
- reduce verbosity for short name segments (now actually working, was supposed to do so before)
Patrick Niklaus
Michael Krasnyk
Moritz Kobitzsch
Daniel J. Hofmann
Daniel Patterson
Daniel J. H
Lev Dragunov
karenzshea
Kajari Ghosh