Profiling, Code Coverage, Debugging, Unused code

Basic timing of event throughput in a Q&A stress test:

time ./NAR shell < ./examples/nal/example1.nal

Events in example: 220

Timing result: 4.73s

Used CPU: M2 CPU

Event throughput: ~ 47 events per second on average.

Commit: 00606f9af2c8431c9ac5f365da415c8bc5b8bf3d

Profiling:

./build.sh -pg
now run NAR, and after it's done:
gprof NAR gmon.out > analysis.txt

Profiling can significantly help to increase performance. For instance, term comparison via for loop versus memcmp revealed:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total           
 time   seconds   seconds    calls  ms/call  ms/call  name    
 49.18      0.30     0.30 70549890     0.00     0.00  Term_Equal
 26.23      0.46     0.16    15086     0.01     0.04  Memory_addEvent
  6.56      0.50     0.04 17221989     0.00     0.00  Event_Equal
  4.92      0.53     0.03  1308239     0.00     0.00  Term_ExtractSubterm
  3.28      0.55     0.02    15513     0.00     0.04  RuleTable_Apply

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total           
 time   seconds   seconds    calls  ms/call  ms/call  name    
 48.15      0.13     0.13 70549890     0.00     0.00  Term_Equal
 18.52      0.18     0.05    15086     0.00     0.01  Memory_addEvent
  7.41      0.20     0.02 17221989     0.00     0.00  Event_Equal
  7.41      0.22     0.02    15513     0.00     0.02  RuleTable_Apply
  3.70      0.23     0.01  1308239     0.00     0.00  Term_ExtractSubterm

Code Coverage:

./build.sh --coverage
now run NAR tests, and after it's done:
gcov filename.c

now look at filename.c.gcov

Debugging:

./build.sh
gdb
file NAR
run args

if for some reason there is a segfault, type the following to print the stacktrace:

backtrace

Unused code:

The following command helps to detects both unused code and methods which should probably be static (module specific):

./build -ffunction-sections -fdata-sections -Wl,--gc-sections -Wl,--print-gc-sections