tclSE-back-port: timerate and more

[sebres]

This is a interim PR back-porting some features from my tclSE engine.
Currently used as merge-point to #2 to compare performance between original tcl-clock and new tcl-clock engines.

Contains:

• new command timerate, that allows to do the measurement of code executing more precise and in opposition to command time is time-related (not count-related);
• time-drifts bug fix, corresponds http://core.tcl.tk/tcl/info/b6fc234ef3dac6df (see ticket http://core.tcl.tk/tcl/tktview/b87ad7e9146832d505f9a430d779c5313c440256)
• several small fixes to test tcl-clock under Windows (timezone, registry in development env, etc.);

Additionally pros of command "timerate"

• timerate can be calibrated to minimize influence of measurement overhead (and byte code execution overhead) - timerate -calibrate {}.
  Following example diff shows influence of measurement and execution overheads (the times retrieved for execution of empty scope-block resp. of each set i ... is more precise after calibration), e. g. in reality the execution time of set i ... amounts to ca. 0.057 µs/#:

+ % # calibrate to minimize influence of overhead:
+ % timerate -calibrate {}
+ 0.04763798110968011 µs/#-overhead, 0.048467 µs/#, 40275119 #, 20632745 #/sec
  % timerate {}
- 0.049425 µs/#, 20232613 #, 20232613 #/sec
+ 0.003276 µs/#, 19640807 #, 305208960 #/sec
  % timerate {set i 0} 5000
- 0.155394 µs/#, 32176202 #, 6435240 #/sec
+ 0.057090 µs/#, 47953979 #, 17516238 #/sec
  % timerate {set i 0; set i 1} 5000
- 0.221634 µs/#, 22559694 #, 4511938 #/sec
+ 0.113668 µs/#, 31042572 #, 8797543 #/sec
  % timerate {set i 0; set i 1; set i 2} 5000
- 0.293191 µs/#, 17053736 #, 3410747 #/sec
+ 0.190558 µs/#, 21012023 #, 5247746 #/sec

• to reset previously calibrated overhead timerate -calibrate {} 0 can be used;
• it can measure compiled and not compiled code pieces (use option -direct to disable "compilation" outside of measure cycle, and will be executed by TclEvalObjEx instead TclNRExecuteByteCode);
  Normally it uses compiled version to measure (as it would be a part of compiled script or procedure body):

- # measure not-compiled version:
- % timerate -direct {}
- 0.089162 µs/#, 11215515 #, 11215515 #/sec
+ # measure compiled version:
+ % timerate {}
+ 0.049425 µs/#, 20232613 #, 20232613 #/sec

• with option -overhead an external calculated overhead can be specified for the timerate command to measure execution of the part of script. Following example shows possibility to ignore overhead of execution of _init procedure resp. more precise the speed of part string tolower ... only:

% proc _init {} { string trim " test " }
% timerate { set obj [_init]; string tolower $obj }
0.758217 µs/#, 1318883 #, 1318883 #/sec
% timerate \
   -overhead [lindex [timerate { set obj [_init] }] 0] \
           { set obj [_init]; string tolower $obj }
0.097757 µs/#, 1393420 #, 10229413 #/sec

[sebres]

Have a small amend for this trunk-based branch in d6d3cb9, that was forgotten, because would be not necessary in tclSE at all (the objects could own more as one slot with internal representation there, so byte-code will be not released, if e. g. unicode representation needed, so exists as long as the object exists).

Without this fix following code produces a segfault through removed byte-code representation in the meantime:

set code { regexp {\w} $code }
timerate $code