More below.
leo
The Pie-thon state - 4
# date [- range ] title
# description
10.07.2004 Pie-thon 45 - 47
Enhanced complex functionality. More tests.
First handling of Python attributes.
Pie-thon 48
Unified Parrot PMC singleton objects (Env used a different
scheme) A singleton is also its own class now.
Pie-thon 49
Array is_equal() code and get_string() for repr()
11.07.2004
A day for nothing. Tried to rewrite pie-thon.pl (the Bytecode
translator) in an OO fashion. Decided that it would take too
long and stopped that. Alas Dan should have done it weeks ago.
12.07.2004 Pie-thon 50
Implement lambda subs in pie-thon.pl. Parrot runs:
print filter(lambda x: x < 5, range(10)) == range(5)
Pie-thon 51
Implement Python's float() function. Parrot used to have its
own string to number conversion which was inexact. Now atof()
is used (which needs fixing for C99 compilers that might
interpret "0x23" as hex.
Pie-thon 52 - 58
Implement a sort() hook in Parrot - using qsort() for now.
Better vararg handling using argcP (the argument count).
Implement lower(), upper(), titlecase() in Parrot (ASCII
only).
Python's int() working. Improve sort() speed.
Pie-thon 59
Reworked Parrots objects instantiation the Python way:
$P0 = global "Py_int" # aka "int"
$P1 = $P0() # a new int (value 0)
$P2 = $P0(42) # a new int (value 42)
That allows also isa() to work.
Python 60 - 63
Convert almost all Python objects to that new scheme
14.07.2004 Python 64 - 66
Some experiments with objects.
class C(int):
pass
def main():
i = C(42)
works now, giving a new C object with integer value 42.
14.07.2004 *Dan's already starting with the Bytecode converter - SNCR*
15.07.2004 Pie-thon 67 - buffered IO
Parrot bufferd IO was severly broken - i.e. non-existent.
Fixing that boosts printing 1000 digits of PI from around
the speed of Python to *four* times that speed - wheee.
A smaller (~15%) improvement in BigInt multiply was involved
too.
Pie-thon 68 - xrange
Implemented xrange object
Pie-thon 69 - b1.py is running
The very first benchmark b1.py is running, translated by
pie-thon.pl.
It spends tons of seconds in DOD. (The darned test creates
highly self-referential and deeply nested array structures).
Reduced DOD count from several hundreds down to 28. Still
slow. Optimizing code generation brings DOD count down to 12.
Still slow.
16.07.2004 Pie-thon 70 - 72
Analyzing b1.pir code with cachegrind revealed massive L2
data cache misses caused by 12 DOD runs. By reducing the
number of DODs to half and by optimizations in the DOD code
the cache misses are now down to one third of the original
number.
Also enabled (and fixed) now an enhancement to the JIT
run core called JIT/CGP. Instead of calling the function
versions of non-JITted code, this jumps directly into the
CGP (direct threaded) run-core, if possible. This save some
function call overhead.
b1.pir runs here faster then with Python now.
TODO
* finish bigint
* finish complex PMC class
* finish exception class hierarchy
* tons more
# vim: sw=4 tw=70:
