Closed GoogleCodeExporter closed 8 years ago
thanks for the new issues!!! I will try to look into them soon.. as usual, I'm
a bit burned out after the new release.. :-)
Original comment by mark.duf...@gmail.com
on 25 Jun 2010 at 12:40
thanks again for the patch! because I don't have write access to SVN from china
at the moment, I decided to move to GIT, something I've been wanting to do for
a while.. so I applied your patch there, and will be working from there from
now on:
http://gitorious.org/~srepmub/shedskin/shedskin-srepmub
this is also a temporary repo, until I receive write-access to the 'shedskin'
repo there that was already created by someone. btw, please send patches as
diffs instead of decribing them! :) (attach the output of diff -u, svn diff or
git diff, for example..)
oh, I'm also a bit curious as to why you are testing with STLport..? :-)
Original comment by mark.duf...@gmail.com
on 30 Jun 2010 at 3:56
make that http://gitorious.org/shedskin.. :-)
Original comment by mark.duf...@gmail.com
on 30 Jun 2010 at 4:12
STLport lib has better performance than MSVC in vector insert operation :)
BTW, Are you travelling in China for vacation ?
Original comment by jason.mi...@gmail.com
on 1 Jul 2010 at 1:44
I see. are you sure you cannot use a faster dastastructure, if this is an
important operation for you..?
in the end, we'll probably want to write our own list/tuple implementation,
based on CPython, just like FFAO did for the set/dict implementations.. so it
won't be possible to use different STL versions.
it's sort of a vacation, but I'm staying with the family of my wife, so I can
work here most of the time, if necessary.. :-)
Original comment by mark.duf...@gmail.com
on 1 Jul 2010 at 4:34
My algorithm involves many insertion operation for LIST. You know, LIST is
implemented in python in array/vector, O(n) for non-tail insertion. The best
suitable container for my algorithm is double linked list. No such thing in
python std lib.
From my test, the STLport's insertion performance of large size vector is
better than others.
Original comment by jason.mi...@gmail.com
on 1 Jul 2010 at 5:42
My algorithm involves many insertion operation for LIST. You know, LIST is
implemented in python in array/vector, O(n) for non-tail insertion. The best
suitable container for my algorithm is double linked list. No such thing in
python std lib.
From my test, the STLport's insertion performance of large size vector is
better than others.
Original comment by jason.mi...@gmail.com
on 1 Jul 2010 at 5:46
can't you build your own little container, or use object attributes to maintain
an implicit list? (e.g., x.prev = y, x.next= z) I guess you could win a lot
more with a faster datastructure than by using another STL implementation..
have you profiled the result of shedskin? I can heartily recommend using
gprof2dot.
Original comment by mark.duf...@gmail.com
on 1 Jul 2010 at 2:24
Yes, you are right. Seems like I need to write a link list by myself. I'm
confused, why not python provide link list like container? :-)
Original comment by jason.mi...@gmail.com
on 2 Jul 2010 at 4:54
how exactly do you use the list? have you considered using a queue, dict, or
heapq..?
perhaps a linked list it's just not generally useful enough, with all these
powerful other datastructures at your fingertips. or it's generally elegant
enough to just add 'prev/next' attributes to some simple Node class.
I see a lot of algorithmic code in python, but I don't remember the last time I
saw an actual linked list.. thinking about it, is also _feels_ a bit low-level,
perhaps even unpythonic, but that's just my gut feeling..
Original comment by mark.duf...@gmail.com
on 2 Jul 2010 at 9:16
found some words about this from the BDFL:
http://mail.python.org/pipermail/python-dev/2005-December/058767.html
Original comment by mark.duf...@gmail.com
on 2 Jul 2010 at 9:29
My algorithm: scan a list and pop an element according to a criteria(can be any
position) and put it to the end of the list.
I tried deque, worse than list.
Some guy wrote a linked list container, seems not easy to write a lib-quality
code:
http://stackoverflow.com/questions/2154946/python-linked-list-o1-insert-remove
Thanks
Original comment by jason.mi...@gmail.com
on 2 Jul 2010 at 10:13
perhaps the heapq module (also supported by shedskin) is useful here? it keeps
an array sorted according to some criterion, while you can push and pop items
quickly. that way, you might even avoid the scanning, and just take the first
element.. not sure though if you can use it here. it would be easier if I could
see the code.. :P
Original comment by mark.duf...@gmail.com
on 2 Jul 2010 at 11:56
heapq can't help.
Actually, this list contain all active threads, and need to update state of all
threads every cycle in the simulator. In each scan select the first ready
thread to execute and then put it to the end of the list to achieve Fair/RR
scheduling policy. (need do scanning/remove/insertion millions of times)
So linked list based container is the best choice.
Currently, the list size is below 100, so using the normal list don't hurt very
much. After the whole system is close to complete, I'll do through performance
tuning.
Thanks :)
Original comment by jason.mi...@gmail.com
on 2 Jul 2010 at 12:39
Original issue reported on code.google.com by
jason.mi...@gmail.com
on 22 Jun 2010 at 8:34