On 6/28/06, Karl Glazebrook <[EMAIL PROTECTED]> wrote:
Stability is good.
However, the lack of new releases is a red flag to someone who might be looking for a solution to some problem. All things being equal, that person is probably less inclined to use software that is not supported, and a lack of new releases imply that it might not be supported. That was my point.
When I couldn't find mail archives past 2004, that really worried me. Clearly this software is supported, but I had to sign up on the mailing list to find out. Since it's a private list, web searches won't find queries even if my question is already answered on the mailing list. Is there a reason for the list to be private?
I agree that the 2.4.2 release is relatively bug free. CVS is where I had issues. It is true that I am using PDL rather differently than others, which is probably why I hit these bugs.
Re: eigens and ssl, see bug 1505171. The gist being that since 2.4.2 release, a library called ssl was added, which I think was not thouroughly tested. (I only tested the eigens function.)
A workaround is to take note if the input matrix is symmetric, and call the older function if so. I suppose I can put together a fix if anyone is interested in testing and applying the patch. This would take the eigens function back to the capabilities of the 2.4.2 release.
Hear hear.
--
Bill I agree with most of what you say but let me make two positive
observations.
On Jun 27, 2006, at 8:31 PM, Bill Coffman wrote:
> 1. The last release was almost two years ago.
>
I'd describe the PDL code as 'stable'. I can run code I wrote 7 years
ago with no problems. Of course this is not to say it is not time for
fresh efforts which may lead to instability. PDL3 anyone?
Stability is good.
However, the lack of new releases is a red flag to someone who might be looking for a solution to some problem. All things being equal, that person is probably less inclined to use software that is not supported, and a lack of new releases imply that it might not be supported. That was my point.
When I couldn't find mail archives past 2004, that really worried me. Clearly this software is supported, but I had to sign up on the mailing list to find out. Since it's a private list, web searches won't find queries even if my question is already answered on the mailing list. Is there a reason for the list to be private?
> 2. The cvs code is pretty bad. The new eigens function (from ssl
> -- probably not such a great acronym for "small scientific
> library") was apparently never tested on a matrix larger than 2x2.
> Opengl didn't even compile. Since I only looked at a small portion
> of PDL, I must assume that there is a lot more problems where that
> came from.
>
I also have found in years of use that it is remarkably bug free once
you master the curly syntax and despite the very curly underlying
code. I think your case is an isolated instance. I do not get the
reference to 'eigens from ssl'
I agree that the 2.4.2 release is relatively bug free. CVS is where I had issues. It is true that I am using PDL rather differently than others, which is probably why I hit these bugs.
Re: eigens and ssl, see bug 1505171. The gist being that since 2.4.2 release, a library called ssl was added, which I think was not thouroughly tested. (I only tested the eigens function.)
A workaround is to take note if the input matrix is symmetric, and call the older function if so. I suppose I can put together a fix if anyone is interested in testing and applying the patch. This would take the eigens function back to the capabilities of the 2.4.2 release.
> 2. Amazing capabilities and speed. This is really an awesome
> package guys.
>
Achieving speed is no mean feat. There's lots of packages out there
which are godawful slow. (How does numpy fair?) Our speed comes from
the basic design decision that the fundamental data structure would
be a C array and when you come down to it all the basic arithmetic is
straight forward C loops (or optimizes to that). It also makes it
very easy to call C maths subroutines. This is something we SHOULD
hang on to.
Hear hear.
Karl
Bill Coffman
_______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
