Hi Mike,
you wrote about legend.m and myself:
I don't think what you're calling a regression is simply loss of
performance for no good reason. I believe that legend has a lot more
functionality than it used to. Functionality that you may not need, but
that is required for Matlab compatibility. And that may be most of what
is making legend slower in newer versions.
It sounds like you are very interested in this performance loss. If that
translates into a willingness to help improve the function, please do
file a bug report and start investigating what some of the critical
bottlenecks in the legend function are.
and
It's left as an exercise if you or someone wants to try to make the old
legend.m function work with a current version of Octave, I wouldn't
consider that a worthy use of anyone's time though :)
I'm not overly interested in legend.m, it is just the fact that I was able to
pin down some of the slowness (compared to previous versions) to this function
and was able to provide a test case. I thought it was low hanging fruit. Your
statements indicate that those changes have been made because of matlab
compatibility, so I just accept that this is no low hanging fruit at all.
I'll keep an eye on these plotting things and might investigate here or there,
but have to keep in mind what pays off for my employer.
Current versions also introduced a huge penalty that's shown by my script with bAfter=false. 3.2.x
didn't care about that, so users might be happy to read about this in the "tips and
standards" or "plotting" section of the octave manual (setting the legend strings
afterwards is a lot faster than giving them to plot()).
Best regards,
Jens