windows 9x trials

[Paul Kienzle]

All,

Is there anyone else out there who is using oct-files under
Windows 95-98-ME?  

I'm still having difficulty with, e.g., "min.oct has caused 
a error" message box popping up for each oct-file that has 
been loaded whenever octave forks.  It does this a lot, like 
every time the pager is called unless page_screen_output = 0.  
It works for me under Windows 2000.  Does it work for anyone 
under 9x?

I tried writing a trimmed down program which did a dynamic
load then a fork, but I was not able to reproduce the problem.

I'm finding fork under cygwin is painfully slow for windows
9x.  Other than, "so don't do it", are there any suggestions?
"Don't do it" will require a couple of changes to the way
things are done.  We would need to build in a pager.  And our
own ls, but that should be easy enough. Texinfo will continue 
to be a problem.  

I suppose we could preprocess all the texinfo help and put
them in a separate files (e.g., .hlp), possibly on a
separate path.  Octave would have to know to look on the
other path or for the other extension first before grabbing
the doc string from the function.  We may want the ability 
to separate the help from the function anyway so that users
can have help in their own language, assuming someone wants 
to work on the translation.  Translation of error messages 
and other internal strings is a separate problem.

I built a shared version of Andy Adler's most recent octave
for windows install exe.  Install size 130 Mb before stripping 
the oct-files, 65 Mb after.  Distribution size is 13 Mb
compared to Andy's 4.5 Mb.  

Looking through one oct-file prior to stripping with nm, most 
of the symbols seem to be related to iostream.  Indeed, 
recompiling my dynamic loading example from yesterday, dynsub.dll 
jumps from 21 kb to 454 kb just from using iostream.  Is there 
some way of telling g++ to only put iostream in the exe and not 
in each of the dlls?  Or better yet, leave it in a dll and link 
to that?  Cygwin only has libstdc++.a, but this may just be an
interface to some unnamed dll and not the implementation.  Would
compiling libstdc++ as a dll solve oct-file size problem?  Is
there a regular on the cygwin list who could ask the question
there?

Thanks in advance,

Paul Kienzle
address@hidden