Re: Octave-3.6.1 for Windows (VS2008/VS2010) available

[Michael Goffioul]

On Sat, Mar 3, 2012 at 6:36 PM, PhilipNienhuis <address@hidden> wrote:
>
> Nice job! Thank you for doing this.
>
> I suppose you're after some testing results. OK, I tried the MSVC 2010
> version (I selected "SSE3 multithreaded" on my Core Duo box).

Thanks for testing.

> Well, it loads instantaneous (while the MinGW version takes ~25 secs).

This may be due to the fact that no package is loaded by default at
startup. Although 25s is quite large...

> Some preliminary testing showed a few issues:
>
> 1. I selected all packages. None of them seems to be loaded by default.

This is an intentional choice. Even if you install all packages, the
probability that every user wants them all loaded at startup is
probably close to zero. Instead, you can tune that behavior by call
the "pkg" command.

> Seems no problem, because I can do:
>   pkg rebuild -auto
> or
>   pkg rebuild -auto *
> But while I get the usual three "line skipped" messages I know from MinGW
> versions, upon restart of Octave still no packages have been loaded.

I think what's missing is a call to "pkg load auto". I removed it from
the startup script until I sort out the auto-loading issue. At the
moment, any package that is marked auto-loaded by the package author
would be loaded by that command. But that's probably different from
what the user wants.

> 2. Then, doing
>  pkg load all
> I get a long list of warning messages that I don't see on MinGW (except the
> first two and last four):
[snip]

The differences are:
- nan package not present MinGW package: the nan package is known to
shadow a bunch statistical functions.
- ocs: it seems nitzan patched the ocs package, that probably explains
the difference

I don't feel particularly responsible for what the octave-forge
packages are doing. If there are issues, they should be reported to
the package author.

> 3. The io package has no corresponding subdir in /lib/octave/packages (it is
> also lacking in Nitzan's MinGW version, while it was there in 3.4.3).
> We used it to add the io spreadsheet jars to the javaclasspath using
> PKG_ADD.

If you want a lib/octave/packages/io-*/ directory to be present, then
you should put some files in there during the io package installation.
The installer will pick up any files that is in that directory. NSIS
does not support empty directories.

> 4. Do you have plans to include the java spreadsheet .jar files with your
> binary package? AFAICS all of them are covered under Apache or GPL license.

Maybe. It depends the additional size to the installer. If it adds
megs to the installer, I think I'll prefer to keep it separate,
especially if it's not mandatory for the io package normal usage.

> 5. It still portrays bug #34631. I suspect (just a hunch) that this bug is
> caused by some overflow in the fontconfig section. It occurs both on gnuplot
> (hang after 28th plot) and fltk (hang after the 24th plot). Perhaps it is
> related to the amount of fonts installed on my box (151)?
> This bug is still a nasty showstopper for me when trying to persuade my
> colleagues to try Octave. I don't blame you, but

I don't expect any difference between MinGW and MSVC about that bug.
It's not compiler-specific.

> 6. Gnuplot pictures are different (incomplete; e.g., no text) compared to
> fltk ones.

Not sure I understand this one. Do you have an example?

Michael.