Re: 3.2.0

[Jaroslav Hajek]

On Wed, May 20, 2009 at 11:43 PM, Michael Goffioul
<address@hidden> wrote:
> On Wed, May 20, 2009 at 1:56 PM, Jaroslav Hajek <address@hidden> wrote:
>> Fine guess. Cell was still smuggled in by defun.h. While the simplest
>> possible workaround is to just split the file into two parts, I think
>> that instantiating Array<T> for a bazillion of different pointer types
>> just for internal purposes is an useless code bloat anyway. So I
>> changed all the arrays to Array<void *> and added reinterpret_cast
>> conversions. I also included instantiation of Array<void *> in
>> liboctave, as it may be of general interest. Array.cc is no longer
>> included in ov-typeinfo.cc. let me know if you find more problems.
>
> Thanks Jaroslav. I really appreciate your support on these issues.
> Your solution fixes the problem.
>
> I have attached the remaining patches that make octave compilable
> under MSVC. I tried to split them into logical pieces such that it's
> easier to discuss them. Some are trivial, others not.
>
> patch_config: some fixes for configure checks
> patch_export: missing xxx_API tags to export the required symbols
> patch_extern_template: required "extern template" to avoid implicit
>     template instantiation and duplicate symbols
> patch_missing_return: missing return statement
> patch_mkstemp: Win32 does not have mkstemp, so I emulated it
>     with mktemp; this is required as the new help system needs
>     a working version of mkstemp
> patch_pthread: move pthread.h inclusion to the right location
> patch_templates: help MSVC to resolve T determination
> patch_texi2dvi: MiKTeX (that I use under Win32) has too low default
>     value for save_size; I searched hard, but couldn't change that using
>     some config file; this can be kept out octave source tree
> patch_volatile: complete version of the fix from John regarding the
>     problem with volatile variables (discussed earlier today)
> support_cpp_scripts: compile C++ versions of mkoctfile, octave-bug
>     and octave-config; based on non-empty $(EXEEXT); adaptation
>     of Benjamin's patch
>
> Michael.
>

I have no problem with any of these, except the texi2dvi - I suggest
you either keep this one for yourself, or change it so that the stack
size is configurable via configure. But I'd like to ask you to add
ChangeLog entries for all the patches.

You can skip patch_templates.diff - this one seems identical to my
workaround patch for Intel C++, which I kept for myself so far, so
I'll apply that. The code in question is correct C++ AFAIK, but it
doesn't seem to hurt clarity to explicitly qualify the templates. Note
that GCC is apparently smarter here than its commercial competitors.

cheers

-- 
RNDr. Jaroslav Hajek
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz