RE: [Bug-gsl] Bug report with VC7

[Vincent Sacksteder]

Dear Brian Gough,
        I'm not sure that I understand your response.  As I understand it, _DLL
should only be defined when compiling a DLL.  In fact it was defined when I
compiled gsl.dll: I did compiles where the output of the preprocessor were
saved as *.i files, and the correct things happened.

The bug I reported occurred while compiling a client exe which used the gsl
header files to link to gsl.dll.  The sample code I sent you was in the main
function of an exe.  That's why I ended up modifying the GSL_VAR definition
for the case when _DLL is undefined.  As I understand it, _DLL should not be
defined when compiling an exe.

So, the net is: I suspect that MSVC7 is doing the right thing by not
defining _DLL when compiling an exe.  And it is defining _DLL when compiling
a dll.

As an aside, I am a little bit mixed up about why there are two gsl_types.h
files: GSL\include\gsl\gsl_types.h which is used by the exe and
GSL\src\gsl\gsl_types.h which is used by the dll.  At any rate, the file I
changed was  GSL\include\gsl\gsl_types.h.

On another note, I figured out why I was having problems with the #includes:
I was making my search directory be whatever\gsl\inc when it needed to be
whatever\gsl.

The warnings are still there though.  All but six are type conversions.  Two
were reports that gsl_sf_legendre_Plm_deriv_array and
gsl_sf_legendre_sphPlm_deriv_array  might not return a value under certain
code paths.  Also memcpy is reported as undefined when compiling cscal.c.
And in coulomb.c G_lam_G and Gp_lam_G are used without having been
initialized.

There is also the following warning:
\Vincent\code\GSL\src\specfunc\legendre_poly.c(649) : warning C4028: formal
parameter 2 different from declaration

Should I send a bug report for these warning issues?

Regards,
        Vincent Sacksteder

-----Original Message-----
From: Brian Gough [mailto:address@hidden
Sent: Thursday, March 06, 2003 11:20 PM
To: address@hidden
Cc: address@hidden
Subject: Re: [Bug-gsl] Bug report with VC7


Vincent Sacksteder writes:
 > Dear GSL developers, thanks for all your work.  I just installed your
 > library and started using it with the Microsoft VC7 compiler.
 > The bug I am reporting is that the linker is unable to find
gsl_rng_mt19937,
 > or any other constant, in the gsl.lib or gsl.dll.  It is, however, able
to
 > find all the functions exported by gsl.dll.  And I checked that it wasn't
a
 > problem with my compile settings by switching to ANSI compatibility mode
and
 > also requiring the compiler to treat all my code as C code.
 >

Hi,
Thanks for the bug report.

In VC6 the macro _DLL is defined when compiling the application using
a multi-threaded DLL (i.e. using /MD).  This picks up the correct
(dllimport) line.

It looks like VC7 doesn't define _DLL, but defines _WINDLL or _USRDLL
instead.

I will modify GSL to check #ifdef GSL_DLL instead of or in addition to
#ifdef _DLL.  In the meantime the simplest workaround may be to define
_DLL manually.

If you still have problems with #include <gsl/...> vs <...> could you
send a bug report for that.

best regards,

--
Brian Gough

----------------------------------------------------------------------
Network Theory Ltd           Phone: +44 117 3179309 (UK: 0117 3179309)
15 Royal Park                  Fax: +44 117 9048108 (UK: 0117 9048108)
Bristol BS8 3AL                WWW: http://www.network-theory.co.uk/
United Kingdom               Email: address@hidden
----------------------------------------------------------------------