[Mingw-users] Struct of floats returns as NaN values

[David T. <dtu...@st...>]

I am trying to build a program using the proj.4 library on MinGW/WinXP.
The program runs fine in the MinGW shell. When run outside MinGW,
floating point values come back from a function as 1#.IND (undefined, NaN)
and the printfs I added to the function are not run.

This seems to depend on whether I run the program from the directory
it was compiled in. The program produces valid xy values -- but still
does not run the gn_sinu printfs -- if I run it from inside the
directory where it was compiled. It fails if I run it from a copy of
that directory, which makes no sense to me. It fails if the current
working directory is not the compilation directory. MinGW bash runs it
without problems in all these cases.

Brief overview of how proj works:
Each projection is a separate .c file that becomes a separate .o file
in the libproj.a library. Each .c file has macros FORWARD and INVERSE
that are expanded to functions. I am using the "goode" forward function
which in turn calls the "sinu" forward function, which returns a struct
of doubles which are coming back as undefined NaNs.

My full compile line:
g++ -g --static -DPROJ4 -g libdbf.c shapestuff.cc shpopen.o dbfopen.o
shpgeo.o libproj_dmt.a -o ../appoint_part1.cgi

All the other files should not matter. -DPROJ4 does not matter either
by this point; I needed it earlier for the shpgeo file. Everything was
compiled in MinGW on the same machine.


Example output from MinGW bash:

$ ../appoint_part1.cgi
In forward, sinu 414d10 moll 41f120 reg 6, lam0 0.0000, l,p are: -0.9974,
+-0.5407
DEBUG: gn_sinu P->m 0 lam
+13444353561806976000000000000000000000000000000000000.0000,-6696286104189131600
+0000000000000000000000000000000000000000000000000000000000000000000000000000000
+0000000000000000000000000000000000000000000000000000000000000000000000000000000
+0000000000000000000000.0000
DEBUG: gn_sinu post returning xy 0.0427,-0.5407
In forward after mods lam0 0.0000, x,y are: 0.0427, -0.5407 lp is 0.0497,
+-0.5407
In inverse, reg 6, x,y are: -1.0046, -0.5407
In inverse after mods lam0 0.0000, x,y are: 0.0426, -0.5407 lp is 0.0497,
+-0.5407


Example output from Windows cmd:

>appoint_part1.exe
In forward, sinu 3e2758 moll 3e2878 reg 6, lam0 0.0000, l,p are: -0.9974,
+-0.5407
In forward after mods lam0 0.0000, x,y are: -1.#IND, -1.#IND lp is 0.0497,
+-0.5407
(The program then hangs here because it goes into a loop comparing against
the NaN value and I am a bad programmer who forgot to check for that.)


Example output from the compilation directory:

>..\appoint_part1.cgi
In forward, sinu 414c50 moll 41f270 reg 6, lam0 0.0000, l,p are: -0.9974,
+-0.5407
In forward after mods lam0 0.0000, x,y are: 0.0427, -0.5407 lp is 0.0497,
+-0.5407
In inverse, reg 6, x,y are: -1.0046, -0.5407
In inverse after mods lam0 0.0000, x,y are: 0.0426, -0.5407 lp is 0.0497,
+-0.5407


All printfs here go to stderr, in case that matters as to why the
gn_sinu printfs are not appearing when run from cmd. The file extension
(exe or cgi) has no effect on the program's behaviour. 


What might cause the struct to come back full of NaNs? Why would the
working directory matter? Why would it matter whether I am in bash or
cmd? Why aren't the gn_sinu printfs showing up even when it appears
the function is run successfully? 

Thanks in advance for any help,

- David T.