Re: Stuck with compiling 4.1 on windows (again)

[Hans-Bernhard B. <br...@ph...>]

Bastian Maerkisch wrote:
> Hans-Bernhard Broeker wrote:
>  > Marko Brandes wrote:
>  >
>  >> - copy ..\config\config.nt ..\src\config.h
>  >
>  >
>  > That still looks strange.  One of those ..\ should be superfluous.
>  >
>  > It's no longer really needed, either --- makefile.nt now makes the copy
>  > itself, if you didn't.
> 
> Hhm. It looks like it really should, but it doesn't do so on my system.

Comparing makefile.nt (which I can't really debug, because I never had 
MSVC) and makefile.oww (which I derived from the former), I find one 
important difference that might explain both this, and the original 
problem with nmake not building anything even *after* copying config.h, 
even though it doesn't so much as hint at why this issue is coming up 
only *now* --- makefile.nt has changed only very little over time. 
Could NMAKE have been modified behind our backs?  Gentlemen, please 
report your nmake versions.

The issue is that there's no manifest information about dependencies in 
makefile.nt (nor in any files it includes).  I had to put this rule in 
makefile.oww:

$(OBJS): config.h .AUTODEPEND

This achieves two things:

1) it causes OpenWatcom's make program to automatically compute 
dependencies for all the main .obj files that gnuplot consists of, 
thanks to the .AUTODEPEND tag.

2) it declares every single one of them as dependent on config.h, so 
wmake will try to build config.h, and thus execute the copy rule if 
there's no config.h in sight.

Someone who has nmake and its documentation will have to duplicate this 
for makefile.nt.