Re: [Mingw-users] adding my own libraries to mingw

[Keith M. <kei...@us...>]

On 25 November 2010 08:19, William Simpson wrote:
> From:
>
> http://www.mingw.org/wiki/IncludePathHOWTO
> http://www.mingw.org/wiki/SpecsFileHOWTO

You should also look at:

  http://mingw.org/wiki/LibraryPathHOWTO

> it seems I have a few options.
>
> Under djgpp I did as follows, for a library I wrote called rand. I
> have djgpp\lib\src\rand\rand.c
> I have the source rand.c, the header rand.h, and the makefile there.
> In the makefile I have
> rand:   rand.c
>       gcc -c rand.c
>       ar rvs librand.a rand.o
>       del rand.o
>       move librand.a \djgpp\lib
>       copy rand.h \djgpp\include

You should break this up, into separate build and install rules, (and
for preference, also separate clean rules):

  rand: librand.a

  CC = gcc

  libdir = c:/djgpp/lib
  includedir = c:/djgpp/include

  %.o: %.c
          $(CC) -c $<

  rand.o: rand.c rand.h

  librand.a: rand.o # and any other objects you wish it to include
          del $@
          ar rcvs $@ $^

  install:
          copy librand.a $(libdir)
          copy rand.h $(includedir)

  clean:
          del *.o

  distclean: clean
          del *.a

> Then subsequently I put #include "rand" ...

The header file is rand.h, right?  (That's what your makefile installed,
in any case).  Thus, you should use #include "rand.h", (if you want to
be able to override the installed version with a local copy in the same
directory in the including source), or #include <rand.h> to always use
the installed copy from the compiler's include path.

> ... at the top of any program that
> uses it and put -lrand in my makefile when I compile it.

You add -lrand *after* the list of object files, in the gcc command line
in your makefile, at the point where you *link* the final executable.

> First, I think that for mingw I need to put a "\" at the ends of the
> lines like so:
> rand:   rand.c
>       gcc -c rand.c \
>       ar rvs librand.a rand.o \
>       del rand.o \
>       move librand.a \djgpp\lib \
>       copy rand.h \djgpp\include

No, this is not correct.

> By analogy, maybe in mingw I would do:
>       move librand.a c:\mingw\lib \
>       copy rand.h c:mingw\include
>
> I am not at all sure about:
>       ar rvs librand.a rand.o \

The backslashes at the ends of the lines are wrong.

> >From what I read, one possibility is to stick the libraries in
> c:\mingw\local
> or maybe c:\mingw\local\lib ?
> This seems nice, because it keeps my libraries separate from mingw's.
> Would mingw be smart enough to know to look there? Otherwise I guess I
> would need to set LIBRARY_PATH? How?

You can add -L c:/mingw/local/lib to your gcc link command line,
*before* the -lrand, (and before any other -l references into that
directory; one instance of -L suffices for any number of *following* -l
references).  Alternatively, you may adapt the gcc specs file to add
that -L reference automatically, for all gcc invoked linking.

This is explained in more detail, in the LibraryPathHOWTO.

> How about the header files? Where do they go?
> c:\mingw\local

No; not a good idea.

> c:\mingw\local\include ?

That's a viable choice, but...

> Would mingw be smart enough to know to look there?

No; you have to tell it, as explained in the IncludePathHOWTO.

> Maybe there's an INCLUDE_PATH?

Yes, again as explained in the HOWTO; follow the link to CPATH and
related environment variables.

> Fiddling with Specs seems an option too, but it is too sketchy for me
> to figure out.

Well, the SpecsFileHOWTO explains it in comprehensive detail; in
particular, the comment I added as a footnote covers precisely this
customisation.

-- 
Regards,
Keith.