Re: [bug-gnu-libiconv] [PATCH] Append the extension to the executables properly

[Bruno Haible]

Hi Ko,

Reviving this discussion from 2016-12-16:

> > But now I found a README.os2 of some package which reads:
> > 
> > -------------------------------------------------------------------------------
> > Installation on OS/2:
> > 
> > - Requires emx+gcc and a working shell (Bourne shell, Korn shell or GNU 
> > bash).
> > 
> > - Set the environment variables CC=gcc CFLAGS="-Zexe" before calling
> >   `configure'.
> > -------------------------------------------------------------------------------
> > 
> > The documentation of the -Zexe option:
> > 
> > -Zexe
> > 
> >     If the -Zexe option is present, emxomfld deletes the output file
> >     (whose name is given on the command line) and adds .exe to the
> >     output filename.  After calling LINK386, emxomfld creates an empty
> >     output file (without .exe).  This feature is used for minimizing
> >     changes to Unixoid makefiles.  See also the -Zexe option of GCC
> >     and ld.
> > 
> > So it seems that neither your Makefile.in patch nor my 'os2compile' script
> > is needed. Add you need is to set CFLAGS="-Zexe" before calling configure !!
> > 
> 
> Although I don't know the package you are saying, adding -Zexe to CFLAGS
> is not a good way. Of course, I also sometimes use -Zexe mainly for
> testing build works before patching sources such as scripts and
> Makefiles. However, I don't prefer -Zexe because:
> 1. It cannot be used to build DLLs. On OS/2, -Zdll is used to build
> DLLs. -Zexe and -Zdll cannot be used at the same time. In case of
> libiconv, DLLs such as iconv2.dll cannot be built with CFLAGS="-Zexe" by
> libtool using -Zdll.
> 2. Currently, Makefile of libiconv tests requires ../src/iconv_no_i18n
> as a dependency. However, its real target is ../src/iconv_no_i18n.exe on
> OS/2.
> 3. It generates both [file] and [file.exe]. Generally, however, Makefile
> cleans only [file], so [file.exe] remains. This is not true for Makefile
> of libiconv tests.
> 
> Therefore, setting CFLAGS="-Zexe" and calling configure is not a proper
> solution.

Later you said that LDFLAGS="-Zexe" instead of CFLAGS="-Zexe" would work.

Is the issue [1] remaining then?
[2] and [3] are not very disturbing if you use "make" only for building
in batch mode; it's only when you are the main developer of a package that you
care about these, right?

> In case of os2compile:
> 1. This is slow, about 1.5 factor.
> 2. This shares [2] and [3] of -Zexe problems.
> 3. In addition, a target [file] will be generated whenever Makefile is
> called. Because [file.exe] is genereated not [file].

These seem like acceptable compromises to me. Heck, when I run a 'configure'
script in Cygwin, it runs with the same speed as a Linux machine 15-20 years
ago. Speed does not matter much here.

> Moreover, @EXEEXT@ is already used in Makefile of libiconv tests.

Yes, and the few necessary uses already clutter up the Makefile.
It's OK for Automake-generated Makefiles to use $(EXEEXT) all over the place,
but in libiconv the Makefile.in is manually maintained, and clutter is not
welcome here.

Bruno