Re: bin_hook.o and binary.o in makefile.all

[Daniel J S. <dan...@ie...>]

Ethan A Merritt wrote:
> On Saturday 10 June 2006 10:33 am, Petr Mikulik wrote:
> 
>>Somebody has added new file bin_hook.o into makefile.all. 
> 
> 
> Huh?  It was added in 2004.
> 
> 
>>This strange file  
>>does just #include "binary.c". Consequently, makefiles including 
>>makefile.all fail to link gnuplot because of multiple definitions from 
>>bin_hook.o and binary.o (e.g. makefile.mgw).

Hmm, that seems like something isn't programmed correctly.  What platform is makefile.mgw, i.e., what variables are defined for the compilation you are trying?


> According to the ChangeLog comment, bin_hook.o was supposed to replace
> binary.o

Yes, well not so much replace, but be a form of conditionally including binary.c.  But it does replace binary.c in the list of

gnuplot_SOURCES = alloc.c alloc.h ansichek.h axis.c axis.h bin_hook.c \
breaders.c breaders.h bitmap.c bitmap.h color.c color.h command.c \...


> 
>>Which one to remove from makefile.all? Or another action to take?
> 
> 
> But you are correct. That doesn't seem to be what really happened.
> So far as I can tell, bin_hook is not used anywhere and should
> be deleted altogether. Both the source file and the mention of it in
> the Makefiles are superfluous.
> 
> Daniel:
> 	This was your addition, correct?

Yes, that was from pretty far back.  I don't see how Lars' name got tied in with that.


>  Is it really correct
> that bin_hook is not referred to anywhere, in any source file?
> 
> I take it that the idea was to make inclusion of the routines in binary.o
> into the final executable be conditional on BINARY_DATA_FILE.

Yes, the reason for this hook is that binary.c is needed for two programs, that special C program to generate binary data and gnuplot in the case BINARY_DATA_FILE is not active.  When BINARY_DATA_FILE is active, from what I recall, a few extra lines of code achieved what the code in binary.c does.

The reason for the hook is that binary.c must be present for the bf_test.c program that generates binary data.  From the Makefile.am:

bf_test_SOURCES = bf_test.c binary.c binary.h alloc.c

so we can't put any type of defined conditional around the code inside binary.c, i.e., bf_test.c always needs it.  Hence, this bin_hook.c accomplishes the defined conditional for gnuplot while leaving the code compiled in binary.o for bf_test.

> It doesn't seem to have worked, and anyhow I don't think that is the
> proper way to do it.

There may be a better way.  I didn't put too much effort into a better method because I hoped that eventually gnuplot wouldn't need the code inside binary.c anymore, in which case binary.c could be dedicated simply to the C program for creating binary files (bf_test), and bin_hook.c would get the heav-ho from gnuplot_SOURCES.

>  The Makefiles produced by ./configure should either
> include binary.o, or not, depending on whether BINARY_DATA_FILE was selected.
> It should not require any extra source files.

I suppose the best thing is that based upon BINARY_DATA_FILE in ./configure, the file binary.c would or wouldn't be present in the list where bin_hook.c is.  That is:

gnuplot_SOURCES = alloc.c alloc.h ansichek.h axis.c axis.h bin_hook.c \
breaders.c breaders.h bitmap.c bitmap.h color.c color.h command.c \...

becomes

gnuplot_SOURCES = alloc.c alloc.h ansichek.h axis.c axis.h binary.c \
breaders.c breaders.h bitmap.c bitmap.h color.c color.h command.c \...

or

gnuplot_SOURCES = alloc.c alloc.h ansichek.h axis.c axis.h \
breaders.c breaders.h bitmap.c bitmap.h color.c color.h command.c \...

whichever is appropriate.  I don't know how to do that in configure.

Dan