Re: Fun with auto generated dependencies & clean...

[Rick Flower]

Mike Shal wrote:

> The problem is you're building the dependencies before each build,
> rather than during the build. This is because you've made a rule for
> the dependency files (%.d: %.cc) and you're including them. It is only
> necessary to have the dependency files available for the *next* build.
> See: http://make.paulandlesley.org/autodep.html

Sounds good.. I'm still reading the above site.. Thanks for the link

> > %.d: %.cc
> >         @echo "Building depends..."
> >         $(target_gpp) -MM $(GPPFLAGS) $< \
> >         | sed "/.*\.o:/ s%^[ ]*%\$$(OBJDIR)/%" > $@
>
> I would get rid of this rule, or perhaps move the commands into the
> %.o: %.cc rule so the .d and .o files are built at the same time.

I'm working on this..

> > $(OBJDIR)/%.o : %.cc %.d
> >         @echo "Prerequisites are: $^"
> >         $(target_gpp) $(GPPFLAGS) -c $< -o $@
>
> Here I would remove %.d from the rule, since the .o file doesn't need
> the .d file to be built (again, you just want the .d to hang around
> for the next time you build). If you moved your commands from the .d

Done..

> rule here, you would always create the .d file when you create the .o
> file. Or, assuming you're using a recent version of gcc, look into the
> -MMD flag. This flag will create the .o and .d simultaneously (thus
> saving you an extraneous preprocessing step). Also look at the other
> -M flags (like -MT, -MF, etc) to get the .d file in the location you
> want it, and have it specify the correct rule. IIRC these flags
> changed behavior a few times during the 3.X releases, but I think
> they're stable now.

I'm using some pretty old compilers (gcc-2.95.3 for ppc and gcc-2.9 for 
pentium) and both seem to handle -MMD properly and generate the 
requisite .d files, but neither version knows anything about -MF or -MT 
so I guess I'll have to do without.. Normally in the past we would 
create a .depends_${target} file for each src directory we had for our 
sun or ppc builds and generate them before building the objects for that 
directory.  If we go and switch paradigms here and have the compiler 
both build and generate depends at the same time, is it safe to rename 
the files to be something like file1.d_${target} and is there some way 
to teach Make to look for them instead of file1.d without getting into 
the same sort of issues I've got now? Or... Can I still use the 
monolithic single file for all depends and just concatenate each ".d" 
file contents into the bigger file after building each object?  The only 
thing gained by doing that is less clutter I suppose.. I'm not sure how 
 my other group members are going to like the extra files since they're 
used to a single file that we've had up to now for each directory..

> Not sure if that necessarily helps with the issue of specifying both
> clean and all at the same time (it seems like you would still have
> issues with -j), but it would be more efficient :)

It's been very interesting. I'll see where this all takes me.. Thanks 
for the suggestions/info so far!