nsb@THUMPER.BELLCORE.COM (Nathaniel Borenstein) (08/10/90)
OK, so my 3100 compilation finally gets as far as trying to link the runapp binary. The Imakefile for atk/apps is set up so that it tries to get the X11 library simply by using -lX11. Great. Again, we get the non-libg0 version. The fix I found is, in atk/apps/Imakefile, to replace "-lX11" with "$(XLIBDIR)/libX11.a". I trust something like that will be in the next patch. I really don't understand how anyone ever managed to compile Andrew on the 3100 without hitting all these problems. Well, now I'll type "make World" again and see how far it gets this time... -- Nathaniel
cmf@UNIX.CIS.PITT.EDU ("Carl M. Fongheiser") (08/10/90)
Nathaniel Borenstein <nsb@thumper.bellcore.com> writes: > I really don't understand how anyone ever managed to compile Andrew on > the 3100 without hitting all these problems. Well, now I'll type "make > World" again and see how far it gets this time... -- Nathaniel I seriously doubt anybody *did* compile Andrew without hitting these problems. I sure didn't. I fought with it for three days, at least 1 of which was just making sure I had all the pieces to build a -G 0 library. It was that frustration that led me to start work on the G0 converter. Unfortunately, it's still a ways from working right. Once you've finished though, it's well worth it. The fastest fad animation I've ever seen was on my DECstation 2100. The 3100 can only be better. Carl Fongheiser cmf@unix.cis.pitt.edu
sean@DSL.PITT.EDU (Sean McLinden) (08/10/90)
> OK, so my 3100 compilation finally gets as far as trying to link the > runapp binary. The Imakefile for atk/apps is set up so that it tries to > get the X11 library simply by using -lX11. Great. Again, we get the > non-libg0 version. > > The fix I found is, in atk/apps/Imakefile, to replace "-lX11" with > "$(XLIBDIR)/libX11.a". I trust something like that will be in the next > patch. I got around it by defining the -L option at the level of the config files. Basically, the following is what I did to get things to work: in config/site.mcr: MIPSLIBC_G0 = /systems/pmax/usr/lib/mips_G0/libc_G0 XLIBDIR = /systems/pmax/usr/lib/mips_G0 XSRCDIR = /sources/pmax/X11/mit/ XUTILDIR = /systems/pmax/usr/bin/X11 XBASEDIR = /systems/pmax ROOTDIR = /systems/pmax BASEDIR = /systems/pmax/usr/andrew DEFAULT_ANDREW_ENV = /systems/pmax/usr/andrew MAKEDOFLAGS = -L/systems/pmax/usr/lib/mips_G0 ${MAKEDODEBUG} \ -d ${BASEDIR}/lib -b ${BASEDIR}/bin in config/allsys.mcr: (this is needed because these are not read, properly, from the other files in time to make the Makefiles correctly. XBASEDIR = /systems/pmax XSRCDIR = /services/X11/mit/ BASEDIR = /systems/pmax/usr/andrew DESTDIR = /systems/pmax/usr/andrew in config/pmax_3/system.mcr: CC = cc -Dvoid=int -G 0 "-Wl,-D,1000000" \ -L/systems/pmax/usr/lib/mips_G0 Also, if you define AMDS, you'll need to get "mpp" (why anyone chose to implement Yet Another Macro Preprocessor when there are at least half a dozen floating around for Unix is beyond me, (let alone why it is copyrighted by IBM and part of the Transarc code)! > I really don't understand how anyone ever managed to compile Andrew on > the 3100 without hitting all these problems. Well, now I'll type "make > World" again and see how far it gets this time... -- Nathaniel If you have already gotten past the build of the examples, the make world will fail because it quits if the ".eh" files already exist (which I why I also remove /usr/andrew before a "make world"). Also, unless you figure out something that I didn't, you'll need to actually run X11R4 (not DECWindows), to get Andrew to run. Otherwise the applications will start but the fonts, cursors, and icons will be screwed up. Don't forget that building AMDS will destroy your system mailer unless you make the BASEDIR changes (this might have been fixed with Patchlevel 6), since it overwrites /usr/lib/sendmail, /bin/mail, and other things without asking and NOT part of "make install" but rather as part of "make world"). Other than that it was easy. Sean
guy@auspex.auspex.com (Guy Harris) (08/11/90)
>The fix I found is, in atk/apps/Imakefile, to replace "-lX11" with >"$(XLIBDIR)/libX11.a". I trust something like that will be in the next >patch. Make it an option, please; I originally submitted the change to use "-lX11" in order that, when building on systems where "-lX11" gets you a shared library, "runapp" gets built with a shared library.
datri@convex.com (Anthony A. Datri) (08/11/90)
>of which was just making sure I had all the pieces to build a -G 0 library
I've been watching this for a while, waiting for someone to leak a clue
as to what a G0 library is. Could one-who-knows please do so?
--zs01+@ANDREW.CMU.EDU (Zalman Stern) (08/11/90)
Yeah, building ATK on a pmax is a bitch. But many many patches ago, it did build cleanly at lesat once for me following the directions in the README file at the top of the tree. Gary is working on building ATK on his pmax so this stuff should get better. How its supposed to work: You put all the -G 0 libraries in some directory. (The libs that you need are enumerated in the documentation.) If that directory is not $(BASEDIR)/lib/mips_G0, then you must edit the copy of makedo.csh in the source tree. (This is also documented.) Makedo adds this directory to the library search path via a -L switch to ld. External libraries used in makedo lines should use -l instead of absolute library paths. (Note: this may lose on systems that don't support the -L switch to ld. The only real alternative is to replace all explicit library references with macros. To support the -G 0 crud, there will have to be one macro for makedo lines and one for other purposes.) I'm not sure how I got region.do to work. A good fix is to change the Imakefile to use -lX11 instead of $(XLIBDIR)/libx11.a (see aove note). It shouldn't hurt to have runapp linked against the non -G 0 X11 library. Code in runapp can be linked against standard libraries without ill effect. (Unless your standard libraries and -G 0 ones are incompatible, in which case all bets are off.) The MIPSLIBC_G0 docs look pretty broken. In addition it might increase the general sanity in the world to make the macro contain the .a instead of the code in the Imakefile. (On the other hand, changing it now might be more hassle than its worth.) For similar fun, try building CMU CS software on systems that don't support the CPATH and LPATH environment variables sometime... Disclaimers: I don't hack on ATK any more. I hate (I)makefile hacking more than anything else I have to do on a regular basis. Sincerely, Zalman Stern | Internet: zs01+@andrew.cmu.edu | Usenet: I'm soooo confused... Information Technology Center, Carnegie Mellon, Pittsburgh, PA 15213-3890 *** Friends don't let friends program in C++ ***