SUN/IRAF V2.3 Installation, 14 April 1986 Sun-3, software floating point ------------------------------------------------------------------------------ /usr/iraf/ Unpacked a source-only TAR archive made on lyra this afternoon; didn't first delete anything. Made sure we had a .login with $iraf defined as the local IRAF root (/usr/iraf/). (4/14) /usr/include/iraf.h As Superuser, "ln -s unix/hlib/libc/iraf.h ." (4/14) /usr/bin/alloc -> $hlib/alloc.e /usr/bin/generic -> $hlib/generic.e /usr/bin/mkpkg -> $hlib/mkpkg.e /usr/bin/rmbin -> $hlib/rmbin.e /usr/bin/rmfiles -> $hlib/rmfiles.e /usr/bin/rtar -> $hlib/rtar.e /usr/bin/wtar -> $hlib/wtar.e /usr/bin/xc -> $hlib/xc.e /usr/bin/cl -> $hlib/cl.e As Superuser, performed these symbolic links. (4/14) unix/hlib/mkiraf.csh unix/hlib/libc/iraf.h Modified IRAF root pathname, HOST definition; irafuser.csh not modified, as $iraf is defined in the IRAF account .login. (4/14) unix/hlib/mkmlist.csh Changed bgrep to grep, since bgrep not available in standard UNIX. (Not important, doesn't have to work, so the efficiency of bgrep on the development machine is worth the mach. dependence.) (4/14) as/ vaxas/ mc68000/ Moved as/ to $iraf/vaxas/; moved $iraf/mc68000/ishift.SUN, zsvjmp.SUN to as/ishift.s, zsvjmp.s. (Now that we have a SUN-3 with 68020, the BYTPAK etc. assembler sources should also be implemented). (4/14) unix/hlib/mach.h unix/hlib/libc/spp.h (EPSILON) Compiled osb$zzeps.f with "f77 -O -f68881 -o zzeps.e zzeps.f"; ran it to check hardware floating point epsilon; turned out to be same as software floating point to 4 places. Modified machine epsilon and byte-swapping definitions (see unix/mc68000/README). (4/14) unix/hlib/[dir]1mach.f Commented out the VAX lines and uncommented the IEEE floating point lines for the 68000. (4/15) unix/hlib/mkpkg.inc Modified XFLAGS to "-c -O -f68881" for our hardware floating point option. To verify that this works, I examined zzeps.e, compiled with the same flags, with adb and found that, indeed, only single arguments were being passed and actual floating point instructions were being generated. Set USE_CALCOMP to NO. (4/15) unix/hlib/libc/kernel.h Commented out DEBUGMEM in order to use the standard UNIX malloc rather than IRAF's version. This is necessary for some reason on the Sun for SunCGI routines to be callable from SPP programs. (4/15) unix/hlib/libc/iraf.h Added definition for SUN3; the Sun IRAF kernel is still mostly identical with the VAX kernel, but zxwhen.c needs to know if it is being compiled on a Sun-3. (4/15) unix/ Attempted a bootstrap; had to make the following modification to get it to work: (4/15) unix/os/zxwhen.c Renamed structure vax_exception to hwx_exception; added ifdef branch for SUN3 FPE traps-to-strings structure. (The version that compiled under the Sun-2 failed to compile under the Sun-3.) (4/15) unix/ Performed a bootstrap with "sh -x mkpkg.csh >& spool &". The only error showing was in bootlib/mkpkg.csh, "syntax error at line 13: `end of file' unexpected", presumably due to the conditional construct, and possibly a CSHELL bug (libboot.a built correctly and ended up in $hlib as it should anyway). (4/15) unix/hlib/alloc.e This file must belong to root, since it needs root permissions at runtime to allocate devices. Do the following: % cd $hlib % su % /etc/chown 0 alloc.e (4/15) ------------------------------------------- The bootstrap is now complete. The next step is to do a full system compile. Attempted a "mkpkg >& spool &" from the root directory. Got compiler failure in fio$; "Compiler error line 0 of (null): invalid flag s...y...s...t...e...m...=" when trying to compile fowner.f. Also got similar errors in fmtio$gstrcat.f and other places. No further files from the original XC command line were compiled. Common denominator was the length of the command line to the f77 compiler. Made the following change: (4/15) unix/boot/mkpkg/host.c Changed SZ_CMD from 512 to 400; this shortens the command line that MKPKG constructs for XC. Rebuilt MKPKG with "sh -x mkpkg.csh". (4/15) Started another sysgen. Noticed compiler errors on the following files; let the sysgen continue while fixing them. sys/fmtio/gctod.f sys/fmtio/xtoc.f These routines had to be hand-compiled without switch "-f68881". Made copies of the fortran in local/bugs, and hand-compiled each routine in its own directory. Something to do with complex numbers? Since it didn't look like they affected anything else, just did a "mkpkg" to load them back into libsys and let the sysgen continue. (4/15) sys/vops/ak/abeqkx.f sys/vops/ak/abeqx.f sys/vops/ak/abgekx.f sys/vops/ak/abgtkx.f sys/vops/ak/ablekx.f sys/vops/ak/abltkx.f sys/vops/ak/abnekx.f sys/vops/ak/abnex.f sys/vops/ak/advzx.f sys/vops/lz/allnx.f sys/vops/lz/alogx.f sys/vops/lz/arcpx.f sys/vops/lz/arczx.f Same with these routines; again, all contained complex datatypes. Hand-compiled, then a "mkpkg" from vops. (4/16) unix/hlib/mkpkg.inc Changed LFLAGS from "" to "-f68881" to get pick up the references generated by the same switch during default compilation. This may cause a problem with those files hand-compiled without this switch due to compiler errors. (4/16) pkg/dataio/reblock/t_reblock.x line 56 pkg/dataio/cardimage/t_wcardimage.x 39 pkg/images/imutil/imdelete.x 43 pkg/onedspec/t_widstape.x 65 pkg/plot/gkiextract.x 99 pkg/utilities/t_translit.x 41, 68 pkg/imred/vtel/dicoplot.x 324 pkg/imred/vtel/tcopy.x 59 pkg/imred/vtel/vtfix.x 60 pkg/imred/vtel/writevt.x 42 All these lines tested a boolean conditional against "true" or "false". The Sun-3 f77 compiler refused to compile these; all were changed to just the name of the boolean or !boolean as appropriate. This was only done on the Sun, since this is an experimental version; mail will be sent on lyra. (4/16) pkg/images/tv/display/iisers.x No change made, since there is no need for this code on the Sun, and it looked like there were non-obvious consequences in other routines for changing either the values or changing shorts to ints. However, note should be taken on lyra that "erase" is a short integer, and it is being assigned to ERASE (in one of the local include files), which has the value "32768" -- too high by one, apparently. pkg/images/tv/cv/iism70/iiscursor.x pkg/images/tv/cv/iism70/zclear.x Also no change made, but similar problem: the include file definitions IREAD (100000B) and ERASE (32768) are being used in assignments to short integers, in one case "32768+256". (4/16) unix/boot/spp/xc.c Added references for the standard Fortran libraries -lI77 and -lU77, plus a reference to the temporary library $hlib/libk.a, which contains several routines referenced by the Fortran compiler, but not provided in any SUN library. See below. (4/25) unix/hlib/SUN_kludge/ This a temporary directory added to build the kludge library libk.a so that IRAF programs can be linked on the SUN-3. Source for the following routines were added: Fc_eq compare complex numbers for equality Fc_conv_h convert complex to short Fh_conv_c convert short to complex The conversion routines appear to have been overlooked by SUN, and should be added to /usr/lib/libF77.a. (4/25) unix/mkpkg.csh Added a line to compile the kludge library during the bootstrap. (4/25) unix/hlib/mkpkg.inc Added the flag "-f68881" to LFLAGS (the linker flags). This is necessary to get the linker to reference /lib/Mcrt1.o rather than /lib/Fcrt1.o. If this is not done and one attempts to link code compiled with the -f68881 switch, an unresolved reference to the external "f68881_used" will result. (4/25) sys/osb/miiupk16.x sys/osb/miiupk32.x Replaced by the new versions from lyra, which fix the BYTE_SWAP bug. (4/25) sys/gio/nspp/portlib/flush.f sys/gio/nspp/portlib/flash[13].f sys/gio/nsppkern/gktcancel.f sys/gio/nsppkern/gktclear.f sys/gio/nsppkern/gktclose.f sys/gio/ncarutil/sysint/spps.f Changed the name of the NSPP routine "flush" to "mcflsh" to avoid a library conflict with the Fortran FLUSH. (4/25) pkg/images/tv/display/iisers.x pkg/images/tv/cv/iism70/iiscursor.x pkg/images/tv/cv/iism70/zclear.x All these files produced a compiler bug due to a compile time assignment of an unsigned short integer number 1XXXXX into a short variable. Had to change these to run time assignments to get the desired truncation assignment. (4/25) local/.login local/login.cl Aliased XC to 'xc -f68881' since all floating point is handled that way on this machine. (4/25) unix/hlib/libc/finfo.h The fi_owner field of the finfo structure was not dimensioned properly, causing the storage allocated for the structure to be overrun. This was found in a call to pfileopen/filetime in the CL; the finfo structure is allocated space on the stack, hence the stack was being corrupted. This was a fairly serious bug which has been in the system for a long time. (4/25) [[ The system is now up (4/25) ]] dev/pix dev/pix.pix Moved the test image over from lyra: wfits dev$pix /u2/tody/pix.fits (on VAX lyra) rfits lyra!/u2/tody/pix.fits (on SUN) We ought to add a dev$pix.fits file so that the test image is available on source only archives. (4/25)
