path: root/doc/ports/notes.aix

# Begin AIX IRAF configuration.  30Mar91.
# RS/6000 Model 530, deskside, 24 Mb RAM 355+670 Mb disk, 24bit color.

# lslpp -h bos.obj says the following.  This means our current OS is
# 3001.  From the net I see that the latest release is 3003, so we
# are a bit behind.  Also, the system was configured on 19-20 Sep 90.

Option Name          State      Event      Date/Time Release         User Name
-------------------- ---------- ---------- --------- --------------- ----------
bos.obj              INACTIVE   COMMIT     02/02/90  03.01.0000.0000 root
                     INACTIVE   APPLY      09/19/90  03.01.0001.0003 root
                     ACTIVE     COMMIT     09/19/90  03.01.0001.0003 root

unix/as
unix/as.rs6000			+
unix/bin
unix/bin.rs6000			+
	Set up AS and BIN directories for the machine architecture "rs6000.
	(3/30)

unix/as.rs6000/zsvjmp.c		+
unix/os/mkpkg.sh
unix/os/mkpkg
	It won't be easy to generate the assembler version of zsvjmp.s for
	the RS6000, so I set up a C version temporarily just to get things
	to link.  The AIX CC doesn't have a -S flag and neither the AIX
	assembler or architecture is simple, so it will take some work to
	generate the assembler version.  (3/30)

unix/hlib/libc/iraf.h
	Set the iraf root to point to /u1/iraf.  (3/30)

unix/hlib/config.h
unix/hlib/libc/spp.h
	Checked that LEN_JUMPBUF was big enough for this machine; it wasn't,
	had to increase it from 64 to 65.  (3/30)

unix/hlib/irafuser.csh
unix/hlib/mkiraf.csh
unix/hlib/fc.csh
unix/hlib/cl.csh
	Made changes to reflect the new iraf root and architecture.  (3/30)

unix/shlib/mkpkg.sh
unix/gdev/sgidev/mkpkg.sh
unix/boot/rmfiles/mkpkg.sh
unix/boot/mkpkg/mkpkg.sh
unix/boot/rmbin/mkpkg.sh
unix/boot/bootlib/mkpkg.sh
unix/boot/wtar/mkpkg.sh
unix/boot/generic/mkpkg.sh
unix/boot/spp/mkpkg.sh
unix/boot/spp/rpp/mkpkg.sh
unix/boot/spp/rpp/rppfor/mkpkg.sh
unix/boot/spp/rpp/ratlibc/mkpkg.sh
unix/boot/spp/rpp/ratlibf/mkpkg.sh
unix/boot/spp/xpp/mkpkg.sh
unix/boot/xyacc/mkpkg.sh
unix/boot/rtar/mkpkg.sh
unix/os/mkpkg.sh
	Replaced all the "cc" calls by $CC, the "f77" calls by $F77, and the
	"ranlib" calls by $RANLIB.  (3/30)

unix/hlib/irafuser.csh
	Set the architecture (MACH) to rs6000.  Added definitions for CC, F77,
	RANLIB.  (3/30)

---------------
Begin bootstrap attempts.  (3/30)

unix/os/mkpkg.sh
	Changed the "rm alloc.o" to "rm -rf alloc.o".  The AIX CC doesn't
	leave the .o file behind.  (3/30)

unix/hlib/irafuser.csh
	Added compile time switches -D_BSD -D_NO_PROTO -D_BSDINCLUDES to
	make the port go a little easier.  (3/30)

unix/os/zfiomt.c
	AIX has modified the unix tape interface as follows: 1) mtio.h is
	not tape.h, 2) MTIOCTOP is STIOCTOP (ST = streaming tape), 3) struct
	mtop is struct stop, 4) all the MTfoo ioctls are now STfoo, 5),
	BSF/BSR are RSF/RSR.  Other than that things look very much the
	same, just a few gratuitous name changes.  (3/30)

unix/os/zfiopl.c
unix/os/zfiolp.c
	AIX CC insists that initializers be enclosed in { }.  Added brackets
	to the initializers for the variable xnullstr in these two files.
	(3/30)

unix/os/zgtime.c
unix/hlib/libc/kernel.h
	Changed the macro define HZ to CLKFREQ.  The former is defined in
	AIX and was causing a redefined symbol warning.  (3/30)

unix/os/zxhwen.c
	Modified the fcancel macro for the AIX stdio.h.  (3/30)

AIX
	1. The problem with pwd hanging the process in an infinite loop
	recurred.  Once this happens any pwd will hang.  Evidently the only
	way to clear the problem is to reboot.
	2. Once a task hung and after a while I get an error message about
	an NFS timeout, server felis not responding.  felis is the system I
	am working on, the file access should be direct rather than NFS.
	(3/31)

unix/os/zwmsec.c
	AIX likes signal handers to be declared type void rather than int.
	Changed the #ifdef SUNOS4 to #ifdef _AIX.  (3/31)

unix/boot/spp/rpp/ratlibc/ratdef.h
	Removed the trailing underscores from all the Fortran callable C
	externals.  AIX C and Fortran externals share the same name space
	and both are preceeded by a leading period (rather than underscore)
	in the host level symbol table.  (3/31)

unix/hlib/libc/libc.h
unix/hlib/libc/knames.h
unix/hlib/libc/xnames.h
	Same as above.  (3/31)

unix/hlib/libc/setjmp.h
	Commented out some SunOS #pragma stuff.  (3/31)

unix/mkpkg.sh
	Commented out shared library stuff.  (3/31)

unix/hlib/libc/knames.h
	Restored the trailing underscore to ushlib_,vshlib_,vshend_ to avoid
	a name conflict on vshlib in zzstrt.c.  Since these are not Fortran
	generated externals we can have a trailing underscore or not as we
	wish.  (3/31)

unix/boot/mkpkg/scanlib.c
	AIX uses COFF libraries (actually something called extended COFF)
	but for some reason they changed the <ar.h> symbols ARMAG, SARMAG,
	and SARAMAG to AIAMAG, SAIAMAG, and SAIAAMAG (they changed the R to
	IA in each name).  It was not clear what was the reason for this
	change.  A perhaps more justifiable one was that ar_fmag and ar_name
	are now elements of the union _ar_name.  With these changes the file
	at least compiles.  (3/31)

unix/os/irafpath.c
	Added #ifdef _AIX conditional to cause bin.rs6000 to be searched.
	(3/31)

unix/boot/spp/xc.c
	Modified for AIX.  The fortran compiler is xlf, host libraries are
	-lxlf, -lm, [-lbsd].  Default no shared library (yet).  (3/31)

unix/hlib/iraf.h
	Bitwise intrinsic functions are AND,OR,NOT,XOR (same as SPP!) no
	changes needed.  (3/31)

unix/hlib/mach.h
	An "od" test appears to indicate that the RS6000 byte ordering is
	big-endian (same as Sun) so no byte swapping needed.  (3/31)

AIX
	The system IS pretty fast - only takes about 10min for a bootstrap.
	(3/31)

unix/hlib/install
	1. Added #!/bin/csh at the top of the file.  The default root shell
	is ksh and ksh was trying to exec the file.
	2. Commented out the line with `mach` and added one set to rs6000.
	3. Uncommented the exit 0 to skip the Suntools stuff.  (3/31)

---------------------------
Start sysgen attempts.  (3/31)

unix/boot/spp/xpp/decl.c
	XLF cannot compile integer*[2|4] functions (types char, short, and
	long in SPP).  Upon enountering a routine declared, e.g.,

		integer*4 function foo (args)

	the compiler returns the message

	        1.17   1514-105: (S) Variable functionfoo was declared as
		array of type OBJECT-TIME.  This type of array is not
		permitted in a main program.
		1516-036: (S) Variable foo has undefined type.

	this only happens for integer*2 and integer*4 functions.  The
	Fortran manual states that only integer functions are allowed
	(excluding the other types such as real etc.).  The compiler is
	clearly treating the declaration as that of a variable rather
	than a function.

	This is potentially a very serious problem - we may not be able to
	use integer*2 if it is not fully implemented.  As a workaround I am
	going to try modifying the preprocessor to declare all integer-like
	(char,short,int,long) functions as type integer.  This should
	almost work, but will probably fail at runtime if char or short
	functions are passed in the argument list to another procedure.
	(3/31)

unix/boot/mkpkg/scanlib.c
	The archive file changes for AIX turned out to be far more extensive
	than mere name changes, and I ended up spending several hours
	figuring out how AIX archives work and rewriting this file.  A major
	misconception that it took me some time to get over was that AIX
	archives have a linked list of members, whereas UNIX systems have
	a sequential list of members.  You HAVE to follow the links.  (3/31)

AIX
	The AIX implementation of symbolic links is funny, or at least the
	"ls" task is.  A while back I had to add the -L flag to ls to get
	it to list directories pointed to by symbolic links, but with that
	flag set it will always follow links, and a command such as "ls -l"
	on a link lists the pointed to file, not the link.  So far I have
	not figured out how to get it to behave like "other" unix systems I
	am familiar with (it is not clear AIX is really a unix variant so
	I am not sure the "other" is accurate).  (3/31)

unix/gdev/iism75/zrdm75.x
	Towards the end of the file there were a couple of calls to MOD
	with mixed type arguments (short and int).  It was necessary to
	add an INT to the short arguments to make both argments integer
	before XLF would accept the statements.  (3/31)

-----------------
Completed the first sysgen, although with many errors which I haven't begun
to check out yet.  This thing IS fast - the full core system sysgen took
only 2:26.  (4/1)

AIX
	A funny thing about the initial sysgen is that it produced libraries
	and executables even in cases where there were serious errors.  This
	may be big system policy (VMS is like this too to some extent) but
	it is not standard for unix.  It causes problems because the objects,
	even though broken, were produced with recent file, dates which
	causes utilities like make/mkpkg to be useless for recompilation
	following errors.  Maybe I can find a way around this but I shouldn't
	have to.  (4/1)

unix/hlib/iraf.h
unix/hlib/libc/xnames.h
	Added definitions to map the VOS name "getpid" to "xgtpid" at the
	host library level.  On AIX Fortran and C names share the same name
	space, and a name collision on getpid was causing infinite recursion.
	(4/1)

pkg/system/help/lroff/textout.x
	This file failed to compile due to a complaint about the declaration
	for the argument out to the ENTRY point textout.  Rather than look
	into this I installed a version of the file from the AUX port which
	eliminates the entry point, and the problem went away.  (4/1)

dev/hosts
	Added columba to hosts table.  (5/22/91 SRo)

-------------------------------------
Begin AIX/IRAF update to V2.10.  (8/10/92)
AIX version 3.2.2.  Fortran version 2.2.

unix/as.rs6000/zsvjmp.s
	Added the RS/6000 version of zsvjmp.s.  It took over a day to learn
	how to program in assembler on this machine (do to limited and
	misleading documentation and tools) and about 20 minutes to write the
	routine...  (8/09)

-------------------------------------
Performed a diff/merge of the HSI and merged changes into the V2.10 HSI.
The bootstrap proceeded with only minor problems.  Started sysgen.  (8/10)

unix/hlib/mkpkg.inc
	Still tuning default fortran compilation switches.  Current set is
	as follows: "-/qhalt=e -/qflttrap -/NQ20000".  (8/10)

sys/gio/gki/gkiclosews.x
	In the call to zcall2, "Mems[...]" was changed to "Mems[gki+...]".
	(8/10)

sys/imio/imsetr.x
	There were a couple of occurrences of IM_PLFLAGS(im) where the (im)
	was missing.  (8/10)

lib/plio.h
sys/plio/*
	Assorted changes to resolve type clashes; see notes above from
	original port.  (8/10)

sys/qpoe/qpiomkidx.x
	Type clash in two calls to min, referencing type short field of
	bucket header.   (8/10)

sys/mwcs/mwtransd.x
	The AIX compiler did not like a "... * -ltv_1[i]" expression, had to
	put parens around the -ltv_1[i] term.  (8/10)

sys/vops/fftx.f
	This routine contained a declaration "x(2), y(2)" where the arrays
	X,Y are of arbitrary length.  Later on in the code the third and
	fourth elements of each array were referenced and this caused a
	compiler message about an incorrect constant array subscript.
	Changed the declaration to "x(*), y(*)".  (8/10)

sys/gio/ncarutil/agdflt.f
sys/gio/ncarutil/agdash.f
sys/gio/ncarutil/aglbls.f
sys/gio/ncarutil/agsetp.f
	The AIX compiler doesn't like the CHARACTER*504 declarations in these
	files.   Had to change these to CHARACTER*500 to get the files to
	compile.  That could be incorrect, but IRAF doesn't use this code
	anyhow (but we will keep it around until the ncar stuff is phased
	out).  (8/10)

pkg/images/tv/display/iisofm.x
	Added the "int" in "max (int(y[i]),..." to fix a type clash.  (8/10)

pkg/images/tv/iis/ids/idsinit.x
	Fixed a type clash in two calls to MIN.  (8/10)

pkg/images/tv/iis/iism70/iissplit.x
	Another type clash in a call to MIN.  (8/10)

pkg/images/tv/imexamine/iejimexam.x
	Type clash - same one as seen earlier on cephus with the DEC
	compiler.  (8/10)

pkg/images/imarith/iccclip.x
pkg/images/imarith/icsclip.x
pkg/images/imarith/icstat.x
pkg/images/imarith/icaclip.x
	More type clashes, again the same as seen on cephus.  (8/10)

pkg/plot/t_gkidir.x
	Another short/int type mismatch for MIN (that has been that way since
	the dawn of time).  (8/11)

noao/lib/mkpkg.inc
noao/lib/mkpkg.sf.AIX3		+
	Added artdata/t_mkechelle.x and astutil/t_gratings.x to the NOAO
	package special file list.  These files have to be compiled with the
	optimizer turned off on the RS/6000 or the compiler generates the
	following message.

	    1500-008: (S) COMPILER LIMIT EXCEEDED in astgrg: Program too
	    complicated to be compiled. Compilation ended. Reduce the
	    complexity of the program and recompile, or lower the level of
	    optimization and recompile.
	(8/11)

noao/digiphot/daophot/daolib/erf.x
noao/digiphot/daophot/daolib/dpevalpsf.x
noao/digiphot/daophot/psf/dpgaussfit.x
noao/digiphot/daophot/psf/dpsubgauss.x
noao/digiphot/daophot/psf/dpwritepsf.x
	This code defines and uses a function called ERF, but ERF is a
	Fortran intrinsic function (at least on the IBM compiler).  Changed
	the name to DAOERF.  (8/11)

noao/onedspec/dispcor/refgspec.x
	Mixed real/double in call to MOD.  (8/11)

noao/onedspec/splot/fixx.x
noao/onedspec/shdr.x
noao/onedspec/t_deredden.x
	Type clashes in intrinsics, already found and fixed on decstation.
	(8/11)

noao/onedspec/t_deredden.x
	This file had another problem on the RS/6000 - syntax error in
	expression.  Had to add parens in "y * (-2.09002)".  (8/11)

noao/rv/coloncmds.x
	The IBM compiler (like the decstation) also complained about cmd_write
	being a function but no value was being returned.  (8/11)

noao/rv/fftutil.x
	The IBM compiler generated the following messages for this file:

	    1514-047: (S) No initializations for this DATA statement will be
	    done because of presence of syntax error.
	    1515-019: (S) Syntax is incorrect.

	Evidently this was due to the use of the name "data" for an argument
	to the procedure fft_fixwrap.  There was nothing wrong with the code,
	the compiler was in error in this case.  To workaround the problem
	I changed the name of the data vector "data" to "v" in fft_fixwrap
	and in the previous procedure.  (8/11)

noao/imred/ccdred/src/icstat.x
	Type mismatch problems, already run into on cephus.  (8/11)

noao/imred/vtel/mscan.x
	The function mtneedfileno() was declared boolean but used as an integer
	function (which it is).  (8/11)

unix/os/zzstrt.c
unix/os/zxwhen.c
	Added #ifdef _AIX code to these routines to enable and handle the
	IEEE exceptions.  This is quite difficult on the RS/6000, compared
	to most systems.  There really aren't any IEEE exceptions on this
	machine.  To catch things like floating divide by zero it is necessary
	to compile code -qflttrap, which causes a test-and-branch and trap
	instruction to be compiled in line *after every floating point
	instruction*.  If the floating instruction overflows, underflows,
	or whatever, the branch will be skipped and the application generates
	a SIGTRAP.  The ex_handler routine in zxwhen gets called by the trap.
	At that point the floating point status register can be examined to
	determine what "exception" occurred.  This appears to work but I
	am still a bit uncomfortable about all this; the IBM documentation
	is not very definitive and thus far I am unable to determine exactly 
	how the RS/6000 hardware works (e.g., no where does it say exactly
	what happens when, for example, floating underflow occurs).  (8/11)

unix/boot/mkpkg/scanlib.c
	While relinking the CL to test the exception handling code I noticed
	a problem with mkpkg not reading the libpkg.a file.  This was traced
	to a problem in scanlib.c, used to read AIX archive files.  The code
	assumed that the first library member followed the file header.  This
	is normally, but not always the case.  In reality the library member
	list is a linked list and the offset to the list head is stored in
	the file header.  (8/12)

unix/hlib/mkpkg.sf.AIX3
	I had a lot of trouble getting the integer divide-by-zero exception
	to work.  Finally I determined (by studying assembled code with adb,
	not from the info-hider) that the compiler needs to generate extra
	instructions to check for a divide by zero.  The Fortran compiler
	does this by default but the C compiler does not.  Studying the
	C compiler documentation I could not find a switch to enable this
	option, but fortunately a query to usenet revealed that "-qcheck"
	was what I was looking for.  I inserted this in hlib$mkpkg.inc and
	tried to rebuild the CL and got a number of messages such as the
	following.

	compile.c:
	*** PROGRAM ERROR *** No OTHERWISE or WHEN for execution in SELECT...
	*** PROGRAM ERROR *** No OTHERWISE or WHEN for execution in SELECT...
	*** PROGRAM ERROR *** No OTHERWISE or WHEN for execution in SELECT...
	cc: 1501-230 Internal compiler error; please contact your IBM repr...

	Obviously some sort of compiler bug (C doesn't have these statements).
	The workaround was to add only unop.c and binop.c, the CL files that
	evaluate arithmetic expressions in scripts, to mkpkg.sf.AIX3.
	Fortunately these files compile without the error messages.  (8/12)

unix/os/zgcmdl.c
	Modified this file to pick up the argument list on AIX systems.
	There is no portable way to do this, one does it by linking an
	executable and studying the symbol table with nm and adb, looking
	for something which will give the address of the argument list.
	So far every system has proved to have a way to do this (on AIX
	there is a symbol called p_argv).  (8/12)

unix/os/alloc.c
	AIX magtape devices have names like "rmt.1", so I had to change
	the code which checks for simple filenames to permit period.
	This still excludes names like ../path, since / is not permitted.
	(10/17)