Re: config/xconfig problems (was Re: [Math-atlas-devel] Any advice on IRIX dynamic linking?)

[<rw...@cs...>]

Allen,

>OK. I've now tried config (several times over) and found some difficulties:
>    A. config won't take the length of flags et al required; a patch (see
>       below) was necessary. The CC/CCFLAGS/etcetera that I _suspect_ are
>       advisable/required are as follows:
>
>FLINKER: /usr/bin/f77 -n32 -mips4 -L/usr/share/rutgers/lib32/r12000 -L/usr/share/rutgers/lib32 -L/usr/lib32/mips4 -L/usr/lib32 -L/lib32 -Wl,-rpath,/usr/share/rutgers/lib32/r12000:/usr/share/rutgers/lib32:/usr/lib32/mips4:/usr/lib32:/lib32
>
>FLINKER FLAGS: -O3 -LANG:recursive=ON -OPT:Olimit=0:got_call_conversion=OFF:IEEE_arithmetic=1 -LNO:blocking=OFF:opt=1:ou_further=3 -TARG:platform=IP30:processor=r12000 -D_SGI_SOURCE -D_SGI_REENTRANT_FUNCTIONS -I/usr/share/rutgers/include

You reckon you've thrown enough flags at the problem?  Any particular
reason you threw half the flags into the compiler name, rather than putting
them in the flag section?  I seem to vaguely recall that there are problems
on some systems when you dump the flags into the compiler rather than
the flags.  In this case, why not put all those freakin -L flags into
your LD_LIBRARY_PATH?

>	B. Problem 2:
>Assertion system(ln) == 0 failed, line 881 of config.c
>Unable to figure F2C data
>
>Calculated F77/C interoperation conventions:
>   Unable to determine naming conventions
>   Unable to determine F77/C integer correspondence
>   Unable to determine F77/C string interoperation

Have you tried it with the compiler macros set to compilers, with the
flags in the flag macros?  You can remove the redirection so it prints
the error message to the screen by changing redir to "" on line 3980 of
config.c (at least I think that'll work).

>       C. Problem 3:
>The F77 and F77 FLAGS setting is automatically done, not overridable by the
>FLINKER and FLINKER FLAGS set manually within config, and
>incorrect/incompatible with the rest. If ATLAS' testing code requires F77
>(as opposed to FLINKER) to be set correctly to work, then this will need
>fixing. (I couldn't use the command line overrides because of the line
>lengths.)  It is also quite possible indeed, probable at this point - that
>the F2C problem above is due to that the F2C code (findF2C) is using _all_
>of the "F77, F77FLAGS, FLINK, FLFLAGS, CC, CCFLAGS" variables, and a -64
>compile will not work with a -n32 link, nor would a -64 compiled Fortran
>program be able to interface with a -n32 C program.

Yeah, it uses F77.  If you are gonna replace the flags with a dissertation,
go ahead and edit the generated file by hand (the F77/C interface shouldn't,
I wouldn't think, change depending on static/dynamic, so copy them from
defaults).

>Below are the various logfiles and the patch. (A thank you to Yozo Hida for
>the patch mentioned in ATLAS' errata, which set me on the right track on
>this one.) Incidentally, the patch also turns off looking for GCC on IRIX -
>a modification to sped up xconfig; this should probably actually be a
>command-line option... but /usr/bin/cc is the current default compiler for
>ATLAS on IRIX anyway, so looking for gcc doesn't seem to make much sense.

Yeah, maybe you wanna turn off the optimization flags: that speeds up
install even further!  Actually, shouldn't cost you anything on an SGI,
but on most systems, not having a working gcc is bad news . . .

Cheers,
Clint