rogerj@batcomputer.tn.cornell.edu (Roger Jagoda) (04/13/90)
Folks, Here are some bugs and problems I and other users here have collected for a while now. Has anyone seen anything like these before and has suggestions/work-arounds/patches/prayers. Thanks in advance. The bugs: By default, the librarian formats nroff files without pipeing them first through tbl. This is a problem for some manual pages, and for much of the sybase documentation. The man command also fails to nroff some man pages correctly as it leaves them with the wrong permissions (formatted man pages need to be flagged 444, these go into /usr/man/cat1, while nroff source files need to go in /usr/man/man1 and need to be flagged 666). Sometimes man outputs files flagged 600 and users get "permission denied" errors. Running installman, or changing the permissions by hand fixes the problem, but it is a bug. There is a bug in the su command which corrupts the environment table inherited by any sh scripts that one might run, causing them to crash. The bug is always repeatable, and there is an error message (described) below, that occurs each time an sh shell script is run. When the command "su - [username]" is used to simulate a full login as another user (for any user, including root), the environment is corrupted by the addition of an incomplete symbol. The symbol is the value of the pre-su environment variable $TERM (for example, vt100). When a sh script is subsequently run, it exits immediately with the error message ``$TERM: is not an identifier'': for example, from a Terminal application window % su - password: # echo $TERM vt100 # printenv HOME=/users/<original user's home> . vt100 . TERM=vt100 . # sh vt100: is not an identifier # Note that a legitimate environment variable consists of a symbol, an equal sign, and a value (as, in the example above, are HOME and TERM). The entry "vt100" is not a legitimate symbol, and sh does not recognize it and aborts. CC compiler bug -- A bug in the compiler involves the generation of bad instructions by the optimizer. While optimizing, the compiler can generate instructions that the assembler ignores with a warning message. The following short test program demonstrates the bug: ----CUT HERE---- /* This is about the smallest I can make this program and still have it fail. One thing I have noticed, however, is that if I use the following prototype and the error goes away: void polint(float *, float *, int, float, float *, float *); Compile as % cc -c -O xpolint.c /tmp/cc000341.s:31:"instruction/operands mismatch" -- Statement 'fmovel fps,a5' ignored */ /* Driver for routine POLINT */ #include <stdio.h> #include <math.h> void polint(); float *vector(); #define NP 10 /* maximum value of n */ #define PI 3.1415926 main() { int i,n,nfunc; float dy,f,x,y,*xa,*ya; xa=vector(1,NP); ya=vector(1,NP); scanf("%d",&n); for (nfunc=1;nfunc<=2;nfunc++) { if (nfunc == 1) { for (i=1;i<=n;i++) { xa[i]=i*PI/n; ya[i]=sin(xa[i]); } } else if (nfunc == 2) { for (i=1;i<=n;i++) { xa[i]=i*1.0/n; ya[i]=exp(xa[i]); } } for (i=1;i<=10;i++) { if (nfunc == 1) { x=(-0.05+i/10.0)*PI; f=sin(x); } else if (nfunc == 2) { x=(-0.05+i/10.0); f=exp(x); } polint(xa,ya,n,x,&y,&dy); printf("%12.6f %12.6f %12.6f %4s %11f\n", x,f,y," ",dy); } } } ------CUT HERE------ Compile this program with the optimizer, as follows % cc -c -O main.c /tmp/cc002574.s:69:"instruction/operands mismatch" -- Statement 'fmovel fps,a5' ignored The warning does not occur if optimization is not enabled. BTW, this program is an abbreviation of part of the test suite for ``Numerical Recipes in C'' by Press, Flannery, Teukolsky, and Vetterling. WorkSpace Manager bug -- There is a problem with the way WindowManager deals with disk mount-dismount events. When an Optical is mounted or unmounted from WorkSpace Manager, a DPS window comes up from the console (even echos su events too!) asking to insert the OD or press 'n' if it's not available. Well, sometimes this works, and other times it doesn't. When it doesn't though, EVERYTHING is frozen, you can't even move the mouse to click anything else. You CAN, however, remote login from somewhere and perform noremal unix commands...EXCEPT restart the Window Manager. You can remove the offending OD, but the WM will still be stuck solid. a Kill -HUP will not help (no real reason it should). The only thing left to do at that point is reboot which seems rather silly. Thanks for listening! --Roger Jagoda --Cornell University --FQOJ@CORNELLA.CIT.CORNELL.EDU -- ------------------------------------------------------------------------------ Roger Jagoda -- My employers don't even like paying Cornell University me, let alone accept responsibility fqoj@cornella.cit.cornell.edu for anything I say or do! --
judd@blake.acs.washington.edu (Randall Judd) (04/13/90)
In article <10102@batcomputer.tn.cornell.edu>, rogerj@batcomputer.tn.cornell.edu (Roger Jagoda) writes: > > WorkSpace Manager bug -- > > There is a problem with the way WindowManager deals with disk > mount-dismount events. When an Optical is mounted or unmounted from > WorkSpace Manager, a DPS window comes up from the console (even echos su > events too!) asking to insert the OD or press 'n' if it's not available. > Well, sometimes this works, and other times it doesn't. When it doesn't > though, EVERYTHING is frozen, you can't even move the mouse to click > This sounds similar to a problem I just had with an optical disk. The scenario goes something like this. Remove new optical from shrink wrap start initialize name disk and press ok when directed insert disk cube locks up, gray window does not go away, mouse does not move I get out of this with the (right) command tilde key and power off (it take three tries with various messages for each try). I did this three times on my cube, and once on another cube. I have since bought another optical and initialized it on my cube with no problem, so it appears the problem is with the disk and not with the cube, except for what must be some kind of bug with the initialize. Has anybody else seen this or is there a fix out there? To email to me please use judd@sunspot.nosc.mil. Thanks, Randy Judd