bill@ur-cvsvax.UUCP (Bill Vaughn) (09/09/85)
DESCRIPTION:
Awk's version of 'printf' core dumps or prints garbage when a field width
variable is used. The manual page refers to printf(3S) so I assume this
feature of printf should be implemented.
REPEAT BY:
Examine the following script. Try it yourself.
Script started on Mon Sep 9 12:58:06 1985
% cat test.c
main(){printf("%*s\n",5,"a");}
% cc test.c ; a.out
a
% echo 5 |awk '{printf("%*s\n",$1,"a")}'
l$l3lll)l:lHlSl]lhl(m%n
% echo 5 |awk '{printf("%*s\n",5,"a")}'
Segmentation fault (core dumped)
% ^D
script done on Mon Sep 9 13:01:08 1985
FIX:
I don't have one. Can anyone out there help?
I haven't been able to kludge around this problem as yet.
_________________________________
\ Bill Vaughn /
\ Center for Visual Science /
| University of Rochester |
/ Rochester, NY 14627 \
/seismo!rochester!ur-cvsvax!bill\
---------------------------------naiman@pegasus.UUCP (Ephrayim J. Naiman) (09/11/85)
<munch, munch> > Awk's version of 'printf' core dumps or prints garbage when a field width > variable is used. The manual page refers to printf(3S) so I assume this > feature of printf should be implemented. I also found this bug a long time ago and gave up on a fix. Incidentally, I have a version of the unreleased new awk and it also core dumps. I hope someone is listening. -- ==> Ephrayim J. Naiman @ AT&T Information Systems Laboratories (201) 576-6259 Paths: [ihnp4, allegra, mtuxo, maxvax, cbosgd, lzmi, ...]!pegasus!naiman
mp@allegra.UUCP (Mark Plotnick) (09/13/85)
As long as we're telling our favorite awk bugs, here's mine.
(Karl Heuer, whuxlb!kwh, discovered this bug)
In run.c, routine program(), there's a call to tempfree(x) at a point where
x may still be uninitialized if the awk script doesn't have any
BEGIN actions and doesn't have any input.
awk 'END { print NR }' </dev/null
will run into the problem.
On 4.2bsd, x is an automatic variable (a struct), and the garbage it
contains will usually cause nothing bad to happen.
But, on SVR2, x is a register ptr. On a vax it's register 10, and
its value is inherited from whatever was in register 10 at the time awk
was exec'ed (e.g. the variable p in the execs() routine in the shell).
When run under ksh, the above awk program gets a memory fault. When
run under sh or sdb, it works OK. Do you know what a pain it is to have
a program fail under the shell and work fine when run under a debugger?
Mark Plotnick
allegra!mpvijay@topaz.RUTGERS.EDU (P. Vijay) (09/13/85)
> DESCRIPTION: > Awk's version of 'printf' core dumps or prints garbage when a field width > variable is used. The manual page refers to printf(3S) so I assume this > feature of printf should be implemented. > > REPEAT BY: > % echo 5 |awk '{printf("%*s\n",$1,"a")}' > Segmentation fault (core dumped) CAUSE: All printf(fmt,v1,v2,...,vn) statements are tokenised. Eventually, format() is called for each individual tokens, with the format string pointer being updated every invocation. The format string is parsed in a rather simplistic fashion, which does not look out for '%*' like constructs. So, a statement like 'printf("%*s",$1,"a")' results in a call 'sprintf("%*s",$1)', which ends up accessing some bogus pointer somewhere, thus dumping core. FIX: Make the format string parsing more sophisticated, and set booleans to indicate presence of variable fieldwidth and/or precision. Also, conditional to these booleans, call sprintf with appropriate arguments. CAVEAT: Since this situation does not occur very often ("%*" in printf), and since printf itself is typically a heavily called routine in awk programs, the parsing for '%*' is done in a not so robust fashion. Calls such as 'printf("%**.22s\n",$1,"Foo")' might cause the awk program to croak without indicating the proper error. ----------P-A-T-C-H----T-O----RUN.C----F-O-L-L-O-W-S------------ *** run.c.ORIG Fri Feb 10 06:53:37 1984 --- run.c Fri Sep 13 04:04:26 1985 *************** *** 321,326 int flag = 0; awkfloat xf; os = s; p = buf = (char *)malloc(RECSIZE); while (*s) { --- 321,339 ----- int flag = 0; awkfloat xf; + /* [P. Vijay - Sep 1985] Fix the problem of awk dumping core when + '*' is present either in the precision or field width spec. in + the format string. + + Reason: Presently this routine does not check for the presence + of '*' in the said places, and as a result, sprintf is called + with insufficient params. [e.g., "sprintf("%*s\n",x.opt->sval)"]. + */ + + int VarFieldWidth; /* Set 'true' if '*' used to spec. fld. wid. */ + int VarPrecision; /* Set 'true' if '*' used to spec. precision */ + int FieldWidth , Precision; /* Hold value of respective specifier */ + os = s; p = buf = (char *)malloc(RECSIZE); while (*s) { *************** *** 334,340 continue; } for (t=fmt; (*t++ = *s) != '\0'; s++) ! if (*s >= 'a' && *s <= 'z' && *s != 'l') break; *t = '\0'; if (t >= fmt + sizeof(fmt)) --- 347,353 ----- continue; } for (t=fmt; (*t++ = *s) != '\0'; s++) ! if (*s >= 'a' && *s <= 'z' && *s != 'l') break; *t = '\0'; if (t >= fmt + sizeof(fmt)) *************** *** 339,344 *t = '\0'; if (t >= fmt + sizeof(fmt)) error(FATAL, "format item %.20s... too long", os); switch (*s) { case 'f': case 'e': case 'g': flag = 1; --- 352,367 ----- *t = '\0'; if (t >= fmt + sizeof(fmt)) error(FATAL, "format item %.20s... too long", os); + + /* [P. Vijay - Sep 1985] Check for presence of '%*', '%*.', + and/or '%.*'. This is a QuikFix!! No checking for specs. + such as '%**.12', etc. Someday, this should be fixed to + do the task in a more robust fashion... + */ + + VarFieldWidth = fmt[1] == '*'; + VarPrecision = (*(s-2) == '.') && (*(s-1) == '*'); + switch (*s) { case 'f': case 'e': case 'g': flag = 1; *************** *** 368,373 p += strlen(p); continue; } if (a == NULL) error(FATAL, "not enough arguments in printf(%s)", os); x = execute(a); --- 391,410 ----- p += strlen(p); continue; } + if (VarFieldWidth){ + if (a == NULL) + error(FATAL, "not enough arguments in printf(%s)", os); + x = execute(a); + FieldWidth = getfval(x.optr); + a = a->nnext; + } + if (VarPrecision){ + if (a == NULL) + error(FATAL, "not enough arguments in printf(%s)", os); + x = execute(a); + Precision = getfval(x.optr); + a = a->nnext; + } if (a == NULL) error(FATAL, "not enough arguments in printf(%s)", os); x = execute(a); *************** *** 374,383 a = a->nnext; if (flag != 4) /* watch out for converting to numbers! */ xf = getfval(x.optr); ! if (flag==1) sprintf(p, fmt, xf); ! else if (flag==2) sprintf(p, fmt, (long)xf); ! else if (flag==3) sprintf(p, fmt, (int)xf); ! else if (flag==4) sprintf(p, fmt, x.optr->sval==NULL ? "" : getsval(x.optr)); tempfree(x); p += strlen(p); s++; --- 411,434 ----- a = a->nnext; if (flag != 4) /* watch out for converting to numbers! */ xf = getfval(x.optr); ! ! /* [P. Vijay - Sep 1985] To avoid a mess of repetitive ! code sequences, define 'Xsprintf' macro to do the ! actual sprintf's. ! */ ! ! #define Xsprintf(str,fmt,var)\ ! if (!VarFieldWidth)\ ! if (!VarPrecision) sprintf(str,fmt,(var));\ ! else sprintf(str,fmt,Precision,(var));\ ! else\ ! if (!VarPrecision) sprintf(str,fmt,FieldWidth,(var));\ ! else sprintf(str,fmt,FieldWidth,Precision,(var)) ! ! if (flag==1) Xsprintf(p, fmt, xf); ! else if (flag==2) Xsprintf(p, fmt, (long)xf); ! else if (flag==3) Xsprintf(p, fmt, (int)xf); ! else if (flag==4) Xsprintf(p, fmt, x.optr->sval==NULL ? "" : getsval(x.optr)); tempfree(x); p += strlen(p); s++;
ado@elsie.UUCP (Arthur David Olson) (09/19/85)
For fellow conservatives, here's a change to "run.c" that's: small in magnitude; gets rid of core dumps caused by '*' characters in awk format strings (and by excessively long format strings); and ensures that awk scripts that work on your system will be portable to other systems. The disadvantage is that '*' characters in formats are not allowed--if one is found, a fatal error message is produced. See "substr" for workarounds. As usual, the trade secret status of the code involved precludes a clearer posting. #ifdef OLDVERSION for (t=fmt; (*t++ = *s) != '\0'; s++) if (*s >= 'a' && *s <= 'z' && *s != 'l') break; *t = '\0'; if (t >= fmt + sizeof(fmt)) error(FATAL, "format item %.20s... too long", os); #else for (t=fmt; (*t++ = *s) != '\0'; s++) if (t >= fmt + sizeof fmt) error(FATAL, "format item %.20s... too long", os); /* ** Ought to use ctype, but that might change the ** behavior on non-ASCII machines. */ else if (*s >= 'a' && *s <= 'z' && *s != 'l') break; else if (*s == '*') error(FATAL, "format item %s has wild '*'", os); *t = '\0'; #endif -- Bugs is a Warner Brothers trademark. Awk is a Polly the Parrot trademark. -- UUCP: ..decvax!seismo!elsie!ado ARPA: elsie!ado@seismo.ARPA DEC, VAX and Elsie are Digital Equipment and Borden trademarks
ado@elsie.UUCP (Arthur David Olson) (09/20/85)
A simple, portable workaround is to replace awk nonstatements such as printf "%*.*s\n", a, b, c with awk statements such as printf "%" a "." b "s\n", c -- Bugs is a Warner Brothers trademark. Awk is a Polly the Parrot trademark. -- UUCP: ..decvax!seismo!elsie!ado ARPA: elsie!ado@seismo.ARPA DEC, VAX and Elsie are Digital Equipment and Borden trademarks