Bug 15966 - [4.0 Only] ICE and segmentation fault on internal write

Attachment

Description

Type

Created

Size

Actions

Create a New Attachment (proposed patch, testcase, etc.)

View All

Bug 15966 depends on:

Show dependency tree
Show dependency graph

Bug 15966 blocks:

18834 19276 19292

Description:

Last confirmed: 2005-06-18 20:40

Opened: 2004-06-12 22:03

This snippet of code is a f95-ification of old legacy code that
generates an ICE/segmentation fault on this particular kind
of internal write. Other compilers accept it. 


troogle> gfortran -c -v !$
gfortran -c -v n15.f95
Reading specs from /OPT/32/gcc-2004-06-09/lib/gcc/hppa2.0w-hp-hpux11.11/3.5.
0/specs
Configured with: ../configure --enable-languages=c,f95 
--prefix=/opt32/gcc-2004-06-09 --with-gmp=/opt32/gmp --with-gnu-as 
--with-as=/opt32/binutils/bin/as --enable-threads --disable-shared --disable-nls
Thread model: posix
gcc version 3.5.0 20040611 (experimental)
 /OPT/32/gcc-2004-06-09/libexec/gcc/hppa2.0w-hp-hpux11.11/3.5.0/f951 n15.f95 
-quiet -dumpbase n15.f95 -auxbase n15 -version -o /var/tmp//cc5B5831.s
GNU F95 version 3.5.0 20040611 (experimental) (hppa2.0w-hp-hpux11.11)
        compiled by GNU C version 3.5.0 20040611 (experimental).
GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096

n15.f95: In function `a':
n15.f95:7: internal compiler error: Segmentation fault



      subroutine a()

      implicit none
      integer         :: i
      character(100)  :: r(1)

      write(r,*) I

      end subroutine a

------- Comment #1 From Andrew Pinski 2004-06-13 09:11 -------

Confirmed, here is the backtrace:
#0  0x080a40e0 in gfc_conv_scalarized_array_ref (se=0xbffebfe0, ar=0x85a9884) at /home/gates/
pinskia/src/gnu/gcc/src/gcc/fortran/trans-array.c:1417
#1  0x080a418d in gfc_conv_array_ref (se=0xbffebfe0, ar=0x85a9884) at /home/gates/pinskia/src/
gnu/gcc/src/gcc/fortran/trans-array.c:1465
#2  0x08096bfc in gfc_conv_expr (se=0xbffebfe0, expr=0x85a9820) at /home/gates/pinskia/src/gnu/
gcc/src/gcc/fortran/trans-expr.c:327
#3  0x080a0ae7 in set_string (block=0xbffec058, postblock=0xbffec050, var=0x400ef414, 
var_len=0x400ef488, e=0x85a9820)
    at /home/gates/pinskia/src/gnu/gcc/src/gcc/fortran/trans-io.c:387
#4  0x080a1cf3 in build_dt (function=0x84f4ea0, code=Variable "code" is not available.
) at /home/gates/pinskia/src/gnu/gcc/src/gcc/fortran/trans-io.c:865
#5  0x0808d65d in gfc_trans_code (code=0x85a9b20) at /home/gates/pinskia/src/gnu/gcc/src/gcc/
fortran/trans.c:574
#6  0x08091a3d in gfc_generate_function_code (ns=0x85a8ec0) at /home/gates/pinskia/src/gnu/gcc/
src/gcc/fortran/trans-decl.c:2046
#7  0x0808d853 in gfc_generate_code (ns=0x85a8ec0) at /home/gates/pinskia/src/gnu/gcc/src/gcc/
fortran/trans.c:651
#8  0x080735a7 in gfc_parse_file () at /home/gates/pinskia/src/gnu/gcc/src/gcc/fortran/parse.c:2477
#9  0x0808a06c in gfc_be_parse_file (set_yydebug=0) at /home/gates/pinskia/src/gnu/gcc/src/gcc/
fortran/f95-lang.c:266
#10 0x0838c4a8 in toplev_main (argc=0, argv=0xbffec304) at /home/gates/pinskia/src/gnu/gcc/src/
gcc/toplev.c:1666
#11 0x420174d9 in __libc_start_main () from /lib/i686/libc.so.6
#12 0x0804b071 in _start ()

Note that this compiles and runs OK subroutine a() implicit none integer :: i character(100) :: r !! eliminate dimension => works write(r,*) I end subroutine a

Same thing happens with read: integer i character*4 c(1) read (c,fmt='(A4)') i end Program received signal SIGSEGV, Segmentation fault. 0x080a760b in gfc_conv_scalarized_array_ref (se=0xbffff740, ar=0x87b48bc) at ../../../gcc/gcc/fortran/trans-array.c:1578 1576 info = &se->ss->data.info; 1577 if (ar) 1578 n = se->loop->order[0]; 1579 else 1580 n = 0; Seeing the number of duplicates for this one, I'm sure we want to get it on our Ten Most Wanted list.

Well, this one should be so difficult. I think the tweak should be in io.c: /* Resolve everything in a gfc_dt structure. */ try gfc_resolve_dt (gfc_dt * dt) There is some code called "Sanity checks on data transfer statements", where we might check for the case where the io_unit is a character array, and then substitute the code for its lower element. Since, I'm really not used to the guts of the front-end, I can't do more than that (and I'm not sure this analysis is correct), but this doesn't seem so difficult...

------- Comment #16 From CVS Commits 2005-07-07 07:55 -------

Subject: Bug 15966

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	fengwang@gcc.gnu.org	2005-07-07 07:55:00

Modified files:
	libgfortran    : ChangeLog 
	gcc/fortran    : ChangeLog arith.c expr.c intrinsic.c io.c 
	                 misc.c primary.c simplify.c trans-const.c 
	                 trans-io.c arith.h gfortran.h 
	gcc/testsuite  : ChangeLog 
	libgfortran/io : transfer.c 
Added files:
	gcc/testsuite/gfortran.dg: hollerith.f90 hollerith2.f90 
	                           hollerith3.f90 hollerith4.f90 
	                           hollerith_f95.f90 
	                           hollerith_legacy.f90 
	gcc/testsuite/gfortran.dg/g77: cpp4.F 

Log message:
	For the 60th anniversary of Chinese people��s Anti-Japan war victory.
	
	2005-07-07  Feng Wang  <fengwang@nudt.edu.cn>
	
	PR fortran/16531
	PR fortran/15966
	PR fortran/18781
	* arith.c (gfc_hollerith2int, gfc_hollerith2real,
	gfc_hollerith2complex, gfc_hollerith2character, gfc_hollerith2logical):
	New functions.
	(eval_intrinsic): Don't evaluate if Hollerith constant arguments exist.
	* arith.h (gfc_hollerith2int, gfc_hollerith2real,
	gfc_hollerith2complex, gfc_hollerith2character, gfc_hollerith2logical):
	Add prototypes.
	* expr.c (free_expr0): Free memery allocated for Hollerith constant.
	(gfc_copy_expr): Allocate and copy string if Expr is from Hollerith.
	(gfc_check_assign): Enable conversion from Hollerith to other.
	* gfortran.h (bt): Add BT_HOLLERITH.
	(gfc_expr): Add from_H flag.
	* intrinsic.c (gfc_type_letter): Return 'h' for BT_HOLLERITH.
	(add_conversions): Add conversions from Hollerith constant to other.
	(do_simplify): Don't simplify if  Hollerith constant arguments exist.
	* io.c (resolve_tag): Enable array in FORMAT tag under GFC_STD_GNU.
	* misc.c (gfc_basetype_name): Return "HOLLERITH" for BT_HOLLERITH.
	(gfc_type_name): Print "HOLLERITH" for BT_HOLLERITH.
	* primary.c (match_hollerith_constant): New function.
	(gfc_match_literal_constant): Add match Hollerith before Integer.
	* simplify.c (gfc_convert_constant): Add conversion from Hollerith
	to other.
	* trans-const.c (gfc_conv_constant_to_tree): Use VIEW_CONVERT_EXPR to
	convert Hollerith constant to tree.
	* trans-io.c (gfc_convert_array_to_string): Get array's address and
	length to set string expr.
	(set_string): Deal with array assigned Hollerith constant and character
	array.
	* gfortran.texi: Document Hollerith constants as extention support.
	
	2005-07-07  Feng Wang  <fengwang@nudt.edu.cn>
	
	PR fortran/16531
	PR fortran/15966
	PR fortran/18781
	* gfortran.dg/hollerith.f90: New.
	* gfortran.dg/hollerith2.f90: New.
	* gfortran.dg/hollerith3.f90: New.
	* gfortran.dg/hollerith4.f90: New.
	* gfortran.dg/hollerith_f95.f90: New.
	* gfortran.dg/hollerith_legacy.f90: New.
	* gfortran.dg/g77/cpp4.F: New. Port from g77.
	
	2005-07-07  Feng Wang  <fengwang@nudt.edu.cn>
	
	PR fortran/16531
	* io/transfer.c (formatted_transfer): Enable FMT_A on other types to
	support Hollerith constants.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libgfortran/ChangeLog.diff?cvsroot=gcc&r1=1.255&r2=1.256
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/ChangeLog.diff?cvsroot=gcc&r1=1.485&r2=1.486
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/arith.c.diff?cvsroot=gcc&r1=1.29&r2=1.30
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/expr.c.diff?cvsroot=gcc&r1=1.26&r2=1.27
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/intrinsic.c.diff?cvsroot=gcc&r1=1.51&r2=1.52
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/io.c.diff?cvsroot=gcc&r1=1.27&r2=1.28
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/misc.c.diff?cvsroot=gcc&r1=1.9&r2=1.10
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/primary.c.diff?cvsroot=gcc&r1=1.28&r2=1.29
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/simplify.c.diff?cvsroot=gcc&r1=1.29&r2=1.30
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/trans-const.c.diff?cvsroot=gcc&r1=1.28&r2=1.29
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/trans-io.c.diff?cvsroot=gcc&r1=1.37&r2=1.38
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/arith.h.diff?cvsroot=gcc&r1=1.6&r2=1.7
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fortran/gfortran.h.diff?cvsroot=gcc&r1=1.74&r2=1.75
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.5732&r2=1.5733
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/libgfortran/io/transfer.c.diff?cvsroot=gcc&r1=1.46&r2=1.47
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/hollerith.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/hollerith2.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/hollerith3.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/hollerith4.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/hollerith_f95.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/hollerith_legacy.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.dg/g77/cpp4.F.diff?cvsroot=gcc&r1=NONE&r2=1.1

(In reply to comment #19) > Fixed in 4.0.2 and above. Strictly speaking, pr15966 is fixed. However, writing to array internal files is still broken because no record length (aka character length) is being transmitted. This is a bug that comes up in a number of NIST tests (eg. fm909.for tests 6, 7, 8, 10 & 22, 23, 25, 26.). Jerry DeLisle and I are onto it. With: implicit none integer :: i(10),j character(12) :: r(12) i = (/(j,j=1,10)/) write(r,'(5(2x,i4/)/5(3x,i6/))') i write(6,'(a12)') r end Intel and g95 do this [root@localhost mytests]# ./a.out 1 2 3 4 5 6 7 8 9 10 whilst gfortran does.... well, something else [root@localhost mytests]# ./a.out 1 2 3 4 5 6 7 8 9 1 0

GCC Bugzilla Bug 15966

[4.0 Only] ICE and segmentation fault on internal write

Last modified: 2005-07-12 05:44:25

Known to work:
Known to fail: