Your message dated Mon, 17 Dec 2012 14:57:46 -0800
with message-id <20121217225746.ga23...@math.sfu.ca>
and subject line Re: Bug#696058: gfortran: [gfortran] incorrect logic in -O2
old-style
has caused the Debian Bug report #696058,
regarding gfortran: [gfortran] incorrect logic in -O2 old-style loop
to be marked as done.
This means that you claim that the problem has been dealt with.
If this is not the case it is now your responsibility to reopen the
Bug report if necessary, and/or fix the problem forthwith.
(NB: If you are a system administrator and have no idea what this
message is talking about, this may indicate a serious mail system
misconfiguration somewhere. Please contact ow...@bugs.debian.org
immediately.)
--
696058: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=696058
Debian Bug Tracking System
Contact ow...@bugs.debian.org with problems
--- Begin Message ---
Package: gfortran
Version: 4:4.7.2-1exp1
Severity: normal
Comparision logic fails between "wrapped around" integers inside an inverted
while loop hack. This occurs only on optimisation level 2 and if the
loop does not have an upper limit.
I encountered this when trying to emperically check the integer
limits. Attached is a small example code and the unexpected output when
compiled with "gfortran -O2 ltbug.f". The last line should read
"( 127 < -128 )= F" and execution should terminate there, but instead
continues indefinitely.
-- System Information:
Debian Release: 7.0
APT prefers unstable
APT policy: (600, 'unstable'), (500, 'testing'), (400, 'stable'), (300,
'experimental')
Architecture: amd64 (x86_64)
Foreign Architectures: i386
Kernel: Linux 3.6.10.01 (SMP w/4 CPU cores)
Locale: LANG=en_CA.UTF-8, LC_CTYPE=en_CA.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Versions of packages gfortran depends on:
ii cpp 4:4.7.2-1exp1
ii gcc 4:4.7.2-1exp1
ii gfortran-4.7 4.7.2-4
gfortran recommends no packages.
Versions of packages gfortran suggests:
ii gfortran-doc 5:4
pn gfortran-multilib <none>
-- no debconf information
program main
implicit none
integer*1 amax,amin,atmp
integer i
amin=-128
atmp=amin-1
print*,'outside loops, logic is good'
print*,'(',atmp,'<',amin,')=',atmp.lt.amin
print*,'in finite loop, logic is good'
amin=0
do i=1,256
atmp=amin-1
print*,'(',atmp,'<',amin,')=',atmp.lt.amin
if ( atmp .lt. amin) then
amin=atmp
else
goto 100
end if
end do
100 continue
print*,'without loop index, logic is bad'
amin=0
do
atmp=amin-1
print*,'(',atmp,'<',amin,')=',atmp.lt.amin
if ( atmp .lt. amin) then
amin=atmp
else
goto 200
end if
end do
200 continue
stop
end program main
outside loops, logic is good
( 127 < -128 )= F
in finite loop, logic is good
( -1 < 0 )= T
( -2 < -1 )= T
( -3 < -2 )= T
( -4 < -3 )= T
( -5 < -4 )= T
( -6 < -5 )= T
( -7 < -6 )= T
( -8 < -7 )= T
( -9 < -8 )= T
( -10 < -9 )= T
( -11 < -10 )= T
( -12 < -11 )= T
( -13 < -12 )= T
( -14 < -13 )= T
( -15 < -14 )= T
( -16 < -15 )= T
( -17 < -16 )= T
( -18 < -17 )= T
( -19 < -18 )= T
( -20 < -19 )= T
( -21 < -20 )= T
( -22 < -21 )= T
( -23 < -22 )= T
( -24 < -23 )= T
( -25 < -24 )= T
( -26 < -25 )= T
( -27 < -26 )= T
( -28 < -27 )= T
( -29 < -28 )= T
( -30 < -29 )= T
( -31 < -30 )= T
( -32 < -31 )= T
( -33 < -32 )= T
( -34 < -33 )= T
( -35 < -34 )= T
( -36 < -35 )= T
( -37 < -36 )= T
( -38 < -37 )= T
( -39 < -38 )= T
( -40 < -39 )= T
( -41 < -40 )= T
( -42 < -41 )= T
( -43 < -42 )= T
( -44 < -43 )= T
( -45 < -44 )= T
( -46 < -45 )= T
( -47 < -46 )= T
( -48 < -47 )= T
( -49 < -48 )= T
( -50 < -49 )= T
( -51 < -50 )= T
( -52 < -51 )= T
( -53 < -52 )= T
( -54 < -53 )= T
( -55 < -54 )= T
( -56 < -55 )= T
( -57 < -56 )= T
( -58 < -57 )= T
( -59 < -58 )= T
( -60 < -59 )= T
( -61 < -60 )= T
( -62 < -61 )= T
( -63 < -62 )= T
( -64 < -63 )= T
( -65 < -64 )= T
( -66 < -65 )= T
( -67 < -66 )= T
( -68 < -67 )= T
( -69 < -68 )= T
( -70 < -69 )= T
( -71 < -70 )= T
( -72 < -71 )= T
( -73 < -72 )= T
( -74 < -73 )= T
( -75 < -74 )= T
( -76 < -75 )= T
( -77 < -76 )= T
( -78 < -77 )= T
( -79 < -78 )= T
( -80 < -79 )= T
( -81 < -80 )= T
( -82 < -81 )= T
( -83 < -82 )= T
( -84 < -83 )= T
( -85 < -84 )= T
( -86 < -85 )= T
( -87 < -86 )= T
( -88 < -87 )= T
( -89 < -88 )= T
( -90 < -89 )= T
( -91 < -90 )= T
( -92 < -91 )= T
( -93 < -92 )= T
( -94 < -93 )= T
( -95 < -94 )= T
( -96 < -95 )= T
( -97 < -96 )= T
( -98 < -97 )= T
( -99 < -98 )= T
( -100 < -99 )= T
( -101 < -100 )= T
( -102 < -101 )= T
( -103 < -102 )= T
( -104 < -103 )= T
( -105 < -104 )= T
( -106 < -105 )= T
( -107 < -106 )= T
( -108 < -107 )= T
( -109 < -108 )= T
( -110 < -109 )= T
( -111 < -110 )= T
( -112 < -111 )= T
( -113 < -112 )= T
( -114 < -113 )= T
( -115 < -114 )= T
( -116 < -115 )= T
( -117 < -116 )= T
( -118 < -117 )= T
( -119 < -118 )= T
( -120 < -119 )= T
( -121 < -120 )= T
( -122 < -121 )= T
( -123 < -122 )= T
( -124 < -123 )= T
( -125 < -124 )= T
( -126 < -125 )= T
( -127 < -126 )= T
( -128 < -127 )= T
( 127 < -128 )= F
without loop index, logic is bad
( -1 < 0 )= T
( -2 < -1 )= T
( -3 < -2 )= T
( -4 < -3 )= T
( -5 < -4 )= T
( -6 < -5 )= T
( -7 < -6 )= T
( -8 < -7 )= T
( -9 < -8 )= T
( -10 < -9 )= T
( -11 < -10 )= T
( -12 < -11 )= T
( -13 < -12 )= T
( -14 < -13 )= T
( -15 < -14 )= T
( -16 < -15 )= T
( -17 < -16 )= T
( -18 < -17 )= T
( -19 < -18 )= T
( -20 < -19 )= T
( -21 < -20 )= T
( -22 < -21 )= T
( -23 < -22 )= T
( -24 < -23 )= T
( -25 < -24 )= T
( -26 < -25 )= T
( -27 < -26 )= T
( -28 < -27 )= T
( -29 < -28 )= T
( -30 < -29 )= T
( -31 < -30 )= T
( -32 < -31 )= T
( -33 < -32 )= T
( -34 < -33 )= T
( -35 < -34 )= T
( -36 < -35 )= T
( -37 < -36 )= T
( -38 < -37 )= T
( -39 < -38 )= T
( -40 < -39 )= T
( -41 < -40 )= T
( -42 < -41 )= T
( -43 < -42 )= T
( -44 < -43 )= T
( -45 < -44 )= T
( -46 < -45 )= T
( -47 < -46 )= T
( -48 < -47 )= T
( -49 < -48 )= T
( -50 < -49 )= T
( -51 < -50 )= T
( -52 < -51 )= T
( -53 < -52 )= T
( -54 < -53 )= T
( -55 < -54 )= T
( -56 < -55 )= T
( -57 < -56 )= T
( -58 < -57 )= T
( -59 < -58 )= T
( -60 < -59 )= T
( -61 < -60 )= T
( -62 < -61 )= T
( -63 < -62 )= T
( -64 < -63 )= T
( -65 < -64 )= T
( -66 < -65 )= T
( -67 < -66 )= T
( -68 < -67 )= T
( -69 < -68 )= T
( -70 < -69 )= T
( -71 < -70 )= T
( -72 < -71 )= T
( -73 < -72 )= T
( -74 < -73 )= T
( -75 < -74 )= T
( -76 < -75 )= T
( -77 < -76 )= T
( -78 < -77 )= T
( -79 < -78 )= T
( -80 < -79 )= T
( -81 < -80 )= T
( -82 < -81 )= T
( -83 < -82 )= T
( -84 < -83 )= T
( -85 < -84 )= T
( -86 < -85 )= T
( -87 < -86 )= T
( -88 < -87 )= T
( -89 < -88 )= T
( -90 < -89 )= T
( -91 < -90 )= T
( -92 < -91 )= T
( -93 < -92 )= T
( -94 < -93 )= T
( -95 < -94 )= T
( -96 < -95 )= T
( -97 < -96 )= T
( -98 < -97 )= T
( -99 < -98 )= T
( -100 < -99 )= T
( -101 < -100 )= T
( -102 < -101 )= T
( -103 < -102 )= T
( -104 < -103 )= T
( -105 < -104 )= T
( -106 < -105 )= T
( -107 < -106 )= T
( -108 < -107 )= T
( -109 < -108 )= T
( -110 < -109 )= T
( -111 < -110 )= T
( -112 < -111 )= T
( -113 < -112 )= T
( -114 < -113 )= T
( -115 < -114 )= T
( -116 < -115 )= T
( -117 < -116 )= T
( -118 < -117 )= T
( -119 < -118 )= T
( -120 < -119 )= T
( -121 < -120 )= T
( -122 < -121 )= T
( -123 < -122 )= T
( -124 < -123 )= T
( -125 < -124 )= T
( -126 < -125 )= T
( -127 < -126 )= T
( -128 < -127 )= T
( 127 < -128 )= T
...
--- End Message ---
--- Begin Message ---
I was worried it might be something like that. Thankyou for the detailed
response.
Kevin
On Mon 12.12.17 03:05, Tobias Burnus wrote:
Kevin Mitchell wrote:
Comparision logic fails between "wrapped around" integers inside an inverted
while loop hack. This occurs only on optimisation level 2 and if the
loop does not have an upper limit.
I encountered this when trying to emperically check the integer
limits. Attached is a small example code and the unexpected output when
compiled with "gfortran -O2 ltbug.f". The last line should read
"( 127 < -128 )= F" and execution should terminate there, but instead
continues indefinitely.
I believe that that's a bug in the program. Fortran (like C) does not
define the semantic for integer wrapping around [at least not for
signed integers; though Fortran doesn't have unsigned integers].
Work around: -fno-strict-overflow. (-fstrict-overflow is enabled by
default with -O2 or higher).
I have some trouble pin-pointing it in the Fortran standard;
nonetheless some quotes from Fortran 2008:
ftp://ftp.nag.co.uk/sc22wg5/N1801-N1850/N1830.pdf
* The standard doesn't define the semantics, hence, the catch it all
applies: "A program (2.2.2) is a standard-conforming program if it
uses only those forms and relationships described herein and if the
program has an interpretation according to this part of ISO/IEC
1539." (Section "1.5 Conformance").
* "The execution of any numeric operation whose result is not defined
by the arithmetic used by the processor is prohibited." (7.1.5.2.4
Evaluation of numeric intrinsic operations)
* And the numerical model is stated in "13.4 Numeric models".
The following quote is more explicit albeit not normative. It comes
from the draft Fortran appendix to ISO/IEC Technical Report (TR)
24772 "Guidance to avoiding vulnerabilities in programming languages
through language selection and use". [That's one of the few
free-of-charge ISO standards.] The Fortran appendix has been written
by members of the Fortran standardization committee and has been
voted on. See ftp://ftp.nag.co.uk/sc22wg5/N1901-N1950/N1947.pdf
Namely: "if an integer operation results in a value outside the
supported range, the program is not conforming. This might not be
detected. Likewise, assigning a value to an integer variable whose
range does not include the value, renders the program not
conforming."
Tobias
--- End Message ---
