As mentioned in a prior message, the debugger (idb) sometimes cannot
access (print the contents of) a local variable, possibly because it
has been optimized away (despite the fact that the ifort option -O0 was
used). Then I also noticed that access to all local variables
disappears if the code was compiles with the ifort option -save. After
removing the option -save, I find to my surprise that the debugger can
access local variables, but not global variables in a common block. If
there is anything that cannot be optimized away, then it is a variable
in a common block, especially if all routines accessing this variable
are compiled separately.
Has anybody else noticed this problem? Is there a fix?
A bit OT (but maybe not), on VMS I find far too often that the debugger
produces the message "%DEBUG-E-INTERR, debugger error in
DBGADDEXPDETERMINE_TYPE unknown arg type or session corruption" when I
try to access a structure or an element of a structure. But then the
VMS debugger can produce the value that had been assigned to a scalar
variable of the same common block, while the Linux version cannot. And
the VMS debugger has the same problem with structures which have been
passed as subroutine argument.
For a while I thought that the Linux-related problem has something to
do with a warning from the compiler about a few common block variables
, "A common-block-object of a nonsequence derived type is an extension
of Standard F90.". But even accessing a scalar variable from a
different common block does not work.
Comments, suggestions? I am looking for a fix.
