https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119214
--- Comment #9 from Robert Dubner <rdubner at symas dot com> ---
Having said all that, I finally got around to confirming the behavior, which I
did.
After noting that the problem doesn't manifest with "-O0 -ftracer", and noting
further that there is no way obvious to me to check the optimization level,
I did notice that the -ftracer option turns on the flag_tracer variable.
So, I wrapped my ASM_EXPR generator in
if( !flag_tracer )
{
...
That eliminated the problem. It also means somebody compiling with -ftracer is
going to be disappointed if they try to use certain features of the GDB-COBOL
debugger. I can't bring myself to feel a lot of concern.
Until such time as I really solve the problem by moving things to .debug_info,
does seem like an acceptable workaround of my workaround?
