Re: Contributing LLVM.org patches to gud.el

[Stephen J. Turnbull]

 > On 02/11/2015 02:37 AM, David Kastrup wrote:

 > > The issue mostly are values which are not present explicitly
 > > anywhere.  If the compiler can deduce the value/effect of a
 > > variable, it does not need to store it anywhere.

I regularly see this with function arguments, IIRC including across
linkage boundaries.  I don't see how the compiler could deduce such
values or their effects, except for cases like "(char) x < 1000".

Daniel Colascione writes:

 > DWARF's virtual machine contains instructions for regenerating
 > these values from extant values and for building them out of thin
 > air. See section 2.5 of the DWARF 4 specification. GCC could emit
 > enough information to rebuild lost values, but does not.

Eli mentioned DWARF 2.  The manual for GCC 4.8.4 says:

    -g
    Produce debugging information in the operating system's native format
    (stabs, COFF, XCOFF, or DWARF 2). GDB can work with this debugging
    information.

Later there is an option for specifying the version of DWARF, which
cautions that use of DWARF 4 "may require gdb 7 and
-fvar-tracking-assignments".  I'll have to try that.

For those of us who aren't gdb developers, it would be nice if gdb
mentioned that use of DWARF 4 and special gcc options helps in
debugging.