Re: [avr-libc-dev] [RFC] sections stub patch

[Theodore A. Roth]

On Thu, 29 Aug 2002, Joerg Wunsch wrote:

:) The patch makes sense to me, also after reading the description of the
:) respective compiler options.  Basically, it disables those parts of C++
:) by default that are currently unavailable for the AVR target (or that
:) don't seem to make sense at all due to space constraints).
:)
:) I'm attaching a simple C++ test file.  The code doesn't do very much and
:) of course is far away from being a real project, but it contains about
:) everything that is needed to verify constructors and destructors of an
:) object (including a global one).  It's impressive to me that the compiler
:) can even eliminate the explicit call to the constructor routine in the
:) -O2 case (just compile it with -S and see the function _ZN6simpleC1Ei
:) go away).  I tried compiling it on the IA32 platform (by declaring an
:) imaginary external function itoa(int, char *, int) on top), the
:) optimization there looks a lot worse considering that the AVR is an
:) 8-bit machine where a lot of data movement is needed.  But OK, IA32
:) suffers from lack of available registers.
:)
:) Ted, maybe you could run it through the simulator.  I still didn't
:) manage it to also produce FreeBSD ports for avrsim and avr-gdb.

Ok... Looks ok to me. gdb gets all confused on with -O2, but that's to be
expected.

I'd wait for gdb 5.3 to be released before you try to get the ports going.
They're cutting the branch on Sept 4, couple more weeks for the release.
gdb-5.2.1 has a minor bug in it for avr which I've already committed to
the mainline.

Looks like we need to get Marek (you listening??) to commit those two
patches to gcc now...  ;-)

Ted Roth