Dave Korn wrote:
On 11 February 2008 21:29, Joel Sherrill wrote:
We use versioned target names. Do I have to list them all in
the following section?
=====================================
# This is a list of toolchains that are supported on this board.
set_board_info target_install {powerpc-rtems4.9}
=====================================
I don't know for a fact, I've never tried it, but I'd guess...
runtest.exp: if { [lsearch -exact $target_install $target_alias] >= 0 }
.. you need to explicitly enumerate them. (There might be a good argument for
making that --glob or even --regex).
What about arguments to the simulator? As you can see in
the example, we use this to invoke powerpc-rtems-run:
RUN -f DEVICE_TREE_FILE EXECUTABLE
where DEVICE_TREE_FILE has the device configuration. How does
that get passed through? Where should the DEVICE_TREE_FILE
go?
I handle that - and this works for me because, like your case, the
EXECUTABLE is last on the commandline - by doing the equivalent of
set_board_info sim "RUN -f DEVICE_TREE_FILE"
and the EXECUTABLE gets appended by proc sim_spawn. (Looking at that, it
seems there is some way of specifying the sim and the flags separately by
set_board_info sim "RUN"
set_board_info sim,flags "-f DEVICE_TREE_FILE"
but I have never tried it myself.)
+ an extra object file to link with.
set_board_info ldflags
Do I custom compile testrun.c myself? Or is that automatic?
For sure we need the main() wrapper for no return code. I
can have the application start at main() so that isn't an issue.
There's a thing called
set_board_info needs_status_wrapper 1
which I forget how it works, but that's one way to make it work. In my case,
I've added a flag to my simulator so that the sim itself returns the exit
status of the executable (and detects if the simulated app hits abort() or
exit() at the same time, I have no runtime library to speak of).
I assume I have to manually compile an rtems specific support
file before the tests are run right?
If the needs_status_wrapper stuff doesn't do what you want, I think - again,
untested - you could probably invoke the compiler from your baseboard script.
Don't really know about this one though.
+ target CFLAGS and LDFLAGS
set_board_info cflags
set_board_info ldflags
As you can hopefully see now, we need to generically do this:
-B ${prefix}/${target}/BSP/ -specs bsp_specs -qrtems CPU_CFLAGS
where each board has its own CPU_CFLAGS. For psim, it would be
something like this when fully expanded:
-B/home/joel/rtems-4.9-work/bsp-install//powerpc-rtems4.9/psim/lib/ \
-specs bsp_specs -qrtems -mcpu=603e -Dppc603e
I hate to hard-code a path or target since it is versioned. Are there
DejaGNU variables I can use?
I pass external environment vars into the script by using
make check RUNTESTFLAGS=" ... MY_SCRIPT_VAR=${MY_SCRIPT_VAR} ... "
then I can use the value of MY_SCRIPT_VAR in the various set_board_info flags;
that should do for getting things like prefix and target into your flags.
So, how far are you at? Have you got a baseboard script written yet?
Have you read the "Extending DejaGnu" chapter of the manual, specifically
the "Adding a new board" section?
I'm hacking on that. But am not comfortable with this and
still don't know how to invoke runtest properly. :)
I always use
make -k check RUNTESTFLAGS="... whatever ..."
and don't bother trying to invoke it directly. (Can replace "check" by
"check-gcc" if you want to skip the mangler tests etc.)
I have this now to run it:
make check-gcc RUNTESTFLAGS="\
RTEMS_MAKEFILE_PATH=/home/joel/work-gnat/423/bsp-install/powerpc-rtems4.9/psim/
\
--target=rtems-powerpc-sim --all \
"
I got executables but they are not linking in with the RTEMS BSP
from what I see of the symbol table and no surprise do not run. :)
From the log I see this which would seem to indicate that I am
not getting any compiler flags passed in.
======================
Executing on host: /home/joel/work-gnat/svn/b-gcc2-powerpc/gcc/xgcc
-B/home/joel/work-gnat/svn/b-gcc2-powerpc/gcc/ -O2 -w
-fno-show-column -c -o 20001222-1.o
/home/joel/work-gnat/svn/gcc/gcc/testsuite/gcc.c-torture/compile/20001222-1.c
(timeout = 300)
======================
The generated site.exp doesn't look like my board info is having any
impact:
=========================
set CFLAGS ""
set CXXFLAGS ""
set HOSTCC "gcc"
set HOSTCFLAGS "-g"
set TESTING_IN_BUILD_TREE 1
set HAVE_LIBSTDCXX_V3 1
set GCC_EXEC_PREFIX "/home/joel/work-gnat/svn/install/lib/gcc/"
set newlib_cflags
"-I/home/joel/work-gnat/svn/b-gcc2-powerpc/gcc/../powerpc-rtems4.9/newlib/targ-include
-I$srcdir/../newlib/libc/include"
set newlib_ldflags
"-B/home/joel/work-gnat/svn/b-gcc2-powerpc/gcc/../powerpc-rtems4.9/newlib/"
append CFLAGS " $newlib_cflags"
append CXXFLAGS " $newlib_cflags"
=========================
Haven't tried your testcase yet; I'll look at it if we can't get this going
in just a few rounds of email exchanges.
I think we are getting there. Just slow due to the imperfect
reception on this end. :)
--joel
_______________________________________________
DejaGnu mailing list
DejaGnu@gnu.org
http://lists.gnu.org/mailman/listinfo/dejagnu
