[distcc] newbie: problem with distcc and intel C++ 7.1: 0byteobject

[Martin Pool]

On 11 Jun 2003, Markus Werle <numerical.simulation at web.de> wrote:
> Martin Pool wrote:

> So further investigation leads to this:
> Something is messed up with the compiler args,
> seems like distcc (or autotools?) add the -MD argument.

Ah!  -MD is making the preprocessor write dependencies to the .i file,
which is why the compiler can't process it.

distcc certainly does not add it this option.  It must be autotools --
probably automake.

> distcc[2069] (dcc_scan_args) scanning arguments: icpc -DHAVE_CONFIG_H -I. -I/work/tresca/markus/DAIXTROSE/daixtrose -I. -I/work/tresca/markus/DAIXTROSE/daixtrose/src -I/opt/boost-1.30.0 -I/opt/loki-2003May28 -DBOOST_NO_CONFIG -xc++ -g -wd76 -ansi -c -o test_simple_get_value_1-main.o /work/tresca/markus/DAIXTROSE/daixtrose/src/demos/SimpleGetValue.1/main.C -MD

Show that the -MD option was there when distcc was invoked.  distcc
passes them through to the compiler.

Using -MD with distcc is very problematic, because the behaviour of
this flag has changed incompatibly over the history of gcc releases.
There is some discussion of this on the problems page and in the list
archives.

Now that people seem to be stabilizing on gcc 3+ it might be OK for
distcc to try to accomodate the current behaviour.  gcc 2.9x would be
broken, but perhaps no more so than now.  This needs some thought.

Does icpc interpret -MD in the same way as gcc anyhow?  If automake is
adding random gcc options to the command line then you may run into
other kinds of trouble.

Is there a way to avoid using -MD?

--
Martin