Re: Use of set_board_info ldscript

[Ben Elliston <bje at wasabisystems dot com>]

Joern Rennecke <joern.rennecke@superh.com> writes:

> Since C++ test can get pretty big, I've added a line like set
> board_info(sh-hms-sim/-m4-single-only/-ml,ldscript)
> "-Wl,--defsym,_stack=0x7f000"

I feel that the entity best able to predict the requirements for a
test is the tool's own testsuite.  Perhaps there should be a way for
the testsuite to indicate to the test framework how much stack space
is required for a given test, just like it can adjust timeouts?

> - should we have a separate board_info item that gets passed only to the
>   compiler for linking, but not to the linker directly when that is what
>   is used for linking?

No, I don't think this is the right approach.  I'd like to see DejaGnu
less coupled to the application domain of "toolchain testing".

> - should the simulator testsuite use gcc for linking if running from a
>   unified with gcc present? (N.B. I use 'make check' in a unified build
>   tree, which is why I have to use the same site.exp file for gcc and sim.
>   when someone builds & tests components of the toolchain separately,
>   they can use different site.exp files for that.)

This is probably the cleanest solution, but I think it's a bit onerous
to expect users wanting to test a simulator for target FOO to have
FOO-gcc installed.  They might not have ported the compiler yet!

> - should dejagnu process -Wl, when it passes the ldscript options to the
>   linker?

That's probably the most pragmatic approach.  Care to submit a patch?

Ben