Complicated dg test cases

[Janis Johnson]

On Sat, Jan 31, 2004 at 04:26:24PM -0800, Matt Austern wrote:
> OK, now I have to plead for help.  There's one important corner case on 
> Darwin ("important" meaning that you're almost certain to get it wrong 
> if you don't deal with it specially) that isn't covered by any existing 
> test case and that requires some icky stuff: building a static archive 
> and then linking against it.  This test cannot be rewritten to avoid 
> the .a, because what we're testing is very specifically the interaction 
> between static archives and our template implementation.
> 
> Here's one way to write the test case.  What I don't know is how to put 
> this in a form where it can become part of the dg test suite.  Any 
> advice?

Several months ago, or maybe more than a year ago, I submitted a patch
for the binary compatibility tests in gcc.dg/compat and g++dg/compat to
builds pieces of a test into shared objects and static archives as well
as relocatable object files.  It wasn't reviewed and I got sidetracked.
There are lots of problems that could be discovered with tests like
that.  I'll dig it up and let you take a look at it.  I'm very far
behind with everything right now; how urgent is this for you?

The compat tests already break a test into three pieces that are
compiled separately, but now they're just built into separate .o files.
I don't know of them finding any binary compatibility problems between
releases of GCC, which was their original goal, but they have found a
lot of bugs using the same compiler and same options for the two pieces.
Expanding them to use libraries (static and shared) could be very
interesting.  The main roadblock, I think, is that there are so many
ways to build libraries for the operating systems that GCC supports that
it will be difficult to set up.

Janis