PR30372

[Brooks Moses]

Steve Kargl wrote:
> On Mon, Feb 12, 2007 at 03:36:46AM +0100, Daniel Franke wrote:
>> It is to assume that, eventually, there will be loads of (mostly generated) 
>> test files. Shall we consider to add them to the fortran test harness?
> 
> This touchs on an observation of Brooks' that the tests in the
> testsuite (which test intrinsics), typically only test the 
> constant folding in the frontend.  We should devise tests for
> both constant folding and the library routines.  I think the 
> test can go into gfortran.dg for now.  At some point, we may 
> move things into libgfortran.dg.

I think it might also be useful to make an "intrinsics" subdirectory off 
of the gfortran.dg directory, just to keep things a bit more organized, 
especially if most of them will be autogenerated.  That will make it 
easier to move them if we do decide to.

While you're generating things, it would probably be good to generate 
testcases that check for errors with extraneous arguments and missing 
required arguments, along with wrong-type arguments.  (I'm especially 
sensitive to that since the big CALL_EXPR patch we just submitted had a 
bug in it that caused about two dozen C builtins to accept extraneous 
arguments or ICE with missing ones, and the testsuite didn't catch it.)

Thanks for working on this, by the way!

- Brooks