Re: [lmi] Fwd: How did this unit test fail?

[Vadim Zeitlin]

On Thu, 22 Mar 2018 19:18:15 +0000 Greg Chicares <address@hidden> wrote:

GC> >  Right now I don't understand at all what's going on here, I'm going to 
try
GC> > to minimize the test case and open a gcc bug report if I manage to do it
GC> > because it clearly looks like something is very, very wrong with it.
GC> 
GC> Great, thanks, I'll look forward to hearing what you find.

 Unfortunately I had no success with this so far: removing any, even
clearly completely unrelated, parts of the source file make the bug
disappear and I didn't manage to really minimize it at all so far (I could
get rid of a few lines, but this leaves way too much code to think about
submitting this to gcc bug tracker).

 One potentially interesting thing I discovered is that _only_ the
instantiation of test_floating_conversion() for (long double, float) (i.e.
the one which fails) changes when -Wall is added/removed, all the other
ones remain exactly the same. But this one changes completely, e.g. in the
very beginning of the function it can be seen that it allocates 16
less/more bytes for the local variables when -Wall is/isn't specified and
then things just diverge completely, i.e. the code in the 2 versions is
very different, even if it seems to do roughly the same thing (I am not
very fluent in reading x87 assembly, so I'm not completely sure). But this
doesn't really help as I still have no idea why is this type combination so
special.

 I'll try a different approach later, but I have to stop with this for now
because I'm just too fed up with not making any progress right now.

 Sorry for the lack of results so far,
VZ