Re: PATCH: Set xfail on empic for Linux/mips

[cgd at broadcom dot com]

At Mon, 30 Sep 2002 16:43:26 -0700, H. J. Lu wrote:
> On Mon, Sep 30, 2002 at 04:30:23PM -0700, cgd@broadcom.com wrote:
> > At Mon, 30 Sep 2002 14:17:16 -0700, H. J. Lu wrote:
> > > Since Linux doesn't use embedded pic, I set them xfail on Linux/mips.
> > > I will set they pass when they are fixed.
> > 
> > uh, if you're going to do that, why not just avoid testing them on
> > Linux/mips to begin with.
> 
> I can certainly do that.
> 
> > 
> > By setting them XFAIL, what'll happen is that they're broken one way
> > now, and future changes will cause them to be broken in additional
> > ways in the future, and somebody who's looking at it N months from now
> > won't be able to tell one from the other.
> 
> I have no problem if embedded PIC remains broken.
> 
> > 
> > What you're proposing here amounts to saying "They don't work for me
> > right now, so the test is wrong and will be marked expected to fail."
> 
> Linux/mips does't use embedded PIC.
> 
> > Except, that's not the reality.  These tests are recently broken.
> > They used to work.  They should work.  If the new assembler linker is
> > correct, the tests should be adjusted.  If not, there is a real bug.
> > They _should not_ however, be marked XFAIL and be left alone.
> > 
> 
> You are more than welcome to fix them. I don't want to see those
> misleading testcases which Linux/mips doesn't use.

That's certainly fine with me.

Do be sure that you and others test mips-elf or other generic MIPS ELF
targets when you contribute changes in the future.  8-)



> [ ... mips-elf tests]
> It is because those tests are ELF/mips specific. Linux/mips uses different MIPS
> targets and linker scripts. You can't depend on the builtin linker scripts for
> those tests if you want them to pass on both ELF/mips and Linux/mips. As I said,
> I don't care much about embedded PIC.

the linker branch-misc-N tests in particular have _NOTHING_ to do with
embedded-pic.  they test whether or not branches work properly.

IIRC, what got you interested in this over the last couple of days and
proposing patches was that... branches to globals were broken.  That
exact condition is what's tested by branch-misc-2, both the assembler
test _and_ the linker test.  That exact problem is what caused me to
put in those tests.  That exact problem happened to be _shown_ first
in the embedded-pic tests.

Originally, before i added the branch-misc-2 test, the _only_ test of
branches to globals IIRC in the binutils test suite were in the
embedded-pic tests.  However, that doesn't mean that that problem was
an embedded-pic "exclusive" problem.

it's not a matter about caring about embedded-pic.

It's more a matter of realizing that, maybe when these other tests for
this same architecture, but with different flags, stop working, the
_really could_ be showing me a problem that will impact my platform.

or, less strong but still quite useful "they really could be showing
me that i made an unintended change."



cgd