Re: [PATCH] gdb/testsuite: Run test when software watchpoints are used

[Andrew Burgess <andrew.burgess@embecosm.com>]

* Simon Marchi <simark@simark.ca> [2018-07-05 22:53:54 -0400]:

> On 2018-07-04 12:40 PM, Andrew Burgess wrote:
> > The test gdb.base/watchpoint-reuse-slot.exp can be run when software
> > watchpoints are in use, we just need to update one test pattern to
> > look for 'Watchpoint' instead of 'Hardware watchpoint' in one case.
> > 
> > gdb/testsuite/ChangeLog:
> > 
> > 	* gdb.base/watchpoint-reuse-slot.exp: Test can be run using
> > 	software watchpoints, we just need to update a test pattern in one
> > 	place.
> 
> Hi Andrew,
> 
> Reading the test description, it seems like it exists specifically to
> check the hardware watchpoint mechanisms of the targets.  Did you
> find it would also be useful to run it with software watchpoints?
> What was the idea that motivated to do this change in the first
> place?

I think you give me too much credit!

What happened was I had a target without h/w watchpoints, I ran the
GDB testsuite and had a set of passes and fails.  After some
investigation I realised that I'd neglected to mark the target as not
supporting h/w watchpoints in the board file.

Once I'd added the no h/w watchpoint flag in the board file I reran
the tests, and mostly things looked better.  Failures, or unresolved
tests had become unsupported.

However.... in watchpoint-reuse-slot.exp a number of tests that used
to pass had gone away, so I went looking at the test script.

What I saw was that though the test declared a need for h/w
watchpoints, the test would run perfectly fine without them.

You'll notice that with my change if the board file says that h/w
watchpoints are supported then we still look for the full "Hardware
watchpoint" pattern in the output, that is, my change does not mean
that if GDB broke and h/w watchpoints changed to s/w watchpoints (when
they shouldn't) the test would pass.  I think that after my change all
targets that previously ran this test are just as well tested as they
ever were.

But, we have additional s/w watchpoint testing for targets that don't
support h/w watchpoints.  Is this testing anything that's not covered
elsewhere?  Honestly, I don't know.  There probably is a lot of test
duplication, but I can't guarantee that there's nothing unique in
here.

I guess my question is, what's the harm from broadening the test in
this way?  If I've missed something and this change could mean a bug
can now slip into GDB then absolutely, this is not acceptable.  But, I
can't see how (yet)...

Thanks,
Andrew