This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA]: threaded watchpoint test


Mark Kettenis writes:
mark> Why do we allow tests like this to be checked in if we know they
mark> produce massive fails on many systems?

To take the philosophical question first: there's no registry of
which systems are actually tested with gdb.  The list of posts to
gdb-testers@ is the only thing I have.  If a system is not
represented there then I pay less attention to it.

Hmmm, this is in gdb.threads, I should not even have approved this
patch in the first place!  I apologize to Michael Snyder.

mark> Hmm, I'm getting a bit frustrated that folks keep adding things to the
mark> testsuite that assume features that are not available on most things
mark> not i386 GNU/Linux.

I hear you.

Daniel Jacobowitz writes:
drow> Maybe not.  It could be a bug to get a software watchpoint when we
drow> expect a hardware one.  The question is whether we want to keep a
drow> separate list of "do we expect hardware watchpoints" in the testsuite.

I think so, but I'm not sure.

Two possible approaches are:

(1) The test suite does not keep its own info.  The test suite asks
    gdb whether gdb supports feature X, and trusts gdb.

(2) The test suite keeps separate info.  That way, the test suite
    is actually testing gdb, not just asking and trusting it.
    But the info can get out of sync.

    (2A) A whitelist of systems to test.

    (2B) A blacklist of systems not to test.

I'm inclined to some form of (2).  I don't like (1) because the test
suite is supposed to *test* gdb, not just trust gdb's answers.
Also, I don't know any existing gdb command to ask this question.
On the other hand, hardware watchpoints are not meant to work on
all systems, so it might be reasonable to ask gdb about them.

gdb.base/bigcore.exp is a big test that uses (2).  First it blacklists
all the !isnative systems.  Then it blacklists four specific systems:
bsd, hpux, solaris, cygwin.

For hardware watchpoint tests, the blacklist would be pretty long.
That's okay with me, it's just painful to set up the first time.

What do y'all think?

Michael C


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]