Re: [RFA] Some testcases for long long bitfields

[Andrew Cagney <cagney@gnu.org>]

Paul Hilfinger wrote:
> I believe that the following revision of the bitfields2 test (for
> bitfields in long long fields) addresses Andrew's comments.  I have
> tested it on i686 GNU/Linux with GCC 3.2.3.  OK?
> 
> Some specifics:
> 
> 
>>- Why?  Or is this really a known bug?
>>
>>>+    if $no_signed then {
>>>+	setup_xfail "*-*-*"
>>>+    }
>>>+    set test "set long long signed bitfield negative"
>>>+    gdb_test_multiple "print flags.s2 = -1" $test {
>>>+	-re "warning: Value does not fit.*$gdb_prompt $" {
>>>+	    fail "$test"
>>>+	    gdb_suppress_tests
>>>+	}
>>>+	-re "= -1.*$gdb_prompt $" {
>>>+	    pass "$test"
>>>+	}
>>>+    } 
> 
> 
> Well, this is what bitfields does, which is what I used as a model. On 
> reflection, however, it seems to me that XFAIL is inappropriate, since if
> there is a failure here, it is most likely due to the fact the compiler
> being used does not support signed bitfields---that is, a compiler for 
> which !defined(__STDC__) && !defined(__cplusplus) and which interprets
> bitfields as unsigned.  So it seems that the tests ought to be considered
> unsupported instead, and I have made that change.  But that merely changes
> your question to "ARE there any compilers we need worry about with this
> property". I have no idea, but other testcases seem to be written as if
> there are.

True, ok.

>>- delete this:
>>+if [istarget "mips-idt-*"] then {
>>+    # Restart because IDT/SIM runs out of file descriptors.
>>+    gdb_exit
>>+    gdb_start
>>+    gdb_reinitialize_dir $srcdir/$subdir
>>+    gdb_load ${binfile}
>>+}
> 
> 
> Done.  I observe, by the way, that this same code appears in several
> existing tests: bitfields, funcargs, opaque, scope.
> 
> 
>>perhaps think about what I did for the sig*.exp tests - have main as a 
>>loop so that it looped around after each test sequence was finished - 
>>will on remote systems improve the performance somewhat.  But what ever.
> 
> 
> OK. I have made a variant of this change that does not use GDB to modify the
> control flow of the program. 

Thanks! Ok for commit.

Andrew