On Tue, Sep 18, 2007 at 6:59 PM, via RT Nuno Carvalho
<[EMAIL PROTECTED]> wrote:
> While running 'make fulltest', for today's release, we noticed there
> is a test failling for 't/op/string.t' when using jit runcore. This
> test is currently being skipped.
>
> The test that's failling is test number #91 "if_s_ic", the last part 'ok
> 10':
>
> clears
> if S1, BAD10
> branch OK10
>
> the if op for S1, which is null, is giving an assertion error. I tried
> to poke around the jit core source code for a bit, but wasn't able to
> fix it. My parrot's jit knowledge is not that good, I admit. Can
> anyone more confortable on this area shed some light on the subject.
The problem is that the opcode checks for nullness before calling string_bool:
op if (invar STR, labelconst INT) {
if ($1 && string_bool(interp, $1))
goto OFFSET($2);
}
And the jitted code apparently calls string_bool unconditionally.
The easier solution is to redefine string_bool to allow a NULL
argument and return false in that case. Many places in the code does
that check before the call, so this change will simplify it.
I expect to hear some comments before preparing a patch, given that
this will change the signature of a widely used function.
--
Salu2
