Re: gas/i386/inval-equ-2 test

["Jan Beulich" <jbeulich at novell dot com>]

>>> "H.J. Lu" <hjl.tools@gmail.com> 13.11.08 16:48 >>>
>On Thu, Nov 13, 2008 at 12:27 AM, Jan Beulich <jbeulich@novell.com> wrote:
>> H.J.,
>>
>> I'm trying to understand what this is really trying to verify - due to a
>> supposedly unrelated change this test is failing for me. I'm seeing two
>> possible problems with the test itself, though:
>>
>> For one, (%eax+1) can't really be considered a register symbol - I really
>> think that it's an error for the assembler to keep the resulting symbol in
>> reg_section (this should hold at least for all targets where
>> md_register_arithmetic is zero). That's also one of my local changes - to
>> force such things into expr_section.
>>
>> Secondly, using the equated symbols in an instruction implies that you
>> expect the parser to happily accept the (invalid) construct, which I think
>> is not the purpose of the test (and which also fails with my local change,
>> because [validly] only reg_section symbols are accepted by
>> parse_register()).
>>
>> Hence I would think that either the expressions ought to be changed to
>> have plain registers on the right side, or the move instructions ought to
>> be removed.
>>
>> Looking forward to read your opinion,
>
>I added it to avoid assembler crash. I don't mind you change it as long as
>assembler doesn't crash and does reasonable things.

I'm not sure we understand each other: How could changing the test case
make the assembler crash? All I'm asking is how much of the test case is
really relevant (and even valid), because based on my observations I
have to assume that it sets expectations that are tied to the current gas
implementation, not to what correct or validly to be expected behavior
would demand.

Jan