This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: Ping: [PATCH] gas: fix bogus error on .org involving expression
- From: "Jan Beulich" <JBeulich at suse dot com>
- To: "Hans-Peter Nilsson" <hp at bitrange dot com>
- Cc: "Alan Modra" <amodra at gmail dot com>,<augustine dot sterling at gmail dot com>, <binutils at sourceware dot org>
- Date: Tue, 22 Mar 2016 10:49:09 -0600
- Subject: Re: Ping: [PATCH] gas: fix bogus error on .org involving expression
- Authentication-results: sourceware.org; auth=none
- References: <56D6DCAB02000078000D845A at prv-mh dot provo dot novell dot com> <56E99A7F02000078000DD4E5 at prv-mh dot provo dot novell dot com> <20160316234647 dot GA22605 at bubble dot grove dot modra dot org> <56F11C3402000078000DF167 at prv-mh dot provo dot novell dot com> <alpine dot BSF dot 2 dot 02 dot 1603220821260 dot 90382 at arjuna dot pair dot com>
>>> On 22.03.16 at 13:26, <hp@bitrange.com> wrote:
> On Tue, 22 Mar 2016, Jan Beulich wrote:
>> Beyond those, however, this also revealed a few other failures:
>
>> 4) MMIX'es pr12815-* tests both fail (they no longer produce the
>> expected or any other error). Since I have no idea what exactly
>> those tests test, I also have no idea how to deal with this.
>
> You refer to *linker tests* checking for "a meaningful error
> message rather than SEGV" from the *linker*. I suspect your
> assembler patches affects things it shouldn't, i.e. the it
> affects assembler output, not only fixing error cases.
While that's certainly true, it's entirely unclear to me why that would
be. Looking a little more closely, both generated object files have
empty .text and .data sections (yet only the .data section of the
second one was empty before). Assuming that what I see in the
sources are opcodes, I have absolutely no idea why that would
result in no code getting generated at all, but there also not being
any error. This, to me at least, quite clearly hints at a bug elsewhere
which this change uncovers. I could guess (and perhaps work out
by trial and error) whether one of the five uses of undefined_section
in gas/config/tc-mmix.c needs replacing or extending by considering
expr_section, but that's not the kind of work I'd like to do on code I
have no knowledge about at all.
Of course it's generally questionable whether the significantly
lower amount of expr_section uses in gas/config/tc-*.c compared
to undefined_section doesn't indicate further lurking issues. So
unless I can get some help here, I'm almost willing to give up on
this and leave .org broken (I've already managed to find an
acceptable replacement for the use case with which I ran into the
problem originally).
Jan