Re: [PATCH, doc RFA] Remove support for "target m32rsdi" and "target mips/pmon/ddb/rockhopper/lsi"

[Pedro Alves <palves at redhat dot com>]

On 05/03/2016 03:57 PM, Maciej W. Rozycki wrote:
> On Mon, 2 May 2016, Pedro Alves wrote:
> 
>>>  NB it looks to me `mips_r3041_reg_names' is now dead.  We just *might* 
>>> consider rewiring it like `mips_tx39_reg_names', but that would require 
>>> defining another BFD machine type and I doubt anybody cares about the 
>>> R3041 anymore (cf. the relevant comment you've just removed).  So if you 
>>> care to remove it too, then I'll appreciate it and a change to do so is 
>>> preapproved.
>>>
>>>  Given that the variable is static I wonder why it hasn't triggered a 
>>> compilation error in the build actually.
>>
>> That's because gdb doesn't use -Wunused presently.
> 
>  Hmm, I thought it was implied by -Wall.

Yeah, GDB uses -Wno-usused explicitly.

> Perhaps we should add it then?  

Yes, agreed.  That's what Trevor was aiming for IIUC.

> Releases are built without -Werror so the end users will be safe either 
> way, and it'll make us easier to avoid code pollution.

*nod*

>> I never managed to come back to this, and looks like we won't need to.
>> Trevor sent a patch that removes mips_r3041_reg_names among a ton
>> of other unused variables, here:
>>
>>  https://sourceware.org/ml/gdb-patches/2016-04/msg00664.html
> 
>  Great!  As it happens, I made a patch to remove `mips_r3041_reg_names' on 
> Friday, but didn't get to actually pushing it -- and we had a bank holiday 
> yesterday.
> 
>  I'll give Trevor's change precedence then as a more general clean-up, 
> although I'd like to review the MIPS part, as not all variables removed 
> from mips-tdep.c are actually "trivially unused" (those would be lone 
> declarations, possibly with initialisers).  

Thanks!

> Especially the heuristic 
> unwinder bits look highly suspicious to me, where the variable is updated 
> as the analysis proceeds.  It could be that these variables can indeed go, 
> but perhaps something is missing that should be there.

Fair enough.

> 
>  I'll try to get this done by the end of tomorrow.

Thanks.

FWIW, I don't think that this all needs to go in as one big single
patch.  We can push in things incrementally, say, push in the obviously
correct bits, and then push in the bits that remove function calls,
which may have desirable side effects as separate patches.
Trevor, if you'd like to proceed like that, feel free to push in
the patch without the check_typedef, mips, and tracepoint trace status
bits as first step.

Thanks,
Pedro Alves