Re: [PATCH] libgloss: arm: break newlib dependency

[Richard Earnshaw <Richard.Earnshaw@foss.arm.com>]

On 22/12/2022 22:03, Mike Frysinger wrote:
> On 21 Dec 2022 09:24, Corinna Vinschen wrote:
>> On Dec 20 20:47, Mike Frysinger wrote:
>>> On 19 Dec 2022 10:08, Richard Earnshaw wrote:
>>>> On 14/12/2022 09:13, Mike Frysinger wrote:
>>>>> The libgloss port has been reaching back into newlib internals for a
>>>>> single header whose contents have been frozen for almost a decade.
>>>>> To break this backwards libgloss->newlib dependency, duplicate that
>>>>> header here so we can keep libgloss independent as it's meant to be.
>>>>
>>>> This isn't really 'newlib internals', it's a header file that tries to
>>>> provide ACLE[1] compatibility for older versions of GCC that lacked such
>>>> support.  Having two copies of this is a maintenance burden, so I'm not
>>>> entirely sure this is a great thing to do, even if the copies are
>>>> supposed to be identical.
>>>
>>> newlib already has 2 itself.  so this will be a 3rd.  i don't disagree with
>>> the maintenance concern, but the fact the file hasn't changed in a decade,
>>> and seems unlikely to ever change, makes me not worry about it.
>>>
>>>> If we can agree on a common location in the source tree that both newlib
>>>> and libgloss can pull this from, then I'm happy to move it if that would
>>>> make you happier.
>>>
>>> libgloss is supposed to be C library agnostic.  the C library (newlib) itself
>>> relies on the output of libgloss (e.g. the crt and low level syscalls).  since
>>> there is no other tree/project in play that i'm aware of, that means there are
>>> really only three options:
>>> * have the compiler provide it
>>> * have libgloss provide it (and newlib uses that)
>>> * duplicate the header
>>>
>>> i know the libgloss/newlib separation is still pretty unclean due to the two
>>> projects historically being one (i.e. everything in newlib), but i don't think
>>> that's a good reason to keep it messy with libgloss depending on newlib.
>>
>> Why not just <toplevel>/include then?  It already contains target-specific
>> stuff in the opcodes subdir, or the xtensa headers.  Having to share them
>> between newlib and libgloss should be reason enough to move the file there.
> 
> the file is currently installed under machine/.  none of the installed headers
> use it though, so maybe it doesn't really need to be installed.  if that's the
> case, moving it into the top-level include/ would work.
> -mike

The file should not be installed, it's only for the build system.  If it 
is being installed, then that's a bug.

Moving it to <toplevel>/include is ok by me.   We'll need to sync it to 
the binutils repo (to support all-in-one build trees), but that's not a 
real issue as the file doesn't change often as you've already noted.

R.