Re: [PATCH 3/4] gdb: add type::length / type::set_length

[Lancelot SIX <lsix@lancelotsix.com>]

On Wed, Sep 21, 2022 at 10:44:08AM -0600, Tom Tromey wrote:
> >>>>> "Lancelot" == Lancelot SIX via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> >> -		  TYPE_LENGTH (array_type) = (array_bitsize + 7) / 8;
> >> +		  array_type->set_length ((array_bitsize + 7) / 8);
> 
> Lancelot> Do you think that while at it you could also replace the 7 and 8
> Lancelot> literals with appropriate configured values?
> 
> ...
> Lancelot>                             / HOST_CHAR_BIT);
> 
> Normally I think the '8' is a target value, not a host value.
> Like, if HOST_CHAR_BIT != 8, would this do the wrong thing?

Hi,

In this case I believe we do want the host value.  The description of
the m_length field in struct type says:

  Length of storage for a value of this type.  The value is the
  expression in host bytes of what sizeof(type) would return.
  [...]
  Since this field is expressed in host bytes, its value is appropriate
  to pass to memcpy and such.

My understanding from this is that when doing some_type->set_length ()
the argument should be HOST_CHAR_BIT aware.

That being said, I do not think GDB is ever executed on any non-8bit
bytes hosts.  Part of the comment I quoted above even states:

  it is assumed that GDB itself always runs on an 8-bits addressable
  architecture

As a consequence, I think the current literals are valid, but giving
them an explicit name helps as it reminds that this is not a target
byte.

Also, as there are no non-8bit bytes targets supported to my knowledge,
I am not sure if GDB is currently viable for such use-case.  But since
we have some some infrastructure to support this, I try to account for
it were possible (I am not always sure to what extent I have it right
though).

Lancelot.

> 
> Tom