Re: [RFA] New bitflags type and eflags on i386/x86-64

[Daniel Jacobowitz <drow@mvista.com>]

On Mon, Apr 29, 2002 at 06:52:08PM +0200, Pierre Muller wrote:
> At 17:45 22/04/2002 , Daniel Jacobowitz a écrit:
> >On Mon, Apr 22, 2002 at 05:15:34PM +0200, Michal Ludvig wrote:
> > > Hi all,
> > > I've created a new typecode TYPE_CODE_FLAGS with appropriate functions 
> > > and used it in builtin_type_i386_eflags type. I did this to be able to 
> > > print i386's and x86-64's FLAGS register in a symbolic form, instead of 
> > > printing it in a hexadecimal and decimal notation.
> > > 
> > > Now it looks like this:
> > > (gdb) info registers eflags
> > > eflags         0x747    [ DF IF TF ZF PF CF ]
> > > 
> > > I've chosen quite a generic way for implementation, so that the others 
> > > could use this for their types as well. For now I'm using this  type 
> > > only on x86-64, but using it on i386 should be possible without 
> > > modifications. (BTW Should I do it or the maintainer will?)
> > > 
> > > Any comments? Can I commit?
> >
> >First of all, please include ChangeLog entries; it makes patches easier
> >to digest quickly.
> >
> >Second, I see that you assume a TYPE_CODE_FLAGS type is the size of a
> >long.  I'm not fond of that.  I would prefer if you instead added
> >support to c-valprint.c for something like Pascal's TYPE_CODE_SET (see
> >p-valprint.c) and used that.  It should be exactly what you're looking
> >for.  Basically, you create an enum describing the bit position (not
> >mask) for each flag, and then call create_set_type with that type as
> >the domain_type.
> 
> Beware that if you try to use this in C language, 
> you will get an error that C language doesn't know anything about TYPE_CODE_SET.
> Furthermore, if you force the use of the pascal version of printing sets,
> you are still not safe:
> Pascal set valprint code does print a set with ranges:
> if you have a pascal type
> tset = set of [1..16];
> and a variable a of type tset, that contains
> 2,5,6,7,8,15
> it will be dispalyed as 
> [2,5..8,15]
> this is also the case for enumerated flags,
> but this is bad for the flag enumeration.

Sure it is.  I'd rather simply fix this for enumerated flags, which it
does not make sense to abbreviate in this fashion.

> Given this info, I would support the creation of a TYPE_CODE_FLAGS 
> rather than trying to use pascal sets.
> Note that the code submitted does failm to write the flags if language is set to 
> pascal for instance... So this means that the p-valprint and
> maybe others (f-valprint or jv-valprint code should be adapted too).
> So maybe the best would be to move thiscode from c-valprint.c to valprint.c
> and simply call it for the different language-valprint.c sources.

Yes, certainly.  Probably even for TYPE_CODE_SET, which I still prefer.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer