Re: USE_LSB_TAG not supported on this platform

[Paul Eggert]

On 02/09/2016 08:59 AM, Eli Zaretskii wrote:
> Thanks.  Out of curiosity: which hosts can behave like that?

16-bit PDP-11s running 7th Edition Unix. malloc returned only a multiple 
of 2 there. :-)

I think there were 32-bit platforms where malloc returned only a 
multiple of 4, but can't find any offhand now. For what it's worth, 
glibc documents a minimum of 8, for all glibc platforms.

> Also, why are we sure that the loops will end at some point on those
> hosts?  Shouldn't we perhaps set a limit to the loop iterations, to be
> sure we don't infloop there?

We're not absolutely sure. Certainly the C standard doesn't guarantee 
it; malloc can return a pointer that is always odd, on weird platforms 
where alignof always returns 1. I view this as almost purely theoretical 
though, due to the practical performance benefit of alignment to at 
least sizeof(double). It's conceivable (though very unlikely) that Emacs 
will infloop on some truly oddball platform that does not care about 
performance; but if that happens it'll be OK, as the infloop would 
almost surely happen during a build and the builder would then send us a 
bug report and we can deal with it then. I think adding a counter would 
complicate the code (and possibly introduce bugs, in code that's never 
really exercised) for not enough benefit.