bug-gnulib
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Test failures on legacy Darwin/powerpc


From: Bruno Haible
Subject: Re: Test failures on legacy Darwin/powerpc
Date: Fri, 10 Nov 2023 00:21:16 +0100

Jeffrey Walton wrote:
> > If I recall correctly, those old PowerMac use 128-bit floats from
> > IEEE754 by default.
> > 
> > I think you should build with `-mlong-double-64` on the old PowerMacs.

This advice would work for a program; it does not work for a library or for
Gnulib. The reason is that '-mlong-double-64' changes the ABI. This means,
in particular, the program could no longer pass 'long double' values to
*printf (because *printf in libc is compiled with the original ABI).

Sevan Janiyan wrote:
> Thank you for the hint, that did the trick on 10.4 (G5) & 10.5 (G4)
> PASS test-float (exit status: 0)
> 
> Should m4/float_h.m4 be modified to pass -mlong-double-64 on 
> Darwin/powerpc or it's left to the operator?

It can't, because that's an option that changes the ABI.

Even compiler options that don't change the ABI, like -mieee-with-inexact
on Linux/alpha, cannot be enforced, because we don't control (and shouldn't
dictate) the compiler option choices of the package that uses Gnulib.

Bruno






reply via email to

[Prev in Thread] Current Thread [Next in Thread]