Re: Is a Raptor Blackbird (or other Power machine) a good general-purpose desktop?
Le 23/03/2023 à 00:43, Riccardo Mottola a écrit :
Hi,
Didier Kryn wrote:
Another concern is that a software which does run only on one single
endianness proves to be buggy and loosely written. High level software
such as Firefox should be independant of such considerations, exactly as
it should not rely on internal details of the implementation of
libraries, the libc in the first place. In this respect, the revival of
Linux on BE arches -- together with libcs alternative to glibc -- would
be a big service to the Linux ecosystem.
That's the theory. Practice is different.
Yes, always (~:
E.g. if you use GNUstep, an OpenStep/Cocoa reimplementation which has
multiplatform in by design, your life is happy. Most endianness problems
are solved inside, so if you write an application in it, it will be
cross-platform, except if you wrote some low-level code code with
graphics, network byte swapping or such.
You could still have issues, as certain code (e.g. shifts, swaps, casts,
signed/unsigned issues) works in one endiannes and not the other or
vice-versa.
That's the point. casts and swaps should be encapsulated in very
low-level routines. They are overused, by facility il places they should
not. The C language implements implicit type conversion and the compiler
can handle safely sign issues; the programmer should keep her/his hands
off of it; it is a question of discipline. There is, unfortunately a
culture of terse programming in C which goes against safety.
something like a browser, however, is a mess. It handles a lot of stuff
wuite low-level, graphics layers, GL, sound, countless image and video
codec libraries. Plus JS script support for your specific CPU.
I understand that a browser is a gnu ("a big animal", as the LaTeX
manual stands). But, the very low-level graphics components should,
ideally, be encapsulated. Dunno how JS works; I'm not considering myself
a great programmer (~:
Just look at TenFourFox and the various bug reports and patches Cameron
proposed to mozilla which sometimes got accepted, sometimes ignored.
Most noticeably SKIA noit being interested in BE at all, as well as
issues with Cairo.
I am working on the ArcticFox browser and try to import most of these
fixes ftom TenFourFox to make them available on a browser not limited to
Mac.
But it is a pain and a pity to know "upstream" is diverging more and more.
Currently, ArcticFox has only minor issues compared on PPC to itself
built on Intel or ARM. Help appreciated.
I can assure you of my admiration for such a work. And my thanks.
For me, the only real endiannes is Big-Endian, as were many classic
CPUs, Motorola 68k, classic MIPS, PPC, SPARC, HP-PA.
I hoped Risc-V would be... and think that PPC-le is betrayal like
MIPS-le. Like a BE VAX would have been betrayal! But this is personal.
I have also very much progrmmed for BE during my carreer, M6809,
M68k and PPC; done much low-level VME also. all in C. Around the end, I
had a rather big and complex project (with respect to my skills), with 4
persons involved. We made a cultural revolution: made a review of what
language would be best and chose one which was neither C/C++ nor Java.
The result was excellent: high performance, good readability and 99%
bugs detected at compile time. The language you speak decides in part
the way you think; and this is true also for programing languages and
you learn a lot when you learn a new language; I'm sure we could not
have reached such a result in ~ 3 years, with our original culture of
C/C++ programmers.
I remember a discussion about which of LE and BE was natural, maybe
it was on this list. This purely a question of taste (~:
Never used a MacIntosh, but very many single board computers which
had the cpus mentionned above. Nowadays my laptop has an amd64, like
everybody. But I would be ready to pay more for a ppc or a riscv.
Cheers, and many thanks for your work.
Reply to: