emulation has fewer bugs than certain hardware
On Fri, Mar 09, 2007 at 10:57:07PM +0100, Wouter Verhelst wrote:
> On Thu, Mar 08, 2007 at 08:41:43PM -0600, Drew Scott Daniels wrote:
> > > > 68k seems to have elected to skip official etch, but also seems to
> > > > have met the requirements. Some of the non-dd porters still want
> > > > an official etch release.
> > >
> > > (They met the requirements after the architecture freeze, and only using
> > > emulated buildds, which needs some more discussion before being suitable
> > > for release.)
> > This should be discussed then. The consensus I've been reading on
> > debian-68k is that emulated buildds produce better code than real
> > hardware.
> Actually, no, the consensus is that it's a good idea because it's faster
> in some cases. It doesn't produce "better" code, since the compiler is
> (presumably) the same.
I meant "better" as in fewer bugs.
"Aranym is apparently getting faster, but we'll then be facing emulation
bugs as well as real ones. Given our hardware, that's probably not a
big deal"  I guess one could interpret this as meeing speed and not
reliability. We'll likely hear more from him.
"The only way to know is to try, which I did.
I never hit any issue with my distcc/crosscc cross-compiling setup.
With aranym, I hit some issues with the FPU emulation but they did not
affect the packages: Octave build fine but the test-suite displayed
problem inside the emulator, however the test-suite run on the same
binary on real hardware did not, so the package was correctly built."
I recall reading a better quote about hardware having more problems.
Things like bad disks, overheating CPU's, questionable drivers for
certain hardware? Certainly there's a case based on speed. Given Bill
Allombert's results, and the number of build failures I've seen due to
hardware, I think many people think emulation is a good choice. But then
this really isn't a d-vote topic. Let's take this to debian-68k if we want
to say more. I think a more clear concise set of arguments for/against
items needs to be articulated. I may look into doing this on a wiki with
help from the lists.
I think 68k is becoming an embedded architecture by todays standards and
should have the exceptions granted to them. Otoh, SLIND seems to mostly
just endorse using cross compilers to recompile extra packages. It might
be nice to add m68k to their efforts.
I do see some dissent to emulation. Most of it doesn't seem to be based
on any experience/fact/hard date though.
<20070227133144.GS6299@marenka.net>, Stephen R Marenka, 2007
<20070211104918.GF3151@yellowpig>, Bill Allombert, 2007