Port-i386 archive

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

Re: MATH_EMULATE



On Mon, 14 Jan 2008, Daniel Carosone wrote:
Sure, there are plenty of things to consider.  Here are a few more:

- None of the netbsd developers disagrees with you, in that it would
  be nice or desirable to have every possible system supported and
  working.

- However, someone needs to show up and do the work. Offers of
  testing are all very well -- and absolutely essential and welcome --
  but are not, in themselves sufficient, someone needs to write or
  maintain the code that is to be tested too. Perhaps someone will in
  this case.

I sure understand that offering testing is not enough, and I would offer more if I were sure that I could deliver the work - but I'm not.

- So far it seems that the MATH_EMULATE code has bitrotted and is
  broken, and may have been for some time.  So it's not just a matter
  of somebody missing a mailing list posting announcing removal of
  something that worked - it's also a case of nobody having reported
  a bug about the problem in all that time.  That doesn't mean
  nobody's using it on old versions, but it does strengthen the
  perception that maybe it won't be much missed in the next release.

In my case, I was under the impression that the fact fpu emulation has bitrotten was widely know - my mistake probably. But yeah, I understand your point - the interest in it isn't strong - but still, there's some.

- Removing MATH_EMULATE from the kernel *doesn't* mean machines
  without FPU can't run netbsd, or won't be supported. It just means
  that they will need their binaries compiled with -msoft-float,
  rather than relying on emulated cpu insructions.  This will be
  faster anyway, and such machines these days will be used for
  special-purpose cases and will benefit in other ways from
  special-purpose builds specific to their needs.

I understand that, but - are we really sure that it's MATH_EMULATE what is broken? If I understood it correctly, the failing ld.so.conf mechanism for switching libm versions could cause this problem as well. Correct me if I'm mistaken.

Vit Herman



Home | Main Index | Thread Index | Old Index