OS X binaries should issue a better warning on incompatible CPUs

[sagetrac-GeorgSWeber]

Currently, every now and then a user reports on sage-support, that he got an error message like

/Applications/sage-4.0.1-OSX10.5-PowerPC-PowerMacintosh-Darwin/sage/
local/bin/sage-sage: line 198:   407 Illegal instruction     sage-
ipython "$@" -i

This is e.g. the case if a Sage binary built on a MacPPC with a G5 processor (typically the one the OS X 10.5 bdist is created on) is used on a MacPPC with only a G4 processor.

For the *nix world on Intel/AMD processors, the sage-flags.txt file was created for just the purpose to check whether the CPU is sufficient, to let a certain Sage binary run.

We seem to need something in that direction for OS X, too (though momentarily only for the PPC processors, not the Intel ones).

CC: @dimpase

Component: distribution

Reviewer: Jeroen Demeyer

Issue created by migration from https://trac.sagemath.org/ticket/6414

[kcrisman]

comment:1

Is this still an issue with the new checks for moving OSX etc?

[kcrisman]

comment:2

This is essentially the same as #5950, except newer and with a little more info, so I'm keeping it. Both of these should be addressed, I guess. From #5950:

-bdist should add something analog to the SSE flags on Linux so that if someone tried to run a ppc build on an x86 it aborts with a sane error message instead of just blowing up.

[jdemeyer]

Reviewer: Jeroen Demeyer

[jdemeyer]

comment:3

There should be no warning needed, the binaries should be produced to run on any CPU.

[kcrisman]

comment:5

There should be no warning needed, the binaries should be produced to run on any CPU.

? That doesn't really work on OS X all the time, we have seen many error messages on ask.sagemath for this for different instruction sets.

[kcrisman]

comment:7

(By which I mean to say this isn't just a PPC/Intel issue.)

[dimpase]

comment:8

it does not work all the time because of problems in the building process (forgetting to set proper flags), or perhaps toolchain bugs, or both, I don't really know.

SAGE_FAT_BINARY cooked with wrong FAT... :-)

[jdemeyer]

comment:9

Replying to @dimpase:

it does not work all the time because of problems in the building process (forgetting to set proper flags), or perhaps toolchain bugs, or both, I don't really know.

I haven't heard of any such bug recently. In any case, such things should be fixed on a case-by-case basis, I don't think that there is a general fix possible. So I still think that this ticket should be closed.

[dimpase]

comment:10

Replying to @jdemeyer:

Replying to @dimpase:

it does not work all the time because of problems in the building process (forgetting to set proper flags), or perhaps toolchain bugs, or both, I don't really know.

I haven't heard of any such bug recently.

certainly with 6.9 binaries it is the case (e.g. reported today on sage-support). There were also very entertaining bugs like this reported by someone with a Mac box having a non-standard (upgraded?) CPU...
Don't recall about 7.1/2 right now.

By the way: faq-usage says: Nobody has yet figured out how to build Sage in such a way that MPIR and ATLAS work on all hardware.

In any case, such things should be fixed on a case-by-case basis, I don't think that there is a general fix possible. So I still think that this ticket should be closed.

[jdemeyer]

comment:11

OK, then let's close this ticket not as "worksforme" but as "invalid" because it's too general.

[kcrisman]

comment:12

That seems like a good compromise.