atlas. dyld: Symbol not found: _dlartg_

[killserver@…]

I've installed atlas @3.10.1_5+mpclang33+nofortran and octave.

But octave terminates with:

dyld: lazy symbol binding failed: Symbol not found: _dlartg_
  Referenced from: /opt/local/lib/libqrupdate.1.dylib
  Expected in: /opt/local/lib/libtatlas.dylib

dyld: Symbol not found: _dlartg_
  Referenced from: /opt/local/lib/libqrupdate.1.dylib
  Expected in: /opt/local/lib/libtatlas.dylib

panic: Trace/BPT trap: 5 -- stopping myself...

when i use function fminunc.

Command

nm /opt/local/lib/libtatlas.dylib | grep dlartg

returns nothing

[killserver@…]

Cc Me!

[seanfarley (Sean Farley)]

Can you provide a running example so I can debug this on my laptop?

[killserver@…]

code, that produce error

[killserver@…]

Replying to sean@…:

Can you provide a running example so I can debug this on my laptop?

I've attached code.

To run this code you should start octave in the same directory and type 'ex'.

[killserver@…]

I've tested atlas with default options (i.e. atlas @3.10.1_5+mpclang33) and it works fine.

Anyway the atlas version with no fortran seems to be broken.

[Veence (Vincent)]

If you compile Atlas absent the fortran option, you get only the core library (Cblas) and not the husk (i.e. Lapack). QRUpdate one way or another seems to depend on Lapack…

[seanfarley (Sean Farley)]

Should we require atlas-lapack when the +atlas variant of octave is set? I would say 'yes' unless there is a really good reason?

[Veence (Vincent)]

I think the nofortran option of Atlas might be a flub. I seriously plan to removed it with the next update.

[seanfarley (Sean Farley)]

Replying to vince@…:

I think the nofortran option of Atlas might be a flub. I seriously plan to removed it with the next update.

I would tend to argue with you about the +nofortran option. I think atlas should always provide a functional lapack.

[Veence (Vincent)]

I’m going to make the necessary modification, then close this ticket.

[Veence (Vincent)]

• Not relevant anymore -