Installing Atlas Hangs

[jonlederman@…]

Hi,

I am trying to install Octave, upon which Atlas is a dependency. I have a recent macbook pro with SSD and plenty or RAM. The Atlas installation never finishes even though I have let my computer run for several days. It always gets stuck at the point I indicate below. How can I at least diagnose and fix this problem? Please see below.

Thank you.

Jons-MacBook-Pro-2:~ Jon$ sudo port clean atlas
Password:
--->  Cleaning atlas
Jons-MacBook-Pro-2:~ Jon$ sudo port clean atlas
--->  Cleaning atlas
Jons-MacBook-Pro-2:~ Jon$ sudo port install octave -v
--->  Computing dependencies for octave
--->  Dependencies to be installed: atlas bison m4 curl curl-ca-bundle libidn epstool ghostscript autoconf automake jbig2dec libpaper fftw-3 fftw-3-single flex gawk glpk gperf grep pcre gsed hdf5-18 less pstoedit ImageMagick djvulibre urw-fonts webp plotutils qhull qrupdate transfig netpbm jbigkit
--->  Fetching archive for atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33.darwin_13.x86_64.tbz2 from http://packages.macports.org/atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33.darwin_13.x86_64.tbz2 from http://lil.fr.packages.macports.org/atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33.darwin_13.x86_64.tbz2 from http://mse.uk.packages.macports.org/sites/packages.macports.org/atlas
--->  Fetching distfiles for atlas
--->  Verifying checksums for atlas
--->  Extracting atlas
--->  Applying patches to atlas
--->  Configuring atlas
Selected C compiler: /opt/local/bin/clang-mp-3.3
Selected F77 compiler: gfortran48
--->  Building atlas
Jons-MacBook-Pro-2:~ Jon$ sudo port clean atlas
Password:
--->  Cleaning atlas
Jons-MacBook-Pro-2:~ Jon$ sudo port install octave -v
--->  Computing dependencies for octave
--->  Dependencies to be installed: atlas bison m4 curl curl-ca-bundle libidn epstool ghostscript autoconf automake jbig2dec libpaper fftw-3 fftw-3-single flex gawk glpk gperf grep pcre gsed hdf5-18 less pstoedit ImageMagick djvulibre urw-fonts webp plotutils qhull qrupdate transfig netpbm jbigkit
--->  Fetching archive for atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33.darwin_13.x86_64.tbz2 from http://packages.macports.org/atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33.darwin_13.x86_64.tbz2 from http://lil.fr.packages.macports.org/atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33.darwin_13.x86_64.tbz2 from http://mse.uk.packages.macports.org/sites/packages.macports.org/atlas
--->  Fetching distfiles for atlas
--->  Verifying checksums for atlas
--->  Extracting atlas
--->  Applying patches to atlas
--->  Configuring atlas
Selected C compiler: /opt/local/bin/clang-mp-3.3
Selected F77 compiler: gfortran48
--->  Building atlas

[ryandesign (Ryan Carsten Schmidt)]

Atlas should take many hours to compile on modern hardware, but not days. You can check the logfile, whose location is printed by the command:

port logfile atlas

but it's possible the end of the log is truncated. You could run the install with the verbose flag to see exactly what it's doing at the time of the hang:

sudo port -v install atlas

Note that single-letter flags like -v must appear between port and the command (install in this case), and have no effect if used elsewhere on the command line.

[mkae (Marko Käning)]

Cc Me!

[mkae (Marko Käning)]

Same here on Mavericks and with variant nofortran at

make[5]: Entering directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_math_atlas/atlas/work/ATLAS/build/tune/blas/ger'
/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_math_atlas/atlas/work/ATLAS/build/bin/xextract -b /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_math_atlas/atlas/work/ATLAS/build/..//tune/blas/gemv/atlas-l2g.base \
                           order=clmajor rout=r1_sse nu=8 type=SREAL \
                           -def MU 2  -o EXTDIR/sr1_sse.c
make[5]: Leaving directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_math_atlas/atlas/work/ATLAS/build

[Veence (Vincent)]

How long did you wait and what options did you use?

[mkae (Marko Käning)]

6+ hours on an i7. :-(

It was variant "nofortran", because the standard install complained about fortran not installed. (Strange that that wasn't resolved through a dependency.)

[Veence (Vincent)]

I know, that's pretty lengthy, but it's part of Atlas internal deep optimization procedure. Better build the port during the night.

As for the "nofortran" option, the fortran API is not required to build Atlas. So I found this way for people willing to opt out. But, you're right, it might be worth trying to find a better way to tackle this.

I surmise you got Atlas all right in the end, so I close this bug. Reopen it if my assumption is wrong.

[mkae (Marko Käning)]

It hadn't built yet, after those 6+ hrs.

Should I wait even longer???

(So, I don't know whether this issue is to be validly considered invalid.)

[mkae (Marko Käning)]

Replying to mk@…:

It hadn't built yet, after those 6+ hrs.

I gave building atlas a 2nd chance. This time I installed gcc48 before trying it. That way it uses gfortran48 and I don't need to use variant nofortran.

I am beyond 12h build time now, but it is doing something, still...

Right now I realise that it began to actually use more than only one core (running xdlanbsrch_pt it seems to work on 6 cores)!

So, I am hoping it is going to finish at some time. (Still, I can't remember that it ever took that long to build atlas in the past.)

[Veence (Vincent)]

I suspect this is caused by a bug in clang-3.3. I am going to update the port and switch the default compiler to the system clang, we should now be more advanced than clang-3.3. At the same time, I am going to upgrade the clang-3.3 option to clang-3.4. Thanks for you patience.

[mkae (Marko Käning)]

I gave up waiting for the compiler to finish after 20h or so... :-/

So, I hope you can get it going with clang-3.4.

[Veence (Vincent)]

I won’t be able to commit the port before Wednesday. Meanwhile, try the +clang option. Sorry for the delay!

[mkae (Marko Käning)]

I tried that too.

Same result.

Never finishes building... :(

[mf2k (Frank Schima)]

Cc'ing reporter of duplicate #42557.

[Veence (Vincent)]

Please retry with the attached Portfile and tell me what you get.

[Veence (Vincent)]

New "more up-to-date" Portfile

[mkae (Marko Käning)]

You should only post the diff to the current port file. That makes review easier.

[Veence (Vincent)]

Oops, sorry. I attach the diff too.

[mkae (Marko Käning)]

Why have you not blacklisted clang as well as mp-clang 3.3? (Those didn't work for me at least on 10.9.)

[Veence (Vincent)]

I was concerned this could cause "collateral effects" to people who have Atlas +clang33 installed.

[mkae (Marko Käning)]

Fair enough, you're right.

[mkae (Marko Käning)]

It's really a pain that atlas can't be build parallel... :(

[Veence (Vincent)]

I know, but it would interfere with Atlas optimization routines, and defeat the whole purpose of the build…

[mkae (Marko Käning)]

The build finished with errors. The gzipped main.log is rather long, but I guess I have to attach it anyways.

[mkae (Marko Käning)]

unsuccessful build ( sudo port install +mpclang34 )

[Veence (Vincent)]

At least it does not hang forever. I’m going to have a look. On my side, I tried with the Apple clang and it worked fine on a 2013 MBA. Hang on.

[Veence (Vincent)]

Ok, there are several snags in both your and my build. I'm going to have a look later today. Please by patient!

[mkae (Marko Käning)]

Hmm, perhaps it's worth trying clang-3.5?

[Veence (Vincent)]

Oops. Sorry I forgot you. My build worked, eventually. You still have the same problem?

[mkae (Marko Käning)]

Well, since then I've never tried that endavour again. :-)

What did you do to make it work eventually?

[Veence (Vincent)]

Geez, I can't remember. I'll dig into it deeper tomorrow. Sorry for the delay (I live in Europe, it's getting pretty late and I'm pooped). But stay tuned. Apologies once again.

[mkae (Marko Käning)]

Dito. Also heading for bed now. :) Thanks in advance. Good night.

[tim.blass@…]

Replying to vince@…:

I suspect this is caused by a bug in clang-3.3. I am going to update the port and switch the default compiler to the system clang, we should now be more advanced than clang-3.3. At the same time, I am going to upgrade the clang-3.3 option to clang-3.4. Thanks for you patience.

Hello. I ran into a similar (possibly the same!) problem yesterday. I didn't want to start a new ticket because this one is so similar to mine. Has there been any progress on getting atlas to build when trying to install octave?

My info is: I have a macbook air, intel core i7, 8GB of RAM, with OSX 10.9.4.

My octave install failed at first because of no fortran compiler, so I installed one (sudo port install gcc49) then tried the install of octave again (sudo port install octave) and let it run over night. It was "building atlas" for 12+ hours before I finally killed it. According to my activity monitor, there was something going on, but often not very much. Most of the time the my CPU load was below 30%, occasionally spiking up to around 90%.

I decided to retry it with the verbose flag -v to see if that gives any new info ... as of this writing it is still building and not stuck on anything in particular, but has only been going for a few minutes.

[tim.blass@…]

Replying to tim.blass@…:

Replying to vince@…:

I suspect this is caused by a bug in clang-3.3. I am going to update the port and switch the default compiler to the system clang, we should now be more advanced than clang-3.3. At the same time, I am going to upgrade the clang-3.3 option to clang-3.4. Thanks for you patience.

Hello. I ran into a similar (possibly the same!) problem yesterday. I didn't want to start a new ticket because this one is so similar to mine. Has there been any progress on getting atlas to build when trying to install octave?

My info is: I have a macbook air, intel core i7, 8GB of RAM, with OSX 10.9.4.

My octave install failed at first because of no fortran compiler, so I installed one (sudo port install gcc49) then tried the install of octave again (sudo port install octave) and let it run over night. It was "building atlas" for 12+ hours before I finally killed it. According to my activity monitor, there was something going on, but often not very much. Most of the time the my CPU load was below 30%, occasionally spiking up to around 90%.

I decided to retry it with the verbose flag -v to see if that gives any new info ... as of this writing it is still building and not stuck on anything in particular, but has only been going for a few minutes.

After running it again, octave is now installed. It only took the afternoon -- not sure what was different between this install and the previous, but I'm happy it works.

[Veence (Vincent)]

Closing this one too.