[Developers] linux build and install problems/patches

Gareth.Williams at csiro.au Gareth.Williams at csiro.au
Tue Mar 1 17:34:23 PST 2011


I recently installed admb 10.0 from source (gcc and icc) on one of our clusters and have some feedback that could help you improve that process and it's documentation.

1) README.txt could be simpler in describing the icc installation.  I found that icc was used if it was available in my PATH so the install instructions are the same as for gcc. ie. configure; make; make install

2) I found in both cases the build (or verify) failed unless I set up for a 64bit build
"./configure BUILD_64BIT=yes"
Perhaps this should be the default - or are there advantages in having a 32bit build on a 64bit linux system? (and what would I do to make it go?)

3) the install failed on copying scripts/icpc/admb-bin  but I found a script in scripts/admb/admb-bin so changed the makefile to install that.  Is that OK?

4) I wanted to install in /tools/admb/10.0 (or similar) to potentially manage multiple installations. Using 
"./configure --prefix=/tools/admb/10.0" worked nicely except the install umask options cause the install to fail as the install directories are marked unwriteable.  To avoid installing as root, I changed 444 to 644 and 555 to 755 in  scripts/configure/GNUmakefile.in and it all got better.

5) "make verify" ran a lot of successful tests but failed (both icc and gcc) on:
Successful build of executable: binomial_w2
Block diagonal Hessian (Block size =1)

 inner maxg = 2.06882e-05
max separable g 2.06882e-05
Newton raphson 1
 f = 210.2864498683653 max g = 4.938272013532696e-13
max separable g 4.973799150320701e-13
Newton raphson 2
 f = 210.2864498683653 max g = 7.549516567451064e-15
Error: Invalid index 4 used for array range [1, 3] in "double& dvector::operator[] (int i)".
array bound exceeded -- index too high
perhaps that test is broken?


Gareth Williams

More information about the Developers mailing list