ORCA-LIM2 segmentation fault with ln_ctl = true

[bac]

When trying to run ORCA025-LIM2 with NEMO v3 on our local SGI Altix ICE system and the UK National service (CRAY XT4), the code fails with a segmentation fault if one set ln_ctl = true. The problem was traced to the lines

CALL prt_ctl(tab3d_1=p_qla , clinfo1=' blk_ice_core: p_qla : ', tab3d_2=z_qsb , clinfo2=' z_qsb : ', kdim=ijpl)

CALL prt_ctl(tab3d_1=z_qlw , clinfo1=' blk_ice_core: z_qlw : ', tab3d_2=p_dqla , clinfo2=' p_dqla : ', kdim=ijpl)
CALL prt_ctl(tab3d_1=z_dqsb , clinfo1=' blk_ice_core: z_dqsb : ', tab3d_2=z_dqlw , clinfo2=' z_dqlw : ', kdim=ijpl)

in subroutine bb of sbcblk_core.F90 where the arrays z_qsb, z+qlw, z_dqlw have their third dimension SIZE(pst,3) which is 1 (number of ice categories). To solve the problem Gurvan and I have declared jpl = 1 in par_ice_2.F90 and then all the arrays in blk_ice_core can be dimensioned (jpi,jpj,jpl). The changed routines are attached
This makes the code more consistent with LIM-3 and I would suggest that the arrays in sbcice_lim_2.F90 which are now (jpi,jpj,1)could also be (jpi,jpj,jpl) .

Beverly de Cuevas

[rblod]

We change the correction to used jpl as da dummy argument in blk_ice_* and have deffered shape arrays. It is needed when we do no want to use sea-ice.