the definition of the work array zwv is wrong!
the keyword used should be kistart instead of kjstart.

      CALL wrk_alloc( jpi+4, jpj  , zwv, kistart = -1 )

Which means that the array bounds are (1:jpi+4, -1:jpj-2) instead of (-1:jpi+2, 1:jpj) .
I don't know why we never saw this bug which was introduced with the wrk_alloc during changeset:3294
Check bounds compilation option are not working in this case?

Am I wrong somewhere?

Sorry if this is the same all over again, but here is another one from limthd.F90:

! --- Energy received in the lead, zqld is defined everywhere (J.m-2) --- !
zqld = tmask(ji,jj,1) * rdt_ice * &
     & ( pfrld(ji,jj) * qsr_oce(ji,jj) * frq_m(ji,jj) + pfrld(ji,jj) * qns_oce(ji,jj) + qemp_oce(ji,jj) )

qsr_oce is not initialised, at least not in the first execution of the routine. Leads to floating point exception and crash.

Coupled configuration and confirmed by initialising qsr_oce to HUGE() in sbc_ice.F90.

The documentation of the trunk version need to be updated. Two main issues :

(1) - Phasing with the v3.6_STABLE documentation. (2) - Phasing with enhancement and simplifications performed in 2015

——— associated branch: ​none (performed directly on the trunk itself) associated wiki page ​​https://forge.ipsl.jussieu.fr/nemo/wiki/ticket/1673_DOC_trunk).