On Wed, Nov 18, 2009 at 1:21 AM, Barry Smith <[email protected]> wrote:
When starting to code PETSc 2.0 I introduced two nasty habits
1) since we called malloc() directly and some mallocs don't like allocating zero length I would often allocate 1 additional space to make sure I did not request a zero size. Giving things like PetscMalloc((n+1)*sizeof(something),&something); This leads to confusing code because we don't know when the 1 belongs or not.
2) in order to reduce the number of mallocs I would do a single malloc for several variables. For example, PetscMalloc((n+m+1)*sizeof(PetscScalar),&buf1); buf2 = buf1 + n;
Later I changed PetscMalloc()/Free() to allow 0 length mallocs. I also introduced PetscMalloc2-3-4-5() to efficiently and cleanly share one malloc. Thus 1 and 2 became unnecessary. Since they are downright disgusting I am trying to eliminate 2) and would like also to reduce 1)
This fixes will likely introduce some hard to determine bugs in the near future (or already have). Please avoid habits 1 and 2 from now on in the PETSc code base.
I would guess that almost all the (n+1) mallocs are unnecessary, except for "I"-like arrays associated to AIJ structures... -- Lisandro Dalcín --------------- Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC) Instituto de Desarrollo Tecnológico para la Industria Química (INTEC) Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET) PTLC - Güemes 3450, (3000) Santa Fe, Argentina Tel/Fax: +54-(0)342-451.1594