<p dir="ltr"> That is what I implemented for TS. The problem is false dependency in the sense that the enum includes extra stuff. For example, DM function pointers are needed by dmdasnes. With dynamic keys, we could avoid that spurious dependency.</p>

<div class="gmail_quote">On Dec 4, 2012 8:14 PM, "Barry Smith" <<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
On Dec 4, 2012, at 4:43 PM, Jed Brown <<a href="mailto:jedbrown@mcs.anl.gov">jedbrown@mcs.anl.gov</a>> wrote:<br>
<br>
> As usual, anything that is duplicated and not checked by the compiler is broken.<br>
><br>
> $ grep PetscObjectAllocateFortranPointers src/**/*.c<br>
> src/dm/impls/da/ftn-custom/zda2f.c:  PetscObjectAllocateFortranPointers(*da,6);<br>
> src/dm/impls/da/ftn-custom/zda2f.c:  PetscObjectAllocateFortranPointers(*da,6);<br>
> src/dm/impls/shell/ftn-custom/zdmshellf.c:  PetscObjectAllocateFortranPointers(*dm,2);<br>
> src/dm/impls/shell/ftn-custom/zdmshellf.c:  PetscObjectAllocateFortranPointers(*dm,2);<br>
><br>
> Note that changing the type does not reset the function pointers, thus having a DMSHELL, calling DMSetType(dm,DMDA), and then setting a DMDA local function will cause memory corruption.<br>
<br>
> I cannot express how much I hate this system. The full-blown solution is that for each type, we register a (global) token which is the index of that function pointer. That doesn't have any false dependencies, but is more "initialize" code.<br>

><br>
> An alternative, used in the TS and KSP code below, is to have a common enum that lists all the Fortran functions. It's a false header dependency, but not a binary dependency.<br>
> What should we do? The current state is a disaster.<br>
<br>
<br>
   Each object gets an enum for each fortran implementable method in that object, names of the enums reflect the object and method name. The final enum for each object is the count of the number of the enums and is used in the AllocationFortranPointers.  Yes, some brain numbing work to set up initially but then pretty clean and extensible for the future (since the final enum is the count we don't have to worry about always fixing the count in the AllocateFunctionPointers() each time a new method is added).<br>

<br>
   Barry<br>
<br>
><br>
> src/ksp/ksp/impls/gmres/fgmres/ftn-custom/zmodpcff.c:  PetscObjectAllocateFortranPointers(*ksp,3);<br>
> src/ksp/ksp/interface/ftn-custom/zitfuncf.c:  PetscObjectAllocateFortranPointers(*ksp,FTN_MAX);<br>
> src/ksp/ksp/interface/ftn-custom/zitfuncf.c:  PetscObjectAllocateFortranPointers(*ksp,FTN_MAX);<br>
> src/ksp/pc/impls/mg/ftn-custom/zmgfuncf.c:    PetscObjectAllocateFortranPointers(*mat,1);<br>
> src/ksp/pc/impls/shell/ftn-custom/zshellpcf.c:  PetscObjectAllocateFortranPointers(*pc,5);<br>
> src/ksp/pc/impls/shell/ftn-custom/zshellpcf.c:  PetscObjectAllocateFortranPointers(*pc,5);<br>
> src/ksp/pc/impls/shell/ftn-custom/zshellpcf.c:  PetscObjectAllocateFortranPointers(*pc,5);<br>
> src/ksp/pc/impls/shell/ftn-custom/zshellpcf.c:  PetscObjectAllocateFortranPointers(*pc,5);<br>
> src/ksp/pc/impls/shell/ftn-custom/zshellpcf.c:  PetscObjectAllocateFortranPointers(*pc,5);<br>
> src/mat/impls/mffd/ftn-custom/zmffdf.c:  PetscObjectAllocateFortranPointers(*mat,2);<br>
> src/mat/impls/shell/ftn-custom/zshellf.c:  PetscObjectAllocateFortranPointers(*mat,11);<br>
> src/mat/interface/ftn-custom/zmatrixf.c:  PetscObjectAllocateFortranPointers(*sp,1);<br>
> src/snes/interface/ftn-custom/zsnesf.c:  PetscObjectAllocateFortranPointers(*snes,14);<br>
> src/snes/interface/ftn-custom/zsnesf.c:  PetscObjectAllocateFortranPointers(*snes,14);<br>
> src/snes/interface/ftn-custom/zsnesf.c:  PetscObjectAllocateFortranPointers(*snes,14);<br>
> src/snes/interface/ftn-custom/zsnesf.c:  PetscObjectAllocateFortranPointers(*snes,14);<br>
> src/snes/interface/ftn-custom/zsnesf.c:  PetscObjectAllocateFortranPointers(*snes,14);<br>
> src/snes/linesearch/impls/shell/ftn-custom/zlinesearchshellf.c:  PetscObjectAllocateFortranPointers(*linesearch,3);<br>
> src/snes/linesearch/interface/ftn-custom/zlinesearchf.c:  PetscObjectAllocateFortranPointers(*linesearch,3);<br>
> src/snes/linesearch/interface/ftn-custom/zlinesearchf.c:  PetscObjectAllocateFortranPointers(*linesearch,3);<br>
> src/sys/draw/utils/ftn-custom/zzoomf.c:  PetscObjectAllocateFortranPointers(*draw,1);<br>
> src/ts/interface/ftn-custom/ztsf.c:  PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
> src/ts/interface/ftn-custom/ztsf.c:  PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
> src/ts/interface/ftn-custom/ztsf.c:    PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
> src/ts/interface/ftn-custom/ztsf.c:    PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
> src/ts/interface/ftn-custom/ztsf.c:  PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
> src/ts/interface/ftn-custom/ztsf.c:  PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
> src/ts/interface/ftn-custom/ztsf.c:  PetscObjectAllocateFortranPointers(*ts,OUR_COUNT);<br>
<br>
</blockquote></div>