[Openmcl-devel] another make-record question

[Andrew P. Lentvorski, Jr.]

On Aug 20, 2004, at 4:53 PM, Cyrus Harmon wrote:

> On Aug 20, 2004, at 12:57 PM, Andrew P. Lentvorski, Jr. wrote:
>
>> Anybody up for changing ld and dylib? <said only half-jokingly>
>
> But isn't this basically what the .cdb files are for? It doesn't 
> bother me too much that we need to ship cdb files with shared 
> libraries, as long as there's a nice clean way to do so. Certainly 
> seems easier than changing ld and dylib!

There is nothing which links a .cdb file to the particular dynamic 
library which is being used.  The disconnect is that the .h is not 
guaranteed to correspond to the .dylib (or .so).  In fact, it is often 
a real PITA to make them correspond correctly.

This issue comes up in far more arenas than just OpenMCL.  Basically 
every VM-based language has this problem with statically compiled 
binaries.  Just look at how many different solutions people propose to 
this problem for all the different languages.

Personally, I like the .ffi solution used in OpenMCL the best out of 
all of the wrapper solutions (SWIG, Gtk stuff, etc.).  However, none of 
them can solve the fundamental disconnect between .h files and .dylib 
files.  Only by making the equivalent to the .h file available for run 
time queries from the .dylib would this be solved.

It would also have the side benefit of easing the coming 32 bit to 64 
bit migration of code.  But that's a topic for another day.

-a