Re: Bug in the 'mex' garbage collector (octave 2.9.10)

[David Bateman]

John W. Eaton wrote:
> On 29-Apr-2007, David Bateman wrote:
> 
> | As you are talking about build slowness. It seems to me that the *.df
> | files and built and re-built, and for good measure built once again,
> | during a build of octave. We should fix the build process so they are
> | only built once. Also the *.df file is built compared to the old version
> | and then removed. If the *.df is newer than the corresponding *.cc file,
> | I don't think it should be built at all..
> 
> If the .df files is newer than the .cc file, it should not be built.
> I think that part is correct now.  The problem is that the .cc file is
> newer than the .df file, but the change to the .cc file doesn't affect
> the .df file.  The idea was that by comparing and only touching the
> .df file if it has really changed would save updating the things that
> depend on the .df files.  This works well if creating the .df file is
> fast compared to what will happen if the .df file is touched.  But in
> this case, I don't think that is true, and because of the
> dependencies we end up recreating the .df files several times, only to
> see that they haven't changed.  Instead, if we just create them
> whenever the .cc files change, then we will trigger a rebuild of
> builtins.cc, PKG_ADD, and gendoc.cc.  But those files also use
> move-if-change to avoid triggering further unnecessary actions, and
> they are quick to build and there are only a three files, not N, so I
> think it makes sense to keep those using move-if-change.  So I checked
> in the following change.

Thats great.... Debugging changes is much faster with that minor change..

D.