[Developers] cactus feature request: source code saves

[Denis Pollney]

On Tue, 2005-05-24 at 19:09 +0200, Erik Schnetter wrote:
> On Sunday 22 May 2005 13:28, Denis Pollney wrote:
> > Hi,
> > I had some discussion with frank and erik yesterday regarding the
> > reproducability of runs. We often introduce small changes to the code
> > we're running in our local directories without commiting them to CVS,
> > since we're just trying something out. We then might later use the
> > same source tree, forgetting that it is slightly modified.
> >
> > Ideally each run output directory should have a copy of the source
> > code that was used to produce the output.
> >
> > The easiest thing to do would be to just record the entire source
> > tree with each run. This means an overhead of more than 200Mb of
> > data, which for runs with high I/O is no big deal, but for hundreds
> > of small runs would start to add up. And it's a waste, given that
> > most of the recorded source will be identical.
> 
> There is now a rough implementation of this in thorn 
> AEIThorns/Formaline.  If you want to try this, compile Formaline into 
> your executable and activate it at run time.  So far, it works on my 
> local machine, and the code might need some refinement to work on all 
> architectures.

Hi,
This thorn works well on peyote, I was able to easily recreate an
executable from the saved source.
For the particular configuration I've been using, the executable
is 16Mb compared to 11Mb without Formaline, and the saved tar files
are the size of the difference, ie. 5Mb. This is for a run which
will likely produce about a GB of data, so it's a pretty minimal
overhead.
Thanks, Erik!
Denis.

-- 
----------------------------------------------------------------------
Denis Pollney <Denis.Pollney at aei.mpg.de>      office: +49-331-567-7623
Max-Planck-Institut fuer Gravitationsphysik   mobile: +49-170-302-3132
Albert-Einstein-Institut, Am Muehlenberg 1, D-14476, Golm, Germany