Comments on: A proposal for an Sweave service http://www.reproducibleresearch.net/blog/2009/01/23/sweave-build-service-proposal/ Ideas, interesting papers and news items around reproducible research Tue, 30 Mar 2010 14:21:22 +0200 http://wordpress.org/?v=2.8.4 hourly 1 By: James Quirk http://www.reproducibleresearch.net/blog/2009/01/23/sweave-build-service-proposal/comment-page-1/#comment-49 James Quirk Fri, 23 Jan 2009 23:27:43 +0000 http://reproducibleresearch.org/blog/?p=40#comment-49 It's worth recalling that PDF was conceived as a container mechanism for streamlining document workflows. Therefore there is nothing to stop you asking for a self-replicating PDF. This is a PDF that contains everything needed to re-generate itself and the computational work it reports. In your case this would amount to an Sweave program embedding itself, and whatever data files it uses, into the PDF report it generates. The "build server" would then check the integrity of the report by extracting the embedded files and processing them to see if it could replicate the supplied PDF. Thus the "build server" is more of a "check server." And only PDF's which check out (i.e. can actually replicate themselves) would be allowed out into the wild. I demonstrated a self-replicating PDF at Burton Wendroff's 70th birthday celebration, March 8th 2000; see http://www-troja.fjfi.cvut.cz/~liska/bbw . My application was computational fluid dynamics, but the associated software plumbing is independent of the end-application and so if anyone is interested, I could easily put an Sweave example together? Here it's worth noting that PDF has come a very long way since 2000 and a new, exciting opportunity for reproducible research is appearing on the horizon. Specifically, it is now notionally possible to compile R using Adobe's Alchemy SDK so as to produce a SWF file that could be run directly inside a PDF. I say notionally as Alchemy generated SWF's require FlashPlayer 10 and Acroread 9 currently ships with Flash Player 9. Also, as Alchemy is still in its infancy it can easily be stumped by applications that work with shared objects. But fast forwarding to 2018, I have every confidence that PDF reports will routinely house virtualized machines that allow the interested reader to sample the recorded work firsthand. The obvious market for such PDF's will be electronic text books. But the associated infrastructure, as it matures, will be a tremendous boon for true "reproducible research." It’s worth recalling that PDF was conceived as a container mechanism for streamlining document workflows. Therefore there is nothing to stop you asking for a self-replicating PDF. This is a PDF that contains everything needed to re-generate itself and the computational work it reports. In your case this would amount to an Sweave program embedding itself, and whatever data files it uses, into the PDF report it generates. The “build server” would then check the integrity of the report by extracting the embedded files and processing them to see if it could replicate the supplied PDF. Thus the “build server” is more of a “check server.” And only PDF’s which check out (i.e. can actually replicate themselves) would be allowed out into the wild.

I demonstrated a self-replicating PDF at Burton Wendroff’s 70th birthday celebration, March 8th 2000; see http://www-troja.fjfi.cvut.cz/~liska/bbw . My application was computational fluid dynamics, but the associated software plumbing is independent of the end-application and so if anyone is interested, I could easily put an Sweave example together?

Here it’s worth noting that PDF has come a very long way since 2000 and a new, exciting opportunity for reproducible research is appearing on the horizon. Specifically, it is now notionally possible to compile R using Adobe’s Alchemy SDK so as to produce a SWF file that could be run directly inside a PDF. I say notionally as Alchemy generated SWF’s require FlashPlayer 10 and Acroread 9 currently ships with Flash Player 9. Also, as Alchemy is still in its infancy it can easily be stumped by applications that work with shared objects.

But fast forwarding to 2018, I have every confidence that PDF reports will routinely house virtualized machines that allow the interested reader to sample the recorded work firsthand. The obvious market for such PDF’s will be electronic text books. But the associated infrastructure, as it matures, will be a tremendous boon for true “reproducible research.”

]]>