RData File Specification?

Ian,

I am developing a tool for converting a large data frame stored in  
an uncompressed binary (XDR) RData file to a delimited text file.   
The data frame is too large to load() and extract rows from on a  
typical PC.  I'm looking to parse through the file and extract  
individual entries without loading the whole thing into memory.

In terms of some C source functions, instead of doing RestoreToEnv 
(R_Unserialize(connection)) which is essentially what load() does,  
I'm looking to get the documentation I would need to build a  
function "SaveToCSV()" so that I could do SaveToCSV(R_Unserialize 
(connection)).

Where can I get documentation on the RData file format?  Does a  
spec document exist?

I don't think so - basically the sources are all the documentation  
I'm aware of. It's a bit messy, because R supports so many old formats.

However, if you want a stand-alone program that handles  
(uncompressed) XDR2 only, then I may have saved you a bit of work. I  
have a utility (based on the R sources) that allows you to scan  
through XDR2 files and to extract individual objects into a separate  
XDR2 file (this happens to be quite useful when you have a workspace  
that doesn't load into R and yet you want to save some pieces of it).  
Have a look at
http://urbanek.info/rdcopy.c

(you can either run it as "./rdcopy foo" to list the objects or "./ 
rdcopy foo -v" to show the full structure (all SEXPs with their  
offsets) or "./rdcopy foo bar 19" to copy SEXP at offset 19 from foo  
into a separate XDR2 file bar (use offset from the first call to copy  
entire objects).

It's not prefect, but servers its purpose (it resolves references by  
copying them instead of re-indexing, but it doesn't detect loops).  
Maybe it helps, even though the task you describe is still far from  
trivial.

Cheers,
Simon