This was a proposal by Samuel Wales. After the user selected a
refiles target, we make sure that the fully qualified target is in the
history, so that next time, UP will bring back exactly this target.
Currently we can't have nested function calls, and I think that a general fix for that would require going the standard route of constructing a parse tree and evaluating it recursively.
With this change we avoid messing about extracting the output from the
comint buffer in the :results value case (the value has already been
written to file).
Before this patch the completion mechanism was this: TAB let's you
complete through link prefixes (gnus: file: bbdb: ...) then RET allows
completion if a completion mechanism is available for the chosen prefix.
Navigating through the history of stored links was a separate process,
available through the up/down M-n/M-p keys.
Now TAB not only completes through link prefixes but also through stored
links. This behavior matches other Emacs completion mechanisms a bit
more closely.
Adding support for column/row names (as currently being implemented
for R) on a language by language basis is definitely the right way to
handle this hline/header issue. I am going to try to do the same for
gnuplot.
I don't believe this solves it, but chomp is more the right thing to
do than trim. I'd like to retain all the whitespace so that alignment
of columns is correct in stdout.
Unlike the other languages, it's central to R to be able to index
columns of a data frame d, either by d[,"columnname"] of d$columnname.
With this change, if colnames are present in the *input* from
org-babel, the corresponding R variable is *always* constructed with
the colnames.
In addition, with the :colnames header arg, the *output* to elisp/org
buffer contains the colnames separated from the rest of the table by
'hline. This behaviour is not default because other languages may
expect a simple table without the 'hline.