When using ':results value' in certain situations, results are written
to file by the foreign language process and subsequently read from
file by emacs into an elisp table structure. If the foreign language
process is running remotely, then the results are written
remotely. These changes ensure that in that case, an appropriate
remote file name is constructed to read the remote data.
These changes solve two problems: both are discussed in the following thread
http://lists.gnu.org/archive/html/tramp-devel/2010-02/msg00025.html
of which a summary follows.
Firstly, shell-command-on-region does not work with tramp in the same
way that shell-command does. I.e. whereas
(let ((default-directory "/user@remote-host:"))
(shell-command "hostname" t))
gives the remote hostname,
(let ((default-directory "/user@remote-host:"))
(shell-command-on-region (point) (mark) "hostname" t))
does not.
The reason is that shell-command-on-region calls call-process-region,
which does not use a tramp handler for remote files. However, such a
file handler does exist (unused) in the tramp sources:
tramp-handle-call-process-region. There is a slight problem in that
there is a bug in that function definition in current tramp (which has
persisted because the function is not normally used).
Therefore, we define an org-babel version of
tramp-handle-call-process-region which fixes the bug, and we bind
call-process-region to org-babel-tramp-handle-call-process-region for
the duration of org-babel-execute-src-block.
This introduces a new header argument :dir. For the duration of source
block execution, default-directory is set to the value of this header
argument. Consequences include:
- external interpreter processes run in that directory
- new session processes run in that directory (but existing ones are unaffected)
- relative paths for file output are relative to that directory
The name of a directory on a remote machine may be specified with
tramp syntax (/user@host:path), in which case the interpreter
executable will be sought in tramp-remote-path, and if found will
execute on the remote machine in the specified remote directory.
Use shell-command-on-region, bringing R into line with ruby, python,
clojure, shell. In addition to consistency (and perhaps efficiency),
this will also be preferable in the case of a remote R process because
it means that transfer of the input to the remote process is dealt
with automatically by tramp.
Before this change we have
#+begin_src ruby :results value :session
["1", "3"]
#+end_src
#+results:
: 1", "3
After, we have
#+begin_src ruby :results value :session
["1", "3"]
#+end_src
#+results:
| 1 | 3 |
This change provides for better export of named source-code blocks,
with specific support for more attractive html export. The arguments
are included with the source code name, and both the source code name
and code body are wrapped in a div to support styling with css. For
example the following CSS can be used to associate a source-code block
name with it's code body, and to slightly indent the body.
.org-src-container {
border-left: 4px solid gray;
padding: 0.5em 0.5em 0.5em 1em; }
.org-src-container pre {
margin-left: 1em; }
the progress-marker local variable set inside of a let in
org-exp-blocks is used to constrain searches in org-babel-exp to
ensure that a babel block doesn't overwrite non-babel src blocks
When exporting to html the source-code block name will be included
as a <label> immediately preceding the exported code. This <label>
is decorated with the 'org-src-name' css class.
When exporting to LaTeX with the listings package the source-code
block name will be included as a listings \title for the code
block. Currently the name will not be exported when the listings
package is not in use.
So, for example since the following <<example>> noweb reference
appears behind the SQL comment syntax the entire inserted body will
also be SQL commented
-- <<example>>
expands to
-- this is the
-- multi-line body of example
Note that noweb replacement text which does *not* contain any
newlines will not be affected by this change, so it is still
possible to use inline noweb references.
Thanks to Sébastien Vauban for this idea.
