165 KiB
org-babel — facilitating communication between programming languages and people
- Introduction
- Getting started
- Tasks
[41/64]
- org-bable-tangle: no default extension if one already exists
- source-name visible in LaTeX and html exports
- STARTED new results types (org, html, latex)
- PROPOSED raise elisp error when source-blocks return errors
- PROPOSED allow `anonymous' function block with function call args?
- PROPOSED allow :result as synonym for :results?
- PROPOSED allow 'output mode to return stdout as value?
- PROPOSED optional timestamp for output
- make tangle files read-only?
- support for working with
*Org Edit Src Example*
buffers[4/6]
- resolve references to other org buffers/files
- resolve references to other non-org files
- figure out how to handle errors during evaluation
- STARTED figure out how to handle graphic output
- Finalise behaviour regarding vector/scalar output
- STARTED share org-babel
[1/6]
- command line execution
- inline source code blocks
[3/5]
- LoB: re-implement plotting and analysis functions from org-R
- PROPOSED Creating presentations
- PROPOSED conversion between org-babel and noweb (e.g. .Rnw) format
- PROPOSED support for passing paths to files between source blocks
- DEFERRED Support rownames and other org babel table features?
- DEFERRED use textConnection to pass tsv to R?
- DEFERRED Rework Interaction with Running Processes
[2/5]
- DEFERRED improve the source-block snippet
- REJECTED re-implement R evaluation using ess-command or ess-execute
- take default values for header args from properties
- new reference syntax inside source code blocks
- re-work tangling system
\C-c \C-o
to open results of source block- Stop spaces causing vector output
- add
:tangle
family of header arguments - extensible library of callable source blocks
- Column names in R input/output
- use example block for large amounts of stdout output?
- exclusive
exports
params - LoB: allow output in buffer
- allow default header arguments by language
- singe-function tangling and loading elisp from literate org-mode file
[3/3]
- add a function to jump to a source-block by name
- add
:none
session argument (for purely functional execution)[4/4]
- fully purge org-babel-R of direct comint interaction
- Create objects in top level (global) environment
[5/5]
- Remove protective commas from # comments before evaluating
- pass multiple reference arguments into R
- ensure that table ranges work
- global variable indicating default to vector output
- name named results if source block is named
- (simple caching) check for named results before source blocks
- set
:results silent
when eval with prefix argument - results-type header (vector/file)
[3/3]
- results name
- org-babel tests org-babel
[1/1]
- make C-c C-c work anywhere within source code block?
- integration with org tables
- source blocks as functions
- folding of code blocks?
[2/2]
- selective export of text, code, figures
- a header argument specifying silent evaluation (no output)
- assign variables from tables in R
- insert 2-D R results as tables
- allow variable initialization from source blocks
- Add languages
[10/14]
- Bugs
[37/41]
- Python session evaluation bug
- require users to explicitly turn on each language
- problem with newlines in output when :results value
- prompt characters appearing in output with R
- o-b-execute-subtree overwrites heading when subtree is folded
- non-orgtbl formatted lists
- PROPOSED allow un-named arguments
- PROPOSED external shell execution can't isolate return values
- adding blank line when source-block produces no output
- Allow source blocks to be recognised when #+ are not first characters on the line
- are the org-babel-trim s necessary?
- LoB is not populated on startup
- use new merge function here?
- creeping blank lines
- #+srcname arg parsing bug
- Fix nested evaluation and default args
- allow srcname to omit function call parentheses
- avoid stripping whitespace from output when :results output
- DEFERRED weird escaped characters in shell prompt break shell evaluation
- function calls in #+srcname: refs
- LoB: with output to buffer, not working in buffers other than library-of-babel.org
- cursor movement when evaluating source blocks
- LoB: calls fail if reference has single character name
- make :results replace the default?
- ruby evaluation not working under ubuntu emacs 23
- test failing forcing vector results with
test-forced-vector-results
ruby code block - defunct R sessions
- ruby fails on first call to non-default session
- when reading results from
#+resname
line - R-code broke on "org-babel" rename
- error on trivial R results
- ruby new variable creation (multi-line ruby blocks)
- R code execution seems to choke on certain inputs
- org bug/request: prevent certain org behaviour within code blocks
- with :results replace, non-table output doesn't replace table output
- extra quotes for nested string
- simple ruby arrays not working
- space trailing language name
- Args out of range error
- ruby arrays not recognized as such
- REJECTED elisp reference fails for literal number
- Tests
- Sandbox
- Buffer Dictionary
<div id="subtitle"> <p>executable source code blocks in org-mode</p> </div> <div id="logo"> <p> <img src="images/tower-of-babel.png" alt="images/tower-of-babel.png" > <div id="attr">from <a href="http://www.flickr.com/photos/23379658@N05" title=""><b>Martijn Streefkerk</b></a></div> </p> </div>
Org-Babel makes source-code blocks in Org-Mode executable and allows data to pass seamlessly between different programming languages, Org-Mode constructs (tables, file links, example text) and interactive comint buffers.
In this document:
- The Introduction
- provides a brief overview of the design and use of Org-Babel including tutorials and examples.
- In Getting Started
- find instructions for installing org-babel into your emacs configuration.
- The Tasks
- section contains current and past tasks roughly ordered by TODO state, then importance or date-completed. This would be a good place to suggest ideas for development.
- The Bugs
- section contains bug reports.
- The Tests
- section consists of a large table which can be evaluated to run Org-Babel's functional test suite. This provides a good overview of the current functionality with pointers to example source blocks.
- The Sandbox
- demonstrates much of the early/basic functionality through commented source-code blocks.
Also see the Library of Babel, an extensible collection of ready-made and easily-shortcut-callable source-code blocks for handling common tasks.
Introduction
Org-Babel enables communication between programming languages and between people.
Org-Babel provides:
- communication between programs
- Data passes seamlessly between different programming languages, Org-Mode constructs (tables, file links, example text) and interactive comint buffers.
- communication between people
- Data and calculations are embedded in the same document as notes explanations and reports.
communication between programs
Org-Mode supports embedded blocks of source code (in any language) inside of Org documents. Org-Babel allows these blocks of code to be executed from within Org-Mode with natural handling of their inputs and outputs.
simple execution
with both scalar, file, and table output
reading information from tables
reading information from other source blocks (disk usage in your home directory)
This will work for Linux and Mac users, not so sure about shell commands for windows users.
To run place the cursor on the #+begin_src
line of the source block
labeled directory-pie and press \C-c\C-c
.
cd ~ && du -sc * |grep -v total
64 | "Desktop" |
11882808 | "Documents" |
8210024 | "Downloads" |
879800 | "Library" |
57344 | "Movies" |
7590248 | "Music" |
5307664 | "Pictures" |
0 | "Public" |
152 | "Sites" |
8 | "System" |
56 | "bin" |
3274848 | "mail" |
5282032 | "src" |
1264 | "tools" |
pie(dirs[,1], labels = dirs[,2])
operations in/on tables
student | grade | letter |
---|---|---|
1 | 99 | A |
2 | 59 | F |
3 | 75 | C |
4 | 15 | F |
5 | 7 | F |
6 | 13 | F |
case score
when 0..59: "F"
when 60..69: "D"
when 70..79: "C"
when 80..89: "B"
when 90..100: "A"
else "Invalid Score"
end
rand(100)
hist(grades$grade)
communication between people
Quick overview of Org-Mode's exportation abilities, with links to the online Org-Mode documentation, a focus on source-code blocks, and the exportation options provided by Org-Babel.
Interactive tutorial
This would demonstrate applicability to Reproducible Research, and Literate Programming.
Tests embedded in documentation
org-babels own functional tests are contained in a large org-mode table, allowing the test suite to be run be evaluation of the table and the results to be collected in the same table.
Emacs initialization files stored in Org-Mode buffers
Using `org-babel-tangle' it is possible to embed your Emacs initialization into org-mode files. This allows for folding, note-taking, todo's etc… embedded with the source-code of your Emacs initialization, and through org-mode's publishing features aids in sharing your customizations with others.
It may be worthwhile to create a fork of Phil Hagelberg's emacs-starter-kit which uses literate org-mode files for all of the actual elisp customization. These org-mode files could then be exported to html and used to populate the repositories wiki on github.
features
code evaluation (comint buffer sessions and external processes)
There are two main ways to evaluate source blocks with org-babel.
- external
- By default (if the
:session
header argument is not present) all source code blocks are evaluated in external processes. In these cases an external process is used to evaluate the source-code blocks. - session
- Session based evaluation uses persistent sessions in
comint buffers. Sessions can be used across multiple
source blocks setting and accessing variables in the
global environment.
Evaluating source blocks in sessions also allows for
interaction with the code. To jump to the session of a
source block use the `org-babel-pop-to-session' command
or press
M-[down]
while inside of a source code block. When called with a prefix argument `org-babel-pop-to-session' will evaluate all header arguments before jumping to the source-code block.
results (values and outputs)
Either the value or the output of source code blocks can be collected after evaluation.
- value
- The default way to collect results from a source-code block is to return the value of the last statement in the block. This can be thought of as the return value of the block. In this case any printed output of the block is ignored. This can be though of a similar to a "functional" value of evaluation.
- output
- Another way of generating results from a source-code block is to collect the output generated by the execution of the block. In this case all printed output is collected throughout the execution of the block. This can be thought of as similar to a "script" style of evaluation.
Getting started
Add the following lines to your .emacs, replacing the path as appropriate. A good place to check that things are up and running would then be the sandbox.
(add-to-list 'load-path "/path/to/org-babel/lisp")
(require 'org-babel-init)
Tasks [41/64]
TODO org-bable-tangle: no default extension if one already exists
TODO source-name visible in LaTeX and html exports
Maybe this should be done in backend specific manners.
The listings package may provide for naming a source-code block…
STARTED new results types (org, html, latex)
Thanks to Tom Short for this recommendation.
- raw or org
- in which case the results are implemented raw, unquoted into the org-mode file. This would also handle links as source block output.
- html
- the results are inserted inside of a #+BEGIN_HTML block
- latex
- the results are inserted inside of a #+BEGIN_LATEX block
It might look like:
#+begin_src R :session *R* :results org cat("***** This is a table\n") cat("| 1 | 2 | 3 |\n") cat("[[http://google.com][Google it here]]\n" #+end_src #+resname: ***** This is a table | 1 | 2 | 3 |
We actually might want to remove the #+resname
line if the results
type is org-mode, not sure… Either way I don't think there is a
good way to capture/remove org type results.
raw
Added a raw
results header argument, which will insert the results
of a source-code block into an org buffer un-escaped. Also, if the
results look like a table, then the table will be aligned.
puts "| root | square |"
puts "|---"
10.times do |n|
puts "| #{n} | #{n*n} |"
end
root | square |
---|---|
0 | 0 |
1 | 1 |
2 | 4 |
3 | 9 |
4 | 16 |
5 | 25 |
6 | 36 |
7 | 49 |
8 | 64 |
9 | 81 |
PROPOSED raise elisp error when source-blocks return errors
Not sure how/if this would work, but it may be desirable.
PROPOSED allow `anonymous' function block with function call args?
My question here is simply whether we're going to allow
# whatever
but with preference given to #+srcname blockname(arg=ref)
PROPOSED allow :result as synonym for :results?
PROPOSED allow 'output mode to return stdout as value?
Maybe we should allow this. In fact, if block x is called with :results output, and it references blocks y and z, then shouldn't the output of x contain a concatenation of the outputs of y and z, together with x's own output? That would raise the question of what happens if y is defined with :results output and z with :results value. I guess z's (possibly vector/tabular) output would be inside a literal example block containing the whole lot.
PROPOSED optional timestamp for output
Add option to place an (inactive) timestamp at the #+resname, to record when that output was generated.
source code block timestamps (optional addition)
[Eric] If we did this would we then want to place a timestamp on the source-code block, so that we would know if the results are current or out of date? This would have the effect of caching the results of calculations and then only re-running if the source-code has changed. For the caching to work we would need to check not only the timestamp on a source-code block, but also the timestamps of any tables or source-code blocks referenced by the original source-code block.
[Dan] I do remember getting frustrated by Sweave always having to re-do everything, so this could be desirable, as long as it's easy to over-ride of course. I'm not sure it should be the default behaviour unless we are very confident that it works well.
maintaining source-code block timestamps
It may make sense to add a hook to `org-edit-special' which could update the source-code blocks timestamp. If the user edits the contents of a source-code block directly I can think of no efficient way of maintaining the timestamp.
TODO make tangle files read-only?
With a file-local variable setting, yea that makes sense. Maybe the header should reference the related org-mode file.
TODO
support for working with *Org Edit Src Example*
buffers [4/6]
STARTED Patch against org source.
I've worked on several related changes to source code edit buffer behaviour in the org core. My current patch (below) does the following. Detailed explanation / working notes are below.
- C-x s offers to save edit buffers
- C-x C-c offers to save edit buffers
- C-x k warns that you're killing an edit buffer
- If you do kill an edit buffer, the overlay in the parent buffer is removed
- Edit buffers are named Org Src <orgbuf>[<lang>], where <orgbuf> is the name of the org-mode buffer containing this source code block, and lang is the language major mode. The latter might be unnecessary?
diff --git a/lisp/org-src.el b/lisp/org-src.el index 2083c77..2be21e6 100644 --- a/lisp/org-src.el +++ b/lisp/org-src.el @@ -113,7 +113,7 @@ but which mess up the display of a snippet in Org exported files.") (defvar org-src-mode-map (make-sparse-keymap)) (define-key org-src-mode-map "\C-c'" 'org-edit-src-exit) -(define-key org-src-mode-map "\C-x\C-s" 'org-edit-src-save) +;; (define-key org-src-mode-map "\C-x\C-s" 'org-edit-src-save) (defvar org-edit-src-force-single-line nil) (defvar org-edit-src-from-org-mode nil) (defvar org-edit-src-picture nil) @@ -168,7 +168,8 @@ the edited version." (if (boundp 'org-edit-src-overlay) (org-delete-overlay org-edit-src-overlay))) (kill-buffer buffer)) - (setq buffer (generate-new-buffer "*Org Edit Src Example*")) + (setq buffer (generate-new-buffer + (concat "*Org Src " (file-name-nondirectory buffer-file-name) "[" lang "]*"))) (setq ovl (org-make-overlay beg end)) (org-overlay-put ovl 'face 'secondary-selection) (org-overlay-put ovl 'edit-buffer buffer) @@ -186,8 +187,7 @@ the edited version." '(display nil invisible nil intangible nil)) (org-do-remove-indentation) (let ((org-inhibit-startup t)) - (funcall lang-f) - (org-src-mode)) + (funcall lang-f)) (set (make-local-variable 'org-edit-src-force-single-line) single) (set (make-local-variable 'org-edit-src-from-org-mode) org-mode-p) (when lfmt @@ -201,6 +201,7 @@ the edited version." (org-set-local 'org-edit-src-end-marker end) (org-set-local 'org-edit-src-overlay ovl) (org-set-local 'org-edit-src-nindent nindent) + (org-src-mode) (and org-edit-src-persistent-message (org-set-local 'header-line-format msg))) (message "%s" msg) @@ -400,12 +401,13 @@ the language, a switch telling of the content should be in a single line." (defun org-edit-src-exit () "Exit special edit and protect problematic lines." (interactive) - (unless (string-match "\\`*Org Edit " (buffer-name (current-buffer))) - (error "This is not an sub-editing buffer, something is wrong...")) + (unless org-edit-src-from-org-mode + (error "This is not a sub-editing buffer, something is wrong...")) (let ((beg org-edit-src-beg-marker) (end org-edit-src-end-marker) (ovl org-edit-src-overlay) (buffer (current-buffer)) + (buffer-file-name nil) (nindent org-edit-src-nindent) code line) (untabify (point-min) (point-max)) @@ -464,6 +466,17 @@ the language, a switch telling of the content should be in a single line." (goto-char (min p (point-max))) (message (or msg "")))) +(defun org-src-mode-configure-buffer () + (setq buffer-offer-save t) + (setq buffer-file-name + (concat (buffer-file-name (marker-buffer org-edit-src-beg-marker)) + "[" (buffer-name) "]")) + (setq write-contents-functions '(org-edit-src-save)) + (org-add-hook 'kill-buffer-hook + '(lambda () (org-delete-overlay org-edit-src-overlay)) nil 'local)) + +(org-add-hook 'org-src-mode-hook 'org-src-mode-configure-buffer) + (provide 'org-src) ;; arch-tag: 6a1fc84f-dec7-47be-a416-64be56bea5d8
Detailed working notes to go with that patch
Recap of current org-src-mode
If you use C-c ' to work on code in a begin_source block, the code buffer is put in minor mode org-src-mode, which features the following two useful key-bindings:
C-x s | org-edit-src-save | save the code in the source code block in the parent org file |
C-c ' | org-edit-src-exit | return to the parent org file with new code |
Furthermore, while the edit buffer is alive, the originating code block is subject to a special overlay which links to the edit buffer when you click on it.
This is all excellent, and I use it daily, but I think there's still a couple of improvements that we should make.
Proposed bug I
C-x k kills the buffer without questions; the overlay remains, but now links to a deleted buffer.
Proposed bug II
C-x C-c kills a modified edit buffer silently, without offering to save your work. I have lost work like that a number of times recently.
Proposed bug III
C-x s does not offer to save a modified edit buffer
Notes on solution
A good start seems to be to use org-src-mode-hook to add org-edit-src-save to the write-contents-functions list. This means that when it comes to saving, org-edit-src-save will be called and no subsequent attempt will be made to save the buffer in the normal way. (This should obviate the remapping of C-x C-s to org-edit-src-save in org-src.el)
We also want to set this to t.
- C-x s still does not offer to save the edit buffer. That's
because buffer-file-name is nil.
- C-x C-c does ask us whether we want to save the
edit buffer. However, since buffer-file-name is nil it asks us for a file name. The check in org-edit-src-exit throws an error unless the buffer is named '* Org Edit '…
- C-x k kills the buffer silently, leaving a broken overlay
link. If buffer-file-name were set, it would have warned that the buffer was modified.
So, that all suggests that we need to set buffer-file-name, even though we don't really want to associate this buffer with a file in the normal way. As for the file name, my current suggestion is parent-org-filename[edit-buffer-name].
[I had to move the (org-src-mode) call to the end of org-edit-src-code to make sure that the required variables were defined when the hook was called.]
- C-x s does offer to save the edit buffer, but in saving
produces a warning that the edit buffer is modified.
- C-x k now gives a warning that the edit buffer is modified
(even if it's not).
- C-x C-c is working as desired, except that again we get
warnings that the edit buffer is modified, once when we save, and again just before exiting emacs.
- And C-c ' now issues a warning that the edit buffer is
modified when we leave it, which we don't want.
I've made buffer-file-name nil inside the let binding in org-edit-src-exit.
- C-x s behaves as desired, except that as was already the case,
the edit buffer is always considered modified, and so repeated invocations keep saving it.
- As was already the case, C-x k always gives a warning that the
edit buffer has been modified.
- C-x C-c is as desired (offers to save the edit buffer) except
that it warns of the modified buffer just before exiting.
- C-c ' is as it should be (silent)
Conclusion
We've got the desired behaviour, at the cost of being forced to assign a buffer-file-name to the edit buffer. The consequence is that the edit buffer is considered to always be modified, since a file of that name is never actually written to (doesn't even exist). I couldn't see a way to trick emacs into believing that the buffer was unmodified since last save. But in any case, I think there's an argument that these modifications warnings are a good thing, because one should not leave active edit buffers around: you should always have exited with C-c ' first.
DONE name edit buffer according to #+srcname (and language?)
See above patch agains org.
DONE
optionally evaluate header references when we switch to *Org Edit Src*
buffer
That seems to imply that the header references need to be evaluated and transformed into the target language object when we hit C-c ' to enter the Org Edit Src buffer [DED]
Good point, I heartily agree that this should be supported [Eric]
(or at least before the first time we attempt to evaluate code in that buffer – I suppose there might be an argument for lazy evaluation, in case someone hits C-c ' but is "just looking" and not actually evaluating anything.) Of course if evaluating the reference is computationally intensive then the user might have to wait before they get the Org Edit Src buffer. [DED]
I fear that it may be hard to anticipate when the references will be needed, some major-modes do on-the-fly evaluation while the buffer is being edited. I think that we should either do this before the buffer is opened or not at all, specifically I think we should resolve references if the user calls C-c ' with a prefix argument. Does that sound reasonable? [Eric]
Yes [Dan]
[Dan] So now that we have org-src-mode and org-src-mode-hook, I guess org-babel should do this by using the hook to make sure that, when C-c C-' is issued on a source block, any references are resolved and assignments are made in the appropriate session.
1 | 2 |
3 | 4 |
table.size.times.do |n|
puts n
end
TODO set buffer-local-process variables appropriately [DED]
I think something like this would be great. You've probably already thought of this, but just to note it down: it would be really nice if org-babel's notion of a buffer's 'session/process' played nicely with ESS's notion of the buffer's session/process. ESS keeps the current process name for a buffer in a buffer-local variable ess-local-process-name. So one thing we will probably want to do is make sure that the Org Edit Src Example buffer sets that variable appropriately. [DED]
I had not thought of that, but I agree whole heartedly. [Eric]
Once this is done every variable should be able to dump regions into their inferior-process buffer using major-mode functions.
REJECTED send code to inferior process
Another thought on this topic: I think we will want users to send chunks of code to the interpreter from within the Org Edit Src buffer, and I think that's what you have in mind already. In ESS that is done using the ess-eval-* functions. [DED]
I think we can leave this up to the major-mode in the source code buffer, as almost every source-code major mode will have functions for doing things like sending regions to the inferior process. If anything we might need to set the value of the buffer local inferior process variable. [Eric]
DONE
some possible requests/proposed changes for Carsten [4/4]
While I remember, some possible requests/proposed changes for Carsten come to mind in that regard:
DONE Remap C-x C-s to save the source to the org buffer?
I've done this personally and I find it essential. I'm using
(defun org-edit-src-save ()
"Update the parent org buffer with the edited source code, save
the parent org-buffer, and return to the source code edit
buffer."
(interactive)
(let ((p (point)))
(org-edit-src-exit)
(save-buffer)
(org-edit-src-code)
(goto-char p)))
(define-key org-exit-edit-mode-map "\C-x\C-s" 'org-edit-src-save)
which seems to work.
I think this is great, but I think it should be implemented in the org-mode core
DEFERRED Rename buffer and minor mode?
Something shorter than Org Edit Src Example for the buffer name. org-babel is bringing org's source code interaction to a level of maturity where the 'example' is no longer appropriate. And if further keybindings are going to be added to the minor mode then maybe org-edit-src-mode is a better name than org-exit-edit-mode.
Maybe we should name the buffer with a combination of the source code and the session. I think that makes sense.
[ES] Are you also suggesting a new org-edit-src minor mode? [DED] org-exit-edit-mode is a minor mode that already exists:
Minor mode installing a single key binding, "C-c '" to exit special edit.
org-edit-src-save now has a binding in that mode, so I guess all I'm saying at this stage is that it's a bit of a misnomer. But perhaps we will also have more functionality to add to that minor mode, making it even more of a misnomer. Perhaps something like org-src-mode would be better.
DONE Changed minor mode name and added hooks
DONE a hook called when the src edit buffer is created
This should be implemented in the org-mode core
TODO resolve references to other org buffers/files
This would allow source blocks to call upon tables, source-blocks, and results in other org buffers/files.
See…
TODO resolve references to other non-org files
- tabular data in .csv, .tsv etc format
- files of interpreted code: anything stopping us giving such files similar status to a source code block?
- Would be nice to allow org and non-org files to be remote
TODO figure out how to handle errors during evaluation
I expect it will be hard to do this properly, but ultimately it would be nice to be able to specify somewhere to receive STDERR, and to be warned if it is non-empty.
Probably simpler in non-session evaluation than session? At least the mechanism will be different I guess.
R has a try function, with error handling, along the lines of python. I bet ruby does too. Maybe more of an issue for functional style; in my proposed scripting style the error just gets dumped to the org buffer and the user is thus alerted.
STARTED figure out how to handle graphic output
This is listed under graphical output in out objectives.
This should take advantage of the :results file
option, and
languages which almost always produce graphical output should set
:results file
to true by default (this is currently done for the
gnuplot and ditaa languages). That would handle placing these results
in the buffer. Then if there is a combination of silent
and file
:results
headers we could drop the results to a temp buffer and pop
open that buffer…
Display of file results is addressed in the open-results-task.
TODO R graphics to screen means session evaluation
If R graphical output is going to screen then evaluation must be in a session, otherwise the graphics will disappear as soon as the R process dies.
Adding to a discussion started in email
I'm not deeply wedded to these ideas, just noting them down. I'm probably just thinking of R and haven't really thought about how this fits with the other graphics-generating languages. Dan: > I used the approach below to get graphical file output > today, which is one idea at least. Maybe it could be linked up with > your :results file variable. (Or do we need a :results image for R?) > Eric: I don't think we need a special image results variable, but I may be missing what the code below accomplishes. Would the task I added about adding org-open-at-point functionality to source code blocks take care of this need?
Dan: I'm not sure. I think the ability for a script to generate both text and graphical output might be a natural expectation, at least for R users.
> > Dan > > #+srcname: cohort-scatter-plots-2d(org_babel_graphical_output_file="cohort-scatter-plots-2d.png") > #+begin_src R > if(exists("org_babel_output_file")) > png(filename=org_babel_graphical_output_file, width=1000, height=1000) > ## plotting code in here > if(exists("org_babel_graphical_output_file")) dev.off() > #+end_src
Dan: Yes, the results :file option is nice for dealing with graphical output, and that could well be enough. Something based on the scheme above would have a couple of points in its favour:
- It's easy to switch between output going to on-screen graphics and output going to file: Output will go to screen unless a string variable with a standard name (e.g. ""org_babel_graphical_output_file"") exists in which case it will go to the file indicated by the value of that variable.
- The block can return a result / script output, as well as produce graphical output.
In interactive use we might want to allow the user to choose between screen and file output. In non-interactive use such as export, it would be file output (subject to the :exports directives).
TODO Finalise behaviour regarding vector/scalar output
DONE Stop spaces causing vector output
This simple example of multilingual chaining produces vector output if there are spaces in the message and scalar otherwise.
[Not any more]
paste(msg, "und R", sep=" ")
org-babel speaks elisp y python und R
msg + " y python"
(concat msg " elisp")
STARTED share org-babel [1/6]
how should we share org-babel?
DONE post to org-mode
TODO post to ess mailing list
TODO create a org-babel page on worg
TODO create a short screencast demonstrating org-babel in action
PROPOSED a peer-reviewed publication?
The following notes are biased towards statistics-oriented journals because ESS and Sweave are written by people associated with / in statistics departments. But I am sure there are suitable journals out there for an article on using org mode for reproducible research (and literate programming etc).
Clearly, we would invite Carsten to be involved with this.
ESS is described in a peer-reviewed journal article: Emacs Speaks Statistics: A Multiplatform, Multipackage Development Environment for Statistical Analysis [Abstract] Journal of Computational & Graphical Statistics 13(1), 247-261 Rossini, A.J, Heiberger, R.M., Sparapani, R.A., Maechler, M., Hornik, K. (2004) Journal of Computational and Graphical Statistics
Also Journal of Statistical Software Established in 1996, the Journal of Statistical Software publishes articles, book reviews, code snippets, and software reviews. The contents are freely available online. For both articles and code snippets, the source code is published along with the paper.
Sweave has a paper:
Friedrich Leisch and Anthony J. Rossini. Reproducible statistical research. Chance, 16(2):46-50, 2003. [ bib ]
also
Friedrich Leisch. Sweave: Dynamic generation of statistical reports using literate data analysis. In Wolfgang Härdle and Bernd Rönz, editors, Compstat 2002 - Proceedings in Computational Statistics, pages 575-580. Physica Verlag, Heidelberg, 2002. ISBN 3-7908-1517-9.
also
We could also look at the Journals publishing these Reproducible Research articles.
PROPOSED an article in The R Journal
This looks good. It seems that their main topic to software tools for use by R programmers, and Org-babel is certainly that.
examples
we need to think up some good examples
interactive tutorials
This could be a place to use org-babel assertions.
for example the first step of a tutorial could assert that the version of the software-package (or whatever) is equal to some value, then source-code blocks could be used with confidence (and executed directly from) the rest of the tutorial.
answering a text-book question w/code example
org-babel is an ideal environment enabling both the development and demonstrationg of the code snippets required as answers to many text-book questions.
something using tables
maybe something along the lines of calculations from collected grades
file sizes
Maybe something like the following which outputs sizes of directories
under the home directory, and then instead of the trivial emacs-lisp
block we could use an R block to create a nice pie chart of the
results.
du -sc ~/*
(mapcar #'car sizes)
TODO command line execution
Allow source code blocks to be called form the command line. This
will be easy using the sbe
function in org-babel-table.el.
This will rely upon resolve references to other buffers.
TODO
inline source code blocks [3/5]
Like the \R{ code }
blocks
not sure what the format should be, maybe just something simple
like src_lang[]{}
where lang is the name of the source code
language to be evaluated, []
is optional and contains any header
arguments and {}
contains the code.
(see the-sandbox)
DONE evaluation with \C-c\C-c
Putting aside the header argument issue for now we can just run these with the following default header arguments
-
:results
- silent
-
:exports
- results
DONE inline exportation
Need to add an interblock hook (or some such) through org-exp-blocks
DONE header arguments
We should make it possible to use header arguments.
TODO fontification
we should color these blocks differently
TODO refine html exportation
should use a span class, and should show original source in tool-tip
TODO LoB: re-implement plotting and analysis functions from org-R
I'll do this soon, now that we things are a bit more settled and we have column names in R.
PROPOSED Creating presentations
The recent thread containing posts by Nick Dokos and Sebastian Vaubán on exporting to beamer looked very interesting, but I haven't had time to try it out yet. I would really like it if, eventually, we can generate a presentation (with graphics generated by code blocks) from the same org file that contains all the notes and code etc. I just wanted that to be on record in this document; I don't have anything more profound to say about it at the moment, and I'm not sure to what extent it is an org-babel issue.
PROPOSED conversion between org-babel and noweb (e.g. .Rnw) format
I haven't thought about this properly. Just noting it down. What Sweave uses is called "R noweb" (.Rnw).
I found a good description of noweb in the following article (see the pdf).
I think there are two parts to noweb, the construction of documentation and the extraction of source-code (with notangle).
documentation: org-mode handles all of our documentation needs in a manner that I believe is superior to noweb.
source extraction At this point I don't see anyone writing large applications with 100% of the source code contained in org-babel files, rather I see org-babel files containing things like
- notes with active code chunks
- interactive tutorials
- requirements documents with code running test suites
- and of course experimental reports with the code to run the experiment, and perform analysis
Basically I think the scope of the programs written in org-babel (at least initially) will be small enough that it wont require the addition of a tangle type program to extract all of the source code into a running application.
On the other hand, since we already have named blocks of source code which reference other blocks on which they rely, this shouldn't be too hard to implement either on our own, or possibly relying on something like noweb/notangle.
PROPOSED support for passing paths to files between source blocks
Maybe this should be it's own result type (in addition to scalars and vectors). The reason being that some source-code blocks (for example ditaa or anything that results in the creation of a file) may want to pass a file path back to org-mode which could then be inserted into the org-mode buffer as a link to the file…
This would allow for display of images upon export providing
functionality similar to org-exp-blocks
only in a more general
manner.
DEFERRED Support rownames and other org babel table features?
The full org table features are detailed in the manual here.
rownames
Perhaps add a :rownames header arg. This would be an integer (usually 1) which would have the effect of post-processing all the variables created in the R session in the following way: if the integer is j, set the row names to the contents of column j and delete column j. Perhaps it is artificial to allow this integer to take any value other than 1. The default would be nil which would mean no such behaviour.
Actually I don't know about that. If multiple variables are passed in, it's not appropriate to alter them all in the same way. The rownames specification would normally refer to just one of the variables. For now maybe just say this has to be done in R. E.g.
collection | size | exclude | include | exclude2 | include2 |
---|---|---|---|---|---|
58C | 2936 | 8 | 2928 | 256 | 2680 |
MS | 5852 | 771 | 5081 | 771 | 5081 |
NBS | 2929 | 64 | 2865 | 402 | 2527 |
POBI | 2717 | 1 | 2716 | 1 | 2716 |
58C+MS+NBS+POBI | 13590 | 13004 |
rownames(size) <- size[,1]
size <- size[,-1]
Old notes
[I don't think it's as problematic as this makes out] This is non-trivial, but may be worth doing, in particular to develop a nice framework for sending data to/from R.
Notes
In R, indexing vector elements, and rows and columns, using strings rather than integers is an important part of the language.
- elements of a vector may have names
- matrices and data.frames may have "column names" and "row names" which can be used for indexing
- In a data frame, row names must be unique
Examples
> # a named vector > vec <- c(a=1, b=2) > vec["b"] b 2 > mat <- matrix(1:4, nrow=2, ncol=2, dimnames=list(c("r1","r2"), c("c1","c2"))) > mat c1 c2 r1 1 3 r2 2 4 > # The names are separate from the data: they do not interfere with operations on the data > mat * 3 c1 c2 r1 3 9 r2 6 12 > mat["r1","c2"] [1] 3 > df <- data.frame(var1=1:26, var2=26:1, row.names=letters) > df$var2 [1] 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 > df["g",] var1 var2 g 7 20
So it's tempting to try to provide support for this in org-babel. For example
- allow R to refer to columns of a :var reference by their names
- When appropriate, results from R appear in the org buffer with "named columns (and rows)" However none (?) of the other languages we are currently supporting really have a native matrix type, let alone "column names" or "row names". Names are used in e.g. python and perl to refer to entries in dicts / hashes. It currently seems to me that support for this in org-babel would require setting rules about when org tables are considered to have named columns/fields, and ensuring that (a) languages with a notion of named columns/fields use them appropriately and (b) languages with no such notion do not treat then as data.
- Org allows something that looks like column names to be separated by a hline
- Org also allows a row to function as column names when special markers are placed in the first column. An hline is unnecessary (indeed hlines are purely cosmetic in org [correct?]
- Org does not have a notion of "row names" [correct?] The full org table functionality exeplified here has features that we would not support in e.g. R (like names for the row below).
Initial statement: allow tables with hline to be passed as args into R
This doesn't seem to work at the moment (example below). It would also be nice to have a natural way for the column names of the org table to become the column names of the R data frame, and to have the option to specify that the first column is to be used as row names in R (these must be unique). But this might require a bit of thinking about.
col1 | col2 | col3 |
---|---|---|
1 | 2 | 3 |
4 | schulte | 6 |
1 | 2 | 3 |
4 | schulte | 6 |
tabel
"col1" | "col2" | "col3" |
---|---|---|
1 | 2 | 3 |
4 | "schulte" | 6 |
Another example is in the grades example.
DEFERRED use textConnection to pass tsv to R?
When passing args from the org buffer to R, the following route is used: arg in buffer -> elisp -> tsv on file -> data frame in R. I think it would be possible to avoid having to write to file by constructing an R expression in org-babel-R-assign-elisp, something like this
(org-babel-R-input-command
(format "%s <- read.table(textConnection(\"%s\"), sep=\"\\t\", as.is=TRUE)"
name (orgtbl-to-tsv value '(:sep "\t" :fmt org-babel-R-quote-tsv-field))))
I haven't tried to implement this yet as it's basically just fiddling with something that works. The only reason for it I can think of would be efficiency and I haven't tested that.
This Didn't work after an initial test. I still think this is a good idea (I also think we should try to do something similar when writing out results frmo R to elisp) however as it wouldn't result in any functional changes I'm bumping it down to deferred for now. [Eric]
for quick tests
1 | 2 | 3 |
mean(mean(vec))
2
2
DEFERRED Rework Interaction with Running Processes [2/5]
DONE robust to errors interrupting execution
sleep(10)
:patton_is_an_grumpy
DEFERRED use C-g
keyboard-quit to push processing into the background
This may be possible using the `run-with-timer' command.
I have no idea how this could work…
sleep(10)
:patton_is_an_grumpy
TODO ability to select which of multiple sessions is being used
Increasingly it is looking like we're going to want to run all source code blocks in comint buffer (sessions). Which will have the benefits of
- allowing background execution
-
maintaining state between source-blocks
- allowing inline blocks w/o header arguments
R sessions
(like ess-switch-process in .R buffers)
Maybe this could be packaged into a header argument, something
like :R_session
which could accept either the name of the
session to use, or the string prompt
, in which case we could use
the ess-switch-process
command to select a new process.
TODO evaluation of shell code as background process?
After C-c C-c on an R code block, the process may appear to block, but C-g can be used to reclaim control of the .org buffer, without interrupting the R evalution. However I believe this is not true of bash/sh evaluation. [Haven't tried other languages] Perhaps a solution is just to background the individual shell commands.
The other languages (aside from emacs lisp) are run through the shell, so if we find a shell solution it should work for them as well.
Adding an ampersand seems to be a supported way to run commands in the background (see external-commands). Although a more extensible solution may involve the use of the call-process-region function.
Going to try this out in a new file org-babel-proc.el. This should contain functions for asynchronously running generic shell commands in the background, and then returning their input.
partial update of org-mode buffer
The sleekest solution to this may be using a comint buffer, and then defining a filter function which would incrementally interpret the results as they are returned, including insertion into the org-mode buffer. This may actually cause more problems than it is worth, what with the complexities of identifying the types of incrementally returned results, and the need for maintenance of a process marker in the org buffer.
'working' spinner
It may be nice and not too difficult to place a spinner on/near the evaluating source code block
TODO conversion of output from interactive shell, R (and python) sessions to org-babel buffers
[DED] This would be a nice feature I think. Although an org-babel purist would say that it's working the wrong way round… After some interactive work in a R buffer, you save the buffer, maybe edit out some lines, and then convert it to org-babel format for posterity. Same for a shell session either in a shell buffer, or pasted from another terminal emulator. And python of course.
DEFERRED improve the source-block snippet
any real improvement seems somewhat beyond the ability of yasnippet for now.
file:~/src/emacs-starter-kit/src/snippets/text-mode/rst-mode/chap::name Chapter title
,#name : Chapter title ,# -- ${1:Chapter} ${1:$(make-string (string-width text) ?\=)} $0
waiting for guidance from those more familiar with yasnippets
REJECTED re-implement R evaluation using ess-command or ess-execute
I don't have any complaints with the current R evaluation code or behaviour, but I think it would be good to use the ESS functions from a political point of view. Plus of course it has the normal benefits of an API (insulates us from any underlying changes etc). [DED]
I'll look into this. I believe that I looked at and rejected these functions initially but now I can't remember why. I agree with your overall point about using API's where available. I will take a look back at these and either switch to using the ess commands, or at least articulate under this TODO the reasons for using our custom R-interaction commands. [Eric]
ess-execute
Lets just replace org-babel-R-input-command
with ess-execute
.
I tried this, and although it works in some situations, I find that
ess-command
will often just hang indefinitely without returning
results. Also ess-execute
will occasionally hang, and pops up
the buffer containing the results of the command's execution, which
is undesirable. For now these functions can not be used. Maybe
someone more familiar with the ESS code can recommend proper usage
of ess-command
or some other lower-level function which could be
used in place of org-babel-R-input-command.
ess functions
(ess-command COM &optional BUF SLEEP NO-PROMPT-CHECK)
Send the ESS process command COM and delete the output from the ESS process buffer. If an optional second argument BUF exists save the output in that buffer. BUF is erased before use. COM should have a terminating newline. Guarantees that the value of .Last.value will be preserved. When optional third arg SLEEP is non-nil, `(sleep-for (* a SLEEP))' will be used in a few places where `a' is proportional to `ess-cmd-delay'.
(ess-execute COMMAND &optional INVERT BUFF MESSAGE)
Send a command to the ESS process. A newline is automatically added to COMMAND. Prefix arg (or second arg INVERT) means invert the meaning of `ess-execute-in-process-buffer'. If INVERT is 'buffer, output is forced to go to the process buffer. If the output is going to a buffer, name it BUFF. This buffer is erased before use. Optional fourth arg MESSAGE is text to print at the top of the buffer (defaults to the command if BUFF is not given.)
out current setup
- The body of the R source code block is wrapped in a function
- The function is called inside of a
write.table
function call writing the results to a table - The table is read using
org-table-import
DONE take default values for header args from properties
Use file-wide and subtree wide properties to set default values for header args.
[DED] One thing I'm finding when working with R is that an org file may contain many source blocks, but that I just want to evaluate a subset of them. Typically this is in order to take up where I left off: I need to recreate a bunch of variables in the session environment. I'm thinking maybe we want to use a tag-based mechanism similar to :export: and :noexport: to control evaluation on a per-subtree basis.
test-header with properties
Ahh… as is so often the case, just had to wrap `org-babel-params-from-properties' in a `save-match-data' form.
1 | 2 | 3 | 4 |
(+ (length my-lis) def)
DONE new reference syntax inside source code blocks
This is from an email discussion on the org-mode mailing list with Sébastien. The goal here is to mimic the source-block reference style of Noweb. Upon export and/or tangle these references could be replaced with the actual body of the referenced source-code block.
See the following for an example.
puts "---------------------------header---------------------------"
puts "---------------------------footer---------------------------"
# <<ems-ruby-print-header>>
puts " Ruby "
# <<ems-ruby-print-footer>>
Upon export the previous source-code block would result in a file
being generated at ruby-noweb.rb
with the following contents
puts "---------------------------header---------------------------" puts " Ruby " puts "---------------------------footer---------------------------"
the body of a source-code block with all <<src-name>>
references
expanded can now be returned by `org-babel-expand-noweb-references'.
This function is now called by default on all source-code blocks on
export.
DONE re-work tangling system
Sometimes when tangling a file (e.g. when extracting elisp from a org-mode file) we want to get nearly every source-code block.
Sometimes we want to only extract those source-code blocks which reference a indicate that they should be extracted (e.g. traditional literate programming along the Noweb model)
I'm not sure how we can devise a single simple tangling system that naturally fits both of these use cases.
new setup
the tangle
header argument will default to no
meaning source-code
blocks will not be exported by default. In order for a source-code
block to be tangled it needs to have an output file specified. This
can happen in two ways…
- a file-wide default output file can be passed to `org-babel-tangle' which will then be used for all blocks
- if the value of the
tangle
header argument is anything other thanno
oryes
then it is used as the file name
(org-babel-load-file "test-tangle.org")
(if (string= test-tangle-advert "use org-babel-tangle for all your emacs initialization files!!")
"succeed"
"fail")
succeed
DONE
\C-c \C-o
to open results of source block
by adding a defadvice
to org-open-at-point
we can use the common
\C-c \C-o
keybinding to open the results of a source-code block.
This would be especially useful for source-code blocks which generate
graphical results and insert a file link as the results in the
org-mode buffer. (see TODO figure out how to handle graphic output).
This could also act reasonably with other results types…
- file
- use org-open-at-point to open the file
- scalar
- open results unquoted in a new buffer
- tabular
- export the table to a new buffer and open that buffer
when called with a prefix argument the block is re-run
+---------+
| cBLU |
| |
| +----+
| |cPNK|
| | |
+----+----+
'((1 2) (3 4))
1 | 2 |
3 | 4 |
8.times do |n|
puts "row #{n}"
end
row 0 row 1 row 2 row 3 row 4 row 5 row 6 row 7
DONE Stop spaces causing vector output
This simple example of multilingual chaining produces vector output if there are spaces in the message and scalar otherwise.
[Not any more]
paste(msg, "und R", sep=" ")
org-babel speaks elisp y python und R
msg + " y python"
(concat msg " elisp")
DONE
add :tangle
family of header arguments
values are
- no
- don't include source-code block when tangling
- yes
- do include source-code block when tangling
this is tested in test-tangle.org
DONE extensible library of callable source blocks
Current design
This is covered by the Library of Babel, which will contain ready-made source blocks designed to carry out useful common tasks.
Initial statement [Eric]
Much of the power of org-R seems to be in it's helper functions for the quick graphing of tables. Should we try to re-implement these functions on top of org-babel?
I'm thinking this may be useful both to add features to org-babel-R and also to potentially suggest extensions of the framework. For example one that comes to mind is the ability to treat a source-code block like a function which accepts arguments and returns results. Actually this can be it's own TODO (see source blocks as functions).
Objectives [Dan]
- We want to provide convenient off-the-shelf actions (e.g. plotting data) that make use of our new code evaluation environment but do not require any actual coding.
Initial Design proposal [Dan]
- Input data will be specified using the same mechanism as :var references, thus the input data may come from a table, or another source block, and it is initially available as an elisp data structure.
- We introduce a new #+ line, e.g. #+BABELDO. C-c C-c on that line will apply an action to the referenced data.
- Actions correspond to source blocks: our library of available actions will be a library of org-babel source blocks. Thus the code for executing an action, and the code for dealing with the output of the action will be the same code as for executing source blocks in general
- Optionally, the user can have the relevant source block inserted into the org buffer after the (say) #+BABELDO line. This will allow the user to fine tune the action by modifying the code (especially useful for plots).
-
So maybe a #+BABELDO line will have header args
- :data (a reference to a table or source code block)
- :action (or should that be :srcname?) which will be something like :action pie-chart, referring to a source block which will be executed with the :data referent passed in using a :var arg.
- :showcode or something controlling whether to show the code
Modification to design
I'm implementing this, at least initially, as a new interpreter named 'babel', which has an empty body. 'babel' blocks take a :srcname header arg, and look for the source-code block with that name. They then execute the referenced block, after first appending their own header args on to the target block's header args.
If the target block is in the library of babel (a.o.t. e.g. the current buffer), then the code in the block will refer to the input data with a name dictated by convention (e.g. data (something which is syntactically legal in all languages…). Thus the babel block will use a :var data = whatever header arg to reference the data to be plotted.
DONE Column names in R input/output
This has been implemented: Automatic on input to R; optional in output. Note that this equates column names with the header row in an org table; whereas org actually has a mechanism whereby a row with a '!' in the first field defines column names. I have not attempted to support these org table mechanisms yet. See this DEFERRED todo item.
DONE use example block for large amounts of stdout output?
We're currently `examplizing' with : at the beginning of the line, but should larger amounts of output be in a \#+begin_example…\#+end_example block? What's the cutoff? > 1 line? This would be nice as it would allow folding of lengthy output. Sometimes one will want to see stdout just to check everything looks OK, and then fold it away.
I'm addressing this in branch 'examplizing-output'.
Yea, that makes sense. (either that or allow folding of large
blocks escaped with :
).
Proposed cutoff of 10 lines, we can save this value in a user customizable variable.
DONE add ability to remove such results
DONE
exclusive exports
params
:this_is_a_test
:this_is_a_test
DONE LoB: allow output in buffer
DONE allow default header arguments by language
DONE
singe-function tangling and loading elisp from literate org-mode file [3/3]
This function should tangle the org-mode file for elisp, and then call `load-file' on the resulting tangled file.
(setq test-tangle-advert nil)
(setq test-tangle-loading nil)
(setq results (list :before test-tangle-loading test-tangle-advert))
(org-babel-load-file "test-tangle.org")
(setq results (list (list :after test-tangle-loading test-tangle-advert) results))
(delete-file "test-tangle.el")
(reverse results)
:before | nil | nil |
:after | "org-babel tangles" | "use org-babel-tangle for all your emacs initialization files!!" |
DONE add optional language limiter to org-babel-tangle
This should check to see if there is any need to re-export
DONE ensure that org-babel-tangle returns the path to the tangled file(s)
(mapcar #'file-name-nondirectory (org-babel-tangle-file "test-tangle.org" "emacs-lisp"))
"test-tangle.el" |
DONE only tangle the file if it's actually necessary
DONE add a function to jump to a source-block by name
I've had an initial stab at that in org-babel-find-named-block (library-of-babel branch).
At the same time I introduced org-babel-named-src-block-regexp, to match src-blocks with srcname.
This is now working with the command `org-babel-goto-named-source-block', all we need is a good key binding.
DONE
add :none
session argument (for purely functional execution) [4/4]
This would allow source blocks to be run in their own new process
- These blocks could then also be run in the background (since we can detach and just wait for the process to signal that it has terminated)
- We wouldn't be drowning in session buffers after running the tests
- we can re-use much of the session code to run in a more functional mode
While session provide a lot of cool features, like persistent environments, pop-to-session, and hints at exportation for org-babel-tangle, they also have some down sides and I'm thinking that session-based execution maybe shouldn't be the default behavior.
Down-sides to sessions
-
much more complicated than functional evaluation
- maintaining the state of the session has weird issues
- waiting for evaluation to finish
- prompt issues like shell-prompt-escapes-bug
- can't run in background
- litter emacs with session buffers
DONE ruby
puts :eric
puts :schulte
[1, 2, 3]
"eric" |
"schulte" |
DONE python
print 'something'
print 'output'
[1, 2, 3]
1 | 2 | 3 |
DONE sh
echo "first"
echo "second"
"first" |
"second" |
DONE R
a <- 8
b <- 9
a + b
b - a
"[1]" | 17 |
"[1]" | 1 |
DONE fully purge org-babel-R of direct comint interaction
try to remove all code under the ;; functions for evaluation of R code line
DONE
Create objects in top level (global) environment [5/5]
sessions
initial requirement statement [DED]
At the moment, objects created by computations performed in the code block are evaluated in the scope of the code-block-function-body and therefore disappear when the code block is evaluated {unless you employ some extra trickery like assign('name', object, env=globalenv()) }. I think it will be desirable to also allow for a style wherein objects that are created in one code block persist in the R global environment and can be re-used in a separate block.
This is what Sweave does, and while I'm not saying we have to be the same as Sweave, it wouldn't be hard for us to provide the same behaviour in this case; if we don't, we risk undeservedly being written off as an oddity by some.
IOW one aspect of org-babel is that of a sort of functional meta-programming language. This is crazy, in a very good way. Nevertheless, wrt R I think there's going to be a lot of value in providing for a working style in which the objects are stored in the R session, rather than elisp/org buffer. This will be a very familiar working style to lots of people.
There are no doubt a number of different ways of accomplishing this, the simplest being a hack like adding
for(objname in ls())
assign(objname, get(objname), envir=globalenv())
to the source code block function body. (Maybe wrap it in an on.exit() call).
However this may deserve to be thought about more carefully, perhaps with a view to having a uniform approach across languages. E.g. shell code blocks have the same semantics at the moment (no persistence of variables across code blocks), because the body is evaluated in a new bash shell process rather than a running shell. And I guess the same is true for python. However, in both these cases, you could imagine implementing the alternative in which the body is evaluated in a persistent interactive session. It's just that it's particularly natural for R, seeing as both ESS and org-babel evaluate commands in a single persistent R session.
sessions [Eric]
Thanks for bringing this up. I think you are absolutely correct that we should provide support for a persistent environment (maybe called a session) in which to evaluate code blocks. I think the current setup demonstrates my personal bias for a functional style of programming which is certainly not ideal in all contexts.
While the R function you mention does look like an elegant solution, I think we should choose an implementation that would be the same across all source code types. Specifically I think we should allow the user to specify an optional session as a header variable (when not present we assume a default session for each language). The session name could be used to name a comint buffer (like the R buffer) in which all evaluation would take place (within which variables would retain their values –at least once I remove some of the functional method wrappings currently in place– ).
This would allow multiple environments to be used in the same buffer, and once this setup was implemented we should be able to fairly easily implement commands for jumping between source code blocks and the related session buffers, as well as for dumping the last N commands from a session into a new or existing source code block.
Please let me know if you foresee any problems with this proposed setup, or if you think any parts might be confusing for people coming from Sweave. I'll hopefully find some time to work on this later in the week.
can functional and interpreted/interactive models coexist?
Even though both of these use the same *R*
buffer the value of a
is not preserved because it is assigned inside of a functional
wrapper.
a <- 9
b <- 21
a + b
a
This functional wrapper was implemented in order to efficiently return the results of the execution of the entire source code block. However it inhibits the evaluation of source code blocks in the top level, which would allow for persistence of variable assignment across evaluations. How can we allow both evaluation in the top level, and efficient capture of the return value of an entire source code block in a language independent manner?
Possible solutions…
- we can't so we will have to implement two types of evaluation depending on which is appropriate (functional or imperative)
- we remove the functional wrapper and parse the source code block into it's top level statements (most often but not always on line breaks) so that we can isolate the final segment which is our return value.
- we add some sort of "#+return" line to the code block
- we take advantage of each languages support for meta-programming
through
eval
type functions, and use said to evaluate the entire blocks in such a way that their environment can be combined with the global environment, and their results are still captured. - I believe that most modern languages which support interactive
sessions have support for a
last_result
type function, which returns the result of the last input without re-calculation. If widely enough present this would be the ideal solution to a combination of functional and imperative styles.
None of these solutions seem very desirable, but for now I don't see what else would be possible.
Of these options I was leaning towards (1) and (4) but now believe that if it is possible option (5) will be ideal.
(1) both functional and imperative evaluation
Pros
- can take advantage of built in functions for sending regions to the inferior process
- retains the proven tested and working functional wrappers
Cons
- introduces the complication of keeping track of which type of evaluation is best suited to a particular context
- the current functional wrappers may require some changes in order to include the existing global context
(4) exploit language meta-programming constructs to explicitly evaluate code
Pros
- only one type of evaluation
Cons
- some languages may not have sufficient meta-programming constructs
(5) exploit some last_value
functionality if present
Need to ensure that most languages have such a function, those without will simply have to implement their own similar solution…
language | last_value function |
---|---|
R | .Last.value |
ruby | _ |
python | _ |
shell | see last command for shells |
emacs-lisp | see special-case |
82 + 18
last command for shells
Do this using the tee
shell command, and continually pipe the output
to a file.
Got this idea from the following email-thread.
suggested from mailing list
while read line
do
bash -c "$line" | tee /tmp/last.out1
mv /tmp/last.out1 /tmp/last.out
done
another proposed solution from the above thread
#!/bin/bash
# so - Save Output. Saves output of command in OUT shell variable.
OUT=`$*`
echo $OUT
and another
.inputrc: "^[k": accept-line "^M": " | tee /tmp/h_lastcmd.out ^[k"
.bash_profile: export __=/tmp/h_lastcmd.out
If you try it, Alt-k will stand for the old Enter; use "command $__" to access the last output.
Best,
–
Herculano de Lima Einloft Neto
emacs-lisp will be a special case
While it is possible for emacs-lisp to be run in a console type
environment (see the elim
function) it is not possible to run
emacs-lisp in a different session. Meaning any variable set top
level of the console environment will be set everywhere inside
emacs. For this reason I think that it doesn't make any sense to
worry about session support for emacs-lisp.
Further thoughts on 'scripting' vs. functional approaches
These are just thoughts, I don't know how sure I am about this. And again, perhaps I'm not saying anything very radical, just that it would be nice to have some options supporting things like receiving text output in the org buffer.
I can see that you've already gone some way down the road towards the 'last value' approach, so sorry if my comments come rather late. I am concerned that we are not giving sufficient attention to stdout / the text that is returned by the interpreters. In contrast, many of our potential users will be accustomed to a 'scripting' approach, where they are outputting text at various points in the code block, not just at the end. I am leaning towards thinking that we should have 2 modes of evaluation: 'script' mode, and 'functional' mode.
In script mode, evaluation of a code block would result in all text output from that code block appearing as output in the org buffer, presumably as an #+begin_example…#+end_example. There could be an :echo option controlling whether the input commands also appear in the output. [This is like Sweave].
In functional mode, the result of the code block is available as an elisp object, and may appear in the org buffer as an org table/string, via the mechanisms you have developed already.
One thing I'm wondering about is whether, in script mode, there simply should not be a return value. Perhaps this is not so different from what exists: script mode would be new, and what exists currently would be functional mode.
I think it's likely that, while code evaluation will be exciting to people, a large majority of our users in a large majority of their usage will not attempt to actually use the return value from a source code block in any meaningful way. In that case, it seems rather restrictive to only allow them to see output from the end of the code block.
Instead I think the most accessible way to introduce org-babel to people, at least while they are learning it, is as an immensely powerful environment in which to embed their 'scripts', which now also allows them to 'run' their 'scripts'. Especially as such people are likely to be the least capable of the user-base, a possible design-rule would be to make the scripting style of usage easy (default?), perhaps requiring a special option to enable a functional style. Those who will use the functional style won't have a problem understanding what's going on, whereas the 'skript kiddies' might not even know the syntax for defining a function in their language of choice. And of course we can allow the user to set a variable in their .emacs controlling the preference, so that functional users are not inconveniennced by having to provide header args the whole time.
Please don't get the impression that I am down-valuing the functional style of org-babel. I am constantly horrified at the messy 'scripts' that my colleagues produce in perl or R or whatever! Nevertheless that seems to be how a lot of people work.
I think you were leaning towards the last-value approach because it offered the possibility of unified code supporting both the single evaluation environment and the functional style. If you agree with any of the above then perhaps it will impact upon this and mean that the code in the two branches has to differ a bit. In that case, functional mode could perhaps after all evaluate each code block in its own environment, thus (re)approaching 'true' functional programming (side-effects are hard to achieve).
ls > files
echo "There are `wc -l files` files in this directory"
even more thoughts on evaluation, results, models and options
Thanks Dan, These comments are invaluable.
What do you think about this as a new list of priorities/requirements for the execution of source-code blocks.
-
Sessions
- we want the evaluation of the source code block to take place in a session which can persist state (variables, current directory, etc…).
- source code blocks can specify their session with a header argument
- each session should correspond to an Emacs comint buffer so that the user can drop into the session and experiment with live code evaluation.
-
Results
- each source-code block generates some form of results which (as we have already implemented) is transfered into emacs-lisp after which it can be inserted into the org-mode buffer, or used by other source-code blocks
- when the results are translated into emacs-lisp, forced to be interpreted as a scalar (dumping their raw values into the org-mode buffer), as a vector (which is often desirable with R code blocks), or interpreted on the fly (the default option). Note that this is very nearly currently implemented through the results-type-header.
- there should be two means of collecting results from the execution of a source code block. Either the value of the last statement of the source code block, or the collection of all that has been passed to STDOUT during the evaluation.
header argument or return line (header argument)
Rather than using a header argument to specify how the return value
should be passed back, I'm leaning towards the use of a #+RETURN
line inside the block. If such a line is not present then we
default to using STDOUT to collect results, but if such a line is
present then we use it's value as the results of the block. I
think this will allow for the most elegant specification between
functional and script execution. This also cleans up some issues
of implementation and finding which statement is the last
statement.
Having given this more thought, I think a header argument is
preferable. The #+return:
line adds new complicating syntax for
something that does little more than we would accomplish through
the addition of a header argument. The only benefit being that we
know where the final statement starts, which is not an issue in
those languages which contain 'last value' operators.
new header :results
arguments
- script
- explicitly states that we want to use STDOUT to initialize our results
- return_last
- stdout is ignored instead the value of the final statement in the block is returned
- echo
- means echo the contents of the source-code block along
with the results (this implies the script
:results
argument as well)
DONE
rework evaluation lang-by-lang [4/4]
This should include…
- functional results working with the comint buffer
-
results headers
- script
-
return the output of STDOUT
- write a macro which runs the first redirection, executes the body, then runs the second redirection
- last
-
return the value of the last statement
- sessions in comint buffers
DONE
Ruby [4/4]
- functional results working with comint
- script results
- ensure scalar/vector results args are taken into consideration
- ensure callable by other source block
a = 2
b = 4
c = a + b
[a, b, c, 78]
2 | 4 | 6 | 78 |
last.flatten.size + 1
5
ruby sessions
schulte = 27
schulte + 3
schulte
DONE
R [4/4]
- functional results working with comint
- script results
- ensure scalar/vector results args are taken into consideration
- ensure callable by other source block
To redirect output to a file, you can use the sink()
command.
a <- 9
b <- 10
b - a
a + b
83
twoentyseven + 9
28
DONE
Python [4/4]
- functional results working with comint
- script results
- ensure scalar/vector results args are taken into consideration
- ensure callable by other source block
8
9
10
tasking + 2
12
DONE
Shells [4/4]
- functional results working with comint
- script results
- ensure scalar/vector results args are taken into consideration
- ensure callable by other source block
echo 'eric'
date
echo $other ' is the old date'
$ Fri Jun 12 13:08:37 PDT 2009 is the old date
DONE
implement a session header argument [4/4]
:session
header argument to override the default session buffer
DONE ruby
schulte = :in_schulte
:in_schulte
schulte
:in_schulte :in_schulte :in_schulte
DONE python
what = 98
what
98
DONE shell
WHAT='patton'
echo $WHAT
patton
DONE R
a <- 9
b <- 8
a + b
17
a + b
DONE
function to bring up inferior-process buffer [4/4]
This should be callable from inside of a source-code block in an
org-mode buffer. It should evaluate the header arguments, then bring
up the inf-proc buffer using pop-to-buffer
.
For lack of a better place, lets add this to the `org-metadown-hook' hook.
To give this a try, place the cursor on a source block with variables, (optionally git a prefix argument) then hold meta and press down.
DONE ruby
num.times{|n| puts another}
DONE python
another * num
DONE R
a * b
DONE shell
echo $NAME
DEFERRED function to dump last N lines from inf-proc buffer into the current source block
Callable with a prefix argument to specify how many lines should be dumped into the source-code buffer.
REJECTED comint notes
Implementing comint integration in org-babel-comint.el.
Need to have…
-
handling of outputs
- split raw output from process by prompts
- a ring of the outputs, buffer-local, `org-babel-comint-output-ring'
- a switch for dumping all outputs to a buffer
- inputting commands
Lets drop all this language specific stuff, and just use org-babel-comint to split up our outputs, and return either the last value of an execution or the combination of values from the executions.
comint filter functions
;; comint-input-filter-functions hook process-in-a-buffer ;; comint-output-filter-functions hook function modes. ;; comint-preoutput-filter-functions hook ;; comint-input-filter function ...
1
2
3
4
5
DONE Remove protective commas from # comments before evaluating
org inserts protective commas in front of ## comments in language modes that use them. We need to remove them prior to sending code to the interpreter.
,# this one might break it??
:comma_protection
DONE pass multiple reference arguments into R
Can we do this? I wasn't sure how to supply multiple 'var' header args. Just delete this if I'm being dense.
This should be working, see the following example…
n + m
10
DONE ensure that table ranges work
when a table range is passed to org-babel as an argument, it should be interpreted as a vector.
1 | 2 | simple |
2 | 3 | Fixnum:1 |
3 | 4 | Array:123456 |
4 | 5 | |
5 | 6 | |
6 | 7 |
"simple"
"#{n.class}:#{n}"
n
Array:123
ar.size
3
DONE global variable indicating default to vector output
how about an alist… org-babel-default-header-args
this may already
exist… just execute the following and all source blocks will default
to vector output
(setq org-babel-default-header-args '((:results . "vector")))
DONE name named results if source block is named
currently this isn't happening although it should be
:namer
:namer
DONE (simple caching) check for named results before source blocks
see the TODO comment in org-babel-ref.el#org-babel-ref-resolve-reference
DONE
set :results silent
when eval with prefix argument
'silentp
DONE
results-type header (vector/file) [3/3]
In response to a point in Dan's email. We should allow the user to force scalar or vector results. This could be done with a header argument, and the default behavior could be controlled through a configuration variable.
:scalar
":scalar" |
since it doesn't make sense to turn a vector into a scalar, lets just add a two values…
- vector
- forces the results to be a vector (potentially 1 dimensional)
- file
- this throws an error if the result isn't a string, and tries to treat it as a path to a file.
I'm just going to cram all of these into the :results
header
argument. Then if we allow multiple header arguments it should
work out, for example one possible header argument string could be
:results replace vector file
, which would replace any existing
results forcing the results into an org-mode table, and
interpreting any strings as file paths.
DONE
multiple :results
headers
:schulte
DONE file result types
When inserting into an org-mode buffer create a link with the path
being the value, and optionally the display being the
file-name-nondirectory
if it exists.
"something"
This will be useful because blocks like ditaa
and dot
can return
the string path of their files, and can add file
to their results
header.
DONE vector result types
8
8 |
DONE results name
In order to do this we will need to start naming our results.
Since the source blocks are named with #+srcname:
lines we can
name results with #+resname:
lines (if the source block has no
name then no name is given to the #+resname:
line on creation,
otherwise the name of the source block is used).
This will have the additional benefit of allowing results and source blocks to be located in different places in a buffer (and eventually in different buffers entirely).
'schulte
Once source blocks are able to find their own #+resname:
lines
we then need to…
(sbe "developing-resnames")
schulte
TODO change the results insertion functions to use these lines
TODO
teach references to resolve #+resname
lines.
DONE
org-babel tests org-babel [1/1]
since we are accumulating this nice collection of source-code blocks in the sandbox section we should make use of them as unit tests. What's more, we should be able to actually use org-babel to run these tests.
We would just need to cycle over every source code block under the sandbox, run it, and assert that the return value is equal to what we expect.
I have the feeling that this should be possible using only org-babel functions with minimal or no additional elisp. It would be very cool for org-babel to be able to test itself.
This is now done, see /mirrors/org-mode/src/commit/c23e62cd09049b993b32454837c472fc754f4d49/%2A%20Tests.
DEFERRED org-babel assertions (may not be necessary)
These could be used to make assertions about the results of a source-code block. If the assertion fails then the point could be moved to the block, and error messages and highlighting etc… could ensue
DONE make C-c C-c work anywhere within source code block?
This seems like it would be nice to me, but perhaps it would be inefficient or ugly in implementation? I suppose you could search forward, and if you find #+end_src before you find #+begin_src, then you're inside one. [DED]
Agreed, I think inside of the #+srcname: line
would be useful as
well.
'schulte
DONE integration with org tables
We should make it easy to call org-babel source blocks from org-mode table formulas. This is practical now that it is possible to pass arguments to org-babel source blocks.
See the related sandbox header for tests/examples.
digging in org-table.el
In the past org-table.el has proven difficult to work with.
Should be a hook in org-table-eval-formula.
Looks like I need to change this if statement (line 2239) into a cond expression.
DONE source blocks as functions
Allow source code blocks to be called like functions, with arguments specified. We are already able to call a source-code block and assign it's return result to a variable. This would just add the ability to specify the values of the arguments to the source code block assuming any exist. For an example see
When a variable appears in a header argument, how do we differentiate between it's value being a reference or a literal value? I guess this could work just like a programming language. If it's escaped or in quotes, then we count it as a literal, otherwise we try to look it up and evaluate it.
DONE
folding of code blocks? [2/2]
[DED] In similar way to using outline-minor-mode for folding function bodies, can we fold code blocks? #+begin whatever statements are pretty ugly, and in any case when you're thinking about the overall game plan you don't necessarily want to see the code for each Step.
DONE folding of source code block
Sounds good, and wasn't too hard to implement. Code blocks should now be fold-able in the same manner as headlines (by pressing TAB on the first line).
REJECTED folding of results
So, lets do a three-stage tab cycle… First fold the src block, then fold the results, then unfold.
There's no way to tell if the results are a table or not w/o actually executing the block which would be too expensive of an operation.
DONE selective export of text, code, figures
[DED] The org-babel buffer contains everything (code, headings and notes/prose describing what you're up to, textual/numeric/graphical code output, etc). However on export to html / LaTeX one might want to include only a subset of that content. For example you might want to create a presentation of what you've done which omits the code.
[EMS] So I think this should be implemented as a property which can
be set globally or on the outline header level (I need to review
the mechanics of org-mode properties). And then as a source block
header argument which will apply only to a specific source code
block. A header argument of :export
with values of
-
code
- just show the code in the source code block
-
none
- don't show the code or the results of the evaluation
-
results
- just show the results of the code evaluation (don't show the actual code)
-
both
- show both the source code, and the results
this will be done in (sandbox) selective export.
DONE a header argument specifying silent evaluation (no output)
This would be useful across all types of source block. Currently
there is a :replace t
option to control output, this could be
generalized to an :output
option which could take the following
options (maybe more)
-
t
- this would be the default, and would simply insert the results after the source block
-
replace
- to replace any results which may already be there
-
silent
- this would inhibit any insertion of the results
This is now implemented see the example in the sandbox
DONE assign variables from tables in R
This is now working (see (sandbox-table)-R). Although it's not that impressive until we are able to print table results from R.
DONE insert 2-D R results as tables
everything is working but R and shell
DONE shells
DONE R
This has already been tackled by Dan in org-R:check-dimensions. The functions there should be useful in combination with R-export-to-csv as a means of converting multidimensional R objects to emacs lisp.
It may be as simple as first checking if the data is multidimensional,
and then, if so using write
to write the data out to a temporary
file from which emacs can read the data in using org-table-import
.
Looking into this further, is seems that there is no such thing as a scalar in R R-scalar-vs-vector In that light I am not sure how to deal with trivial vectors (scalars) in R. I'm tempted to just treat them as vectors, but then that would lead to a proliferation of trivial 1-cell tables…
DONE allow variable initialization from source blocks
Currently it is possible to initialize a variable from an org-mode
table with a block argument like table=sandbox
(note that the
variable doesn't have to named table
) as in the following example
1 | 2 | 3 |
4 | schulte | 6 |
(message (format "table = %S" table))
"table = ((1 2 3) (4 \"schulte\" 6))"
It would be good to allow initialization of variables from the results
of other source blocks in the same manner. This would probably
require the addition of #+SRCNAME: example
lines for the naming of
source blocks, also the table=sandbox
syntax may have to be expanded
to specify whether the target is a source code block or a table
(alternately we could just match the first one with the given name
whether it's a table or a source code block).
At least initially I'll try to implement this so that there is no need to specify whether the reference is to a table or a source-code block. That seems to be simpler both in terms of use and implementation.
This is now working for emacs-lisp, ruby and python (and mixtures of the three) source blocks. See the examples in the sandbox.
This is currently working only with emacs lisp as in the following example in the emacs lisp source reference.
TODO
Add languages [10/14]
I'm sure there are many more that aren't listed here. Please add them, and bubble any that you particularly care about up to the top.
Any new language should be implemented in a org-babel-lang.el file. Follow the pattern set by org-babel-script.el, org-babel-shell.el and org-babel-R.el.
STARTED ocaml [2/3]
- Working for the simple case (no arguments, simple output)
- correct handling of vector/list output
- ability to import arguments
let rec fib x =
match x with
| 0 -> 1
| 1 -> 1
| n -> fib(n - 1) + fib(n - 2) in
fib 12
233
"string"
"string"
[1; 2; 3; 4]
1 | 2 | 3 | 4 |
[|"ocaml"; "array"|]
"ocaml" | "array" |
TODO perl
This could probably be added to org-babel-script.el
TODO java
STARTED SQL
Things left to do
- support for sessions
- add more useful header arguments (user, passwd, database, etc…)
- support for more engines (currently only supports mysql)
- what's a reasonable way to drop table data into SQL?
show databases
"Database" |
"information_schema" |
"test" |
DONE CSS
trivial org-babel-css.el
DONE ditaa
DONE
gnuplot [7/7]
(see file result types)
independent var | first dependent var | second dependent var |
---|---|---|
0.1 | 0.425 | 0.375 |
0.2 | 0.3125 | 0.3375 |
0.3 | 0.24999993 | 0.28333338 |
0.4 | 0.275 | 0.28125 |
0.5 | 0.26 | 0.27 |
0.6 | 0.25833338 | 0.24999993 |
0.7 | 0.24642845 | 0.23928553 |
0.8 | 0.23125 | 0.2375 |
0.9 | 0.23333323 | 0.2333332 |
1 | 0.2225 | 0.22 |
1.1 | 0.20909075 | 0.22272708 |
1.2 | 0.19999998 | 0.21458333 |
1.3 | 0.19615368 | 0.21730748 |
set title "Implementing Gnuplot"
plot data using 1:2 with lines
DONE add variables
gnuplot 4.2 and up support user defined variables. This is how we will handle variables with org-babel (meaning we will need to require gnuplot 4.2 and up for variable support, which can be install using macports on Mac OSX).
- scalar variables should be replaced in the body of the gnuplot code
- vector variables should be exported to tab-separated files, and the variable names should be replaced with the path to the files
DONE direct plotting w/o session
DEFERRED gnuplot support for column/row names
This should be implemented along the lines of the R-colname-support.
We can do something similar to the :labels param in org-plot, we just have to be careful to ensure that each label is aligned with the related data file.
This may be walking too close to an entirely prebuilt plotting tool rather than straight gnuplot code evaluation. For now I think this can wait.
DONE
a file
header argument
DONE helpers from org-plot.el
There are a variety of helpers in org-plot which can be fit nicely into custom gnuplot header arguments.
These should all be in place by now.
DEFERRED header argument specifying 3D data
0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
0 | 0 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 0 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 1 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 1 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 1 | 1 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 1 | 0 |
0 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 0 | 1 | 0 | 1 | 0 | 0 | 0 | 1 | 0 |
0 | 0 | 1 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 0 | 0 | 1 | 1 | 1 | 0 | 0 |
0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
DONE gnuplot sessions
Working on this, we won't support multiple sessions as `gnuplot-mode' isn't setup for such things.
Also we can't display results with the default :none session, so for gnuplot we really want the default behavior to be :default, and to only run a :none session when explicitly specified.
set title "Implementing Gnuplot Sessions"
plot data using 1:2 with lines
DONE dot
(see file result types)
digraph data_relationships {
"data_requirement" [shape=Mrecord, label="{DataRequirement|description\lformat\l}"]
"data_product" [shape=Mrecord, label="{DataProduct|name\lversion\lpoc\lformat\l}"]
"data_requirement" -> "data_product"
}
DONE asymptote
(see file result types)
for information on asymptote see http://asymptote.sourceforge.net
import graph;
size(0,4cm);
real f(real t) {return 1+cos(t);}
path g=polargraph(f,0,2pi,operator ..)--cycle;
filldraw(g,pink);
xaxis("$x$",above=true);
yaxis("$y$",above=true);
dot("$(a,0)$",(1,0),N);
dot("$(2a,0)$",(2,0),N+E);
DONE ruby
DONE python
DONE R
DONE emacs-lisp
DONE sh
Bugs [37/41]
DONE Python session evaluation bug
The following block evaluates correctly with :session none (set :results to output), but fails with session-based evaluation (with :results value, as below, you see the error message)
I edebug'ed it and it seems fine until we go to comint.
import os
from subprocess import *
chunk = 10000
format = '.gen.gz'
cc = [('58C','NBS'),
('58C','POBI'),
('NBS','POBI')]
for outdir in ['none', 'noscots', 'lax', 'strict']:
outdir = os.path.join('exclusion-study', outdir)
for case, control in cc:
outfile = os.path.join(outdir, '%s-vs-%s-direct' % (case, control))
cmd = 'snptest %s -frequentist 1 -hwe ' % ('-gen_gz' if format == '.gen.gz' else '')
cmd += '-cases %s %s ' % (case + format, case + '.sample')
cmd += '-controls %s %s ' % (control + format, control + '.sample')
cmd += '-exclude_samples %s ' % os.path.join(outdir, 'exclusions')
cmd += '-o %s ' % outfile
cmd += '-chunk %d ' % chunk
cmd += '> %s' % outfile + '.log'
# os.system(cmd)
print(cmd)
snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls NBS.gen.gz NBS.sample -exclude_samples exclusion-study/none/exclusions -o exclusion-study/none/58C-vs-NBS-direct -chunk 10000 > exclusion-study/none/58C-vs-NBS-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/none/exclusions -o exclusion-study/none/58C-vs-POBI-direct -chunk 10000 > exclusion-study/none/58C-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases NBS.gen.gz NBS.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/none/exclusions -o exclusion-study/none/NBS-vs-POBI-direct -chunk 10000 > exclusion-study/none/NBS-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls NBS.gen.gz NBS.sample -exclude_samples exclusion-study/noscots/exclusions -o exclusion-study/noscots/58C-vs-NBS-direct -chunk 10000 > exclusion-study/noscots/58C-vs-NBS-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/noscots/exclusions -o exclusion-study/noscots/58C-vs-POBI-direct -chunk 10000 > exclusion-study/noscots/58C-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases NBS.gen.gz NBS.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/noscots/exclusions -o exclusion-study/noscots/NBS-vs-POBI-direct -chunk 10000 > exclusion-study/noscots/NBS-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls NBS.gen.gz NBS.sample -exclude_samples exclusion-study/lax/exclusions -o exclusion-study/lax/58C-vs-NBS-direct -chunk 10000 > exclusion-study/lax/58C-vs-NBS-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/lax/exclusions -o exclusion-study/lax/58C-vs-POBI-direct -chunk 10000 > exclusion-study/lax/58C-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases NBS.gen.gz NBS.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/lax/exclusions -o exclusion-study/lax/NBS-vs-POBI-direct -chunk 10000 > exclusion-study/lax/NBS-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls NBS.gen.gz NBS.sample -exclude_samples exclusion-study/strict/exclusions -o exclusion-study/strict/58C-vs-NBS-direct -chunk 10000 > exclusion-study/strict/58C-vs-NBS-direct.log snptest -gen_gz -frequentist 1 -hwe -cases 58C.gen.gz 58C.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/strict/exclusions -o exclusion-study/strict/58C-vs-POBI-direct -chunk 10000 > exclusion-study/strict/58C-vs-POBI-direct.log snptest -gen_gz -frequentist 1 -hwe -cases NBS.gen.gz NBS.sample -controls POBI.gen.gz POBI.sample -exclude_samples exclusion-study/strict/exclusions -o exclusion-study/strict/NBS-vs-POBI-direct -chunk 10000 > exclusion-study/strict/NBS-vs-POBI-direct.log
DONE require users to explicitly turn on each language
As we continue to add more languages to org-babel, many of which will require new major-modes we need to re-think how languages are added to org-babel.
Currently we are requiring all available languages in the org-babel-init.el file. I think we need to change this to a user setting so that only the language which have their requirements met (in terms of system executables and emacs major modes) are loaded. It is one more step for install, but it seems to me to be the only solution. Thoughts?
proposed
we add something like the following to the instillation instructions
;; Uncomment each of the following require lines if you want org-babel
;; to support that language. Each language has a comment explaining
;; it's dependencies. See the related files in lisp/langs for more
;; detailed explanations of requirements.
;;
;; (require 'org-babel-R) ;; ess-mode
;; (require 'org-babel-asymptote) ;; asymptote be installed on your system
;; (require 'org-babel-css) ;; none
;; (require 'org-babel-ditaa) ;; ditaa be installed on your system
;; (require 'org-babel-dot) ;; dot be installed on your system
;; (require 'org-babel-gnuplot) ;; gnuplot-mode
;; (require 'org-babel-python) ;; python-mode
;; (require 'org-babel-ruby) ;; inf-ruby mode, ruby and irb must be installed on your system
;; (require 'org-babel-sql) ;; none
note that org-babel-sh
, org-babel-emacs-lisp
are not included in
the list as they can safely be assumed to work on any system.
impetus
we should come up with a way to gracefully degrade when support for a specific language is missing
> To demonstrate creation of documents, open the "test-export.org" file in > the base of the org-babel directory, and export it as you would any > other org-mode file. The "exports" header argument controls how > source-code blocks are exported, with the following options > > - none :: no part of the source-code block is exported in the document > - results :: only the output of the evaluated block is exported > - code :: the code itself is exported > - both :: both the code and results are exported
I have this error showing up:
executing Ruby source code block apply: Searching for program: no such file or directory, irb
DONE problem with newlines in output when :results value
'\n'.join(map(str, range(4)))
0 1 2 3
Whereas I was hoping for
0 |
1 |
2 |
3 |
Note: to generate the above you can try using the new raw
results
header.
'|'+'|\n|'.join(map(str, range(4)))+'|'
0 |
1 |
2 |
3 |
This is now working, it doesn't return as a table because the value returned is technically a string. To return the table mentioned above try something like the following.
[[0], [1], [2], [3]]
0 |
1 |
2 |
3 |
This is some sort of non-printing char / quoting issue I think. Note that
'\\n'.join(map(str, range(4)))
0\n1\n2\n3
Also, note that
print('\n'.join(map(str, range(4))))
0 1 2 3
collapsing consecutive newlines in string output
This is an example of the same bug
"the first line ends here
and this is the second one
even a third"
This doesn't produce anything at all now. I believe that's because I've changed things so that :results output really does not get the value of the block, only the STDOUT. So if we add a print statement this works OK.
print "the first line ends here
and this is the second one
even a third"
the first line ends here and this is the second one even a third
However, the behaviour with :results value is wrong
"the first line ends here
and this is the second one
even a third"
0
TODO prompt characters appearing in output with R
x <- 6
y <- 8
3
> [1] 3
TODO o-b-execute-subtree overwrites heading when subtree is folded
Example
Try M-x org-babel-execute-subtree with the subtree folded and point at the beginning of the heading line.
size=5
DONE non-orgtbl formatted lists
for example
'((:results . "replace"))
(:results . "replace") |
'(eric schulte)
eric | schulte |
PROPOSED allow un-named arguments
x
## produces no output
It's not essential but would be nice for this to work. To do it properly, would mean that we'd have to specify rules for how a string of supplied arguments (some possibly named) interact with the arguments in the definition (some possibly with defaults) to give values to the variables in the funbction body.
PROPOSED external shell execution can't isolate return values
I have no idea how to do this as of yet. The result is that when
shell functions are run w/o a session there is no difference between
the output
and value
result arguments.
Yea, I don't know how to do this either. I searched extensively on how to isolate the last output of a series of shell commands (see [[* last command for shells][last command for shells]]). The results of the search were basically that it was not possible (or at least not accomplish-able with a reasonable amount of effort).
That fact combined with the tenancy to all ways use standard out in
shell scripts led me to treat these two options (output
and value
)
as identical in shell evaluation. Not ideal but maybe good enough for
the moment.
In the `results' branch I've changed this so that they're not quite identical: output results in raw stdout contents, whereas value converts it to elisp, perhaps to a table if it looks tabular. This is the same for the other languages. [Dan]
DONE adding blank line when source-block produces no output
find . \( -path \*/SCCS -o -path \*/RCS -o -path \*/CVS -o -path \*/MCVS -o -path \*/.svn -o -path \*/.git -o -path \*/.hg -o -path \*/.bzr -o -path \*/_MTN -o -path \*/_darcs -o -path \*/\{arch\} \) -prune -o -type f \( -iname \*.el \) -exec grep -i -nH -e org-babel-trim {} \;
DONE Allow source blocks to be recognised when #+ are not first characters on the line
I think Carsten has recently altered the core so that #+ can have preceding whitespace, at least for literal/code examples. org-babel should support this.
(message "i'm indented")
(message "I'm not indented")
(message "I said %s" speech)
DONE are the org-babel-trim s necessary?
at the end of e.g. org-babel-R-evaluate, org-babel-python-evaluate, but not org-babel-ruby-evaluate
I think it depends on the language, if we find that extra blank lines are being inserted in a particular language that is a good indication that the trim or chomp functions may be appropriate.
org-babel-trim and the related org-babel-chomp are use throughout org-babel…
find lisp/ \( -path \*/SCCS -o -path \*/RCS -o -path \*/CVS -o -path \*/MCVS -o -path \*/.svn -o -path \*/.git -o -path \*/.hg -o -path \*/.bzr -o -path \*/_MTN -o -path \*/_darcs -o -path \*/\{arch\} \) -prune -o -type f \( -iname \*.el \) -exec grep -i -nH org-babel-trim {} \;
lisp//langs/org-babel-python.el:49: vars "\n") "\n" (org-babel-trim body) "\n")) ;; then the source block body lisp//langs/org-babel-python.el:143: (org-remove-indentation (org-babel-trim body)) "[\r\n]"))) lisp//langs/org-babel-python.el:158: #'org-babel-trim lisp//langs/org-babel-python.el:166: (reverse (mapcar #'org-babel-trim raw))))))) lisp//langs/org-babel-python.el:169: (output (org-babel-trim (mapconcat #'identity (reverse (cdr results)) "\n"))) lisp//langs/org-babel-python.el:170: (value (org-babel-python-table-or-string (org-babel-trim (car results))))))))) lisp//langs/org-babel-ruby.el:149: (mapcar #'org-babel-trim raw))))))) lisp//langs/org-babel-sh.el:148: (mapcar #'org-babel-trim raw))))))) lisp//langs/org-babel-sh.el:151: (output (org-babel-trim (mapconcat #'org-babel-trim (reverse results) "\n"))) lisp//org-babel-ref.el:161: (mapcar #'org-babel-trim (reverse (cons buffer return))))) lisp//org-babel.el:198: (with-temp-buffer (insert (org-babel-trim body)) (copy-region-as-kill (point-min) (point-max))) lisp//org-babel.el:465: (org-babel-trim lisp//org-babel.el:706:(defun org-babel-trim (string &optional regexp)
find lisp/ \( -path \*/SCCS -o -path \*/RCS -o -path \*/CVS -o -path \*/MCVS -o -path \*/.svn -o -path \*/.git -o -path \*/.hg -o -path \*/.bzr -o -path \*/_MTN -o -path \*/_darcs -o -path \*/\{arch\} \) -prune -o -type f \( -iname \*.el \) -exec grep -i -nH org-babel-chomp {} \;
lisp//langs/org-babel-R.el:122: (full-body (mapconcat #'org-babel-chomp lisp//langs/org-babel-R.el:143: (delete nil (mapcar #'extractor (mapcar #'org-babel-chomp raw))) "\n")))))))) lisp//langs/org-babel-ruby.el:143: #'org-babel-chomp lisp//langs/org-babel-sh.el:142: (full-body (mapconcat #'org-babel-chomp lisp//org-babel-tangle.el:163: (insert (format "\n%s\n" (org-babel-chomp body))) lisp//org-babel.el:362: (org-babel-chomp (match-string 2 arg))) lisp//org-babel.el:698:(defun org-babel-chomp (string &optional regexp) lisp//org-babel.el:707: "Like `org-babel-chomp' only it runs on both the front and back of the string" lisp//org-babel.el:708: (org-babel-chomp (org-babel-reverse-string lisp//org-babel.el:709: (org-babel-chomp (org-babel-reverse-string string) regexp)) regexp))
DONE LoB is not populated on startup
org-babel-library-of-babel is nil for me on startup. I have to evaluate the org-babel-lob-ingest line manually.
1 | 2 |
2 | 4 |
3 | 9 |
4 | 16 |
5 | 25 |
I've added a section to org-babel-init.el which will load the library of babel on startup.
Note that this needs to be done in org-babel-init.el rather than in org-babel-lob.el, not entirely sure why, something about it being required directly?
Also, I'm now having the file closed if it wasn't being visited by a buffer before being loaded.
DONE use new merge function here?
And at other occurrences of org-combine-plists?
DONE creeping blank lines
There's still inappropriate addition of blank lines in some circumstances.
Hmm, it's a bit confusing. It's to do with o-b-remove-result. LoB removes the entire (#+resname and result) and starts from scratch, whereas #+begin_src only removes the result. I haven't worked out what the correct fix is yet. Maybe the right thing to do is to make sure that those functions (o-b-remove-result et al.) are neutral with respect to newlines. Sounds easy, but…
E.g.
b=5
Compare the results of
22
23
23
DONE #+srcname arg parsing bug
arg
2
arg
40
arg
40
DONE Fix nested evaluation and default args
The current parser / evaluator fails with greater levels of nested function block calls (example below).
Initial statement [ded]
If we want to overcome this I think we'd have to redesign some of the evaluation mechanism. Seeing as we are also facing issues like dealing with default argument values, and seeing as we now know how we want the library of babel to behave in addition to the source blocks, now might be a good time to think about this. It would be nice to do the full thing at some point, but otoh we may not consider it a massive priority.
AIui, there are two stages: (i) construct a parse tree, and (ii) evaluate it and return the value at the root. In the parse tree each node represents an unevaluated value (either a literal value or a reference). Node v may have descendent nodes, which represent values upon which node v's evaluation depends. Once that tree is constructed, then we evaluate the nodes from the tips towards the root (a post-order traversal).
[This would also provide a solution for concatenating the STDOUTs of called blocks, which is a task below; we concatenate them in whatever order the traversal is done in.]
In addition to the variable references (i.e. daughter nodes), each node would contain the information needed to evaluate that node (e.g. lang body). Then we would pass a function postorder over the tree which would call o-b-execute-src-block at each node, finally returning the value at the root.
Fwiw I made a very tentative small start at stubbing this out in org-babel-call.el in the 'evaluation' branch. And I've made a start at sketching a parsing algorithm below.
Parse tree algorithm
Seeing as we're just trying to parse a string like f(a=1,b=g(c=2,d=3)) it shouldn't be too hard. But of course there are 'proper' parsers written in elisp out there, e.g. Semantic. Perhaps we can find what we need – our syntax is pretty much the same as python and R isn't it?
Or, a complete hack, but maybe it would be we easy to transform it to XML and then parse that with some existing tool?
But if we're doing it ourselves, something very vaguely like this? (I'm sure there're lots of problems with this)
## we are currently reading a reference name: the name of the root function
whereami = "refname"
node = root = Node()
for c in call_string:
if c == '(':
varnum = 0
whereami = "varname" # now we're reading a variable name
if c == '=':
new = Node()
node.daughters = [node.daughters, new]
new.parent = node
node = new
whereami = "refname"
if c == ',':
whereami = "varname"
varnum += 1
elif c == ')':
node = node.parent
elif c == ' ':
pass
else:
if whereami = "varname":
node.varnames[varnum] += c
elif whereami = "refname":
node.name += c
discussion / investigation
I believe that this issue should be addressed as a bug rather than as a point for new development. The code in org-babel-ref.el already resolves variable references in a recursive manner which should work in the same manner regardless of the depth of the number of nested function calls. This recursive evaluation has the effect of implicitly constructing the parse tree that your are thinking of constructing explicitly.
Through using some of the commented out debugging statements in
org-babel-ref.el I have looked at what may be going wrong in the
current evaluation setup, and it seems that nested variables are being
set using the :var
header argument, and these variables are being
overridden by the default variables which are being entered through
the new functional syntax (see the demonstration header below).
I believe that once this bug is fixed we should be back to fully resolution of nested arguments. We should capture this functionality in a test to ensure that we continue to test it as we move forward. I can take a look at implementing this once I get a chance.
Looks like the problem may be in org-babel-merge-params, which seems to be trampling the provided :vars values.
Nope, now it seems that we are actually looking up the results line, rather than the actual source-code block, which would make sense given that the results-line will return the same value regardless of the arguments supplied. See the output of this debug-statement.
We need to be sure that we don't read from a #+resname:
line when we
have a non-nil set of arguments.
demonstration
After uncommenting the debugging statements located here and more
importantly here, we can see that the current reference code does
evaluate the references correctly, and it uses the :var
header
argument to set a=8
, however the default variables specified using
the functional syntax in adder(a=3, b=2)
is overriding this
specification.
doesn't work with functional syntax
a + b
5
arg
5
still does work with :var
syntax
so it looks like regardless of the syntax used we're not overriding the default argument values.
a + b
5
arg
5
Set of test cases
Both defaults provided in definition
a+b
30
30
## should be 30 ## OK, but
E.g. remove spaces between parens above
updated org-babel-lob-one-liner-regexp
20
## should be 20
## should be 10
10
3
One arg lacks default in definition
a+b
[no output]
## should be error: b has no default
Maybe we should let the programming language handle this case. For
example python spits out an error in the #+lob
line above. Maybe
rather than catching these errors our-selves we should raise an error
when the source-block returns an error. I'll propose a task for this
idea, I'm not sure how/if it would work…
11
## should be 11 ## OK
3
## should be 3
Example that fails
a+b
1
nesting
arg
nil
arg
12
DONE deeply nested arguments still fails
arg
4
Used to result in this error
supplied params=nil new-refere="adder", new-referent="a=adder(a=one(),b=one()),b=adder(a=one(),b=one())" args=((:var . "a=adder(a=one()") (:var . "b=one())") (:var . "b=adder(a=one()") (:var . "b=one())")) type=source-block supplied params=((:var . "a=adder(a=one()") (:var . "b=one())") (:var . "b=adder(a=one()") (:var . "b=one())")) new-refere="adder", new-referent="a=one(" args=((:var . "a=one(")) type=source-block supplied params=((:var . "a=one(")) reference 'one(' not found in this buffer
Need to change the regexp in org-babel-ref-resolve-reference so that it only matches when the parenthesis are balanced. Maybe look at this.
DONE Still some problems with deeply nested arguments and defaults
sandbox
DONE Parsing / defaults bug
Try inserting a space between 'a=0,' and 'b=0' and comparing results
arg
99
arg
10
DONE Nesting problem II
DONE Why does this give 8?
It was picking up the wrong definition of adder
arg
101
DONE Problem with empty argument list
This gives empty list with () and 'no output' with ( )
I think this is OK now.
arg
99
arg
10
DONE Nesting problem II
DONE Why does this give 8?
It was picking up the wrong definition of adder
arg
101
DONE Problem with empty argument list
This gives empty list with () and 'no output' with ( )
I think this is OK now.
arg
99
arg
DONE Arg lacking default
This would be good thing to address soon. I'm imagining that e.g. here, the 'caller' block would return the answer 30. I believe there's a few issues here: i.e. the naked 'a' without a reference is not understood; the default arg b=6 is not understood.
a+b
var
DONE allow srcname to omit function call parentheses
Someone needs to revisit those regexps. Is there an argument for moving some of the regexps used to match function calls into defvars? (i.e. in o-b.el and o-b-ref.el)
This seems to work now. It still might be a good idea to separate out some of the key regexps into defvars at some point.
3.times {puts 'x'}
x x x
DONE avoid stripping whitespace from output when :results output
This may be partly solved by using o-b-chomp rather than o-b-trim in the o-b-LANG-evaluate functions.
DEFERRED weird escaped characters in shell prompt break shell evaluation
E.g. this doesn't work. Should the shell sessions set a sane prompt when they start up? Or is it a question of altering comint-prompt-regexp? Or altering org-babel regexps?
black=30 ; red=31 ; green=32 ; yellow=33 ; blue=34 ; magenta=35 ; cyan=36 ; white=37
prompt_col=$red
prompt_char='>'
export PS1="\[\033[${prompt_col}m\]\w${prompt_char} \[\033[0m\]"
I just pushed a good amount of changes, could you see if your shell problems still exist?
The problem's still there. Specifically, aIui, at this line of org-babel-sh.el, raw gets the value
("" "[0m Sun Jun 14 19:26:24 EDT 2009\n" "[0m org_babel_sh_eoe\n" "[0m ")
and therefore (member org-babel-sh-eoe-output …) fails
I think that `comint-prompt-regexp' needs to be altered to match the shell prompt. This shouldn't be too difficult to do by hand, using the `regexp-builder' command and should probably be part of the user's regular emacs init. I can't think of a way for us to set this automatically, and we are SOL without a regexp to match the prompt.
DONE function calls in #+srcname: refs
My srcname references don't seem to be working for function calls. This needs fixing.
59
srcname function call doesn't work for calling a source block
var1
59
They do work for a simple reference
var1
59
and they do work for :var header arg
var1
58
DONE LoB: with output to buffer, not working in buffers other than library-of-babel.org
Initial report
I haven't fixed this yet. org-babel-ref-resolve-reference moves point around, inside a save-excursion. Somehow when it comes to inserting the results (after possible further recursive calls to org-babel-ref-resolve-reference), point hasn't gone back to the lob line.
1 | 1 |
2 | .5 |
3 | .333 |
11
Now
I think this got fixed in the bugfixes before merging results into master.
DONE cursor movement when evaluating source blocks
E.g. the pie chart example. Despite the save-window-excursion in org-babel-execute:R. (I never learned how to do this properly: org-R jumps all over the place…)
I don't see this now [ded]
DONE LoB: calls fail if reference has single character name
commit 21d058869d
This doesn't work
1 | 1 |
2 | .5 |
3 | .3333 |
4 | .25 |
5 | .2 |
6 | .1666 |
But this is OK
1 | 1 |
2 | .5 |
3 | .3333 |
4 | .25 |
5 | .2 |
6 | .1666 |
DONE make :results replace the default?
I'm tending to think that appending results to pre-existing results creates mess, and that the cleaner `replace' option should be the default. E.g. when a source block creates an image, we would want that to be updated, rather than have a new one be added.
I agree.
DONE ruby evaluation not working under ubuntu emacs 23
With emacs 23.0.91.1 on ubuntu, for C-h f run-ruby I have the following, which seems to conflict with this line in org-babel-ruby.el.
run-ruby is an interactive compiled Lisp function. (run-ruby cmd) Run an inferior Ruby process, input and output via buffer *ruby*. If there is a process already running in `*ruby*', switch to that buffer. With argument, allows you to edit the command line (default is value of `ruby-program-name'). Runs the hooks `inferior-ruby-mode-hook' (after the `comint-mode-hook' is run). (Type C-h m in the process buffer for a list of commands.)
So, I may have a non-standard inf-ruby.el. Here's my version of run-ruby.
run-ruby is an interactive Lisp function in `inf-ruby.el'. (run-ruby &optional COMMAND NAME) Run an inferior Ruby process, input and output via buffer *ruby*. If there is a process already running in `*ruby*', switch to that buffer. With argument, allows you to edit the command line (default is value of `ruby-program-name'). Runs the hooks `inferior-ruby-mode-hook' (after the `comint-mode-hook' is run). (Type C-h m in the process buffer for a list of commands.)
It seems we could either bundle my version of inf-ruby.el (as it's the newest). Or we could change the use of `run-ruby' so that it is robust across multiple distributions. I think I'd prefer the former, unless the older version of inf-ruby is actually bundled with emacs, in which case maybe we should go out of our way to support it. Thoughts?
I think for now I'll just include the latest inf-ruby.el in the newly created utility directory. I doubt anyone would have a problem using the latest version of this file.
DONE
test failing forcing vector results with test-forced-vector-results
ruby code block
Note that this only seems to happen the second time the test table is evaluated
8
triv.class.name
mysteriously this seems to be fixed…
DONE defunct R sessions
Sometimes an old R session will turn defunct, and newly inserted code will not be evaluated (leading to a hang).
This seems to be fixed by using `inferior-ess-send-input' rather than `comint-send-input'.
DONE ruby fails on first call to non-default session
:patton
DONE
when reading results from #+resname
line
Errors when trying to read from resname lines.
8
buggy
DONE R-code broke on "org-babel" rename
8 * 2
DONE error on trivial R results
So I know it's generally not a good idea to squash error without
handling them, but in this case the error almost always means that
there was no file contents to be read by org-table-import
, so I
think it's ok.
pie(c(1, 2, 3), labels = c(1, 2, 3))
8
8
c(1,2,3)
1 |
2 |
3 |
DONE ruby new variable creation (multi-line ruby blocks)
Actually it looks like we were dropping all but the last line.
total = 0
table.each{|n| total += n}
total/table.size
2
DONE R code execution seems to choke on certain inputs
Currently the R code seems to work on vertical (but not landscape) tables
"schulte"
num
schulte
(setq debug-on-error t)
'(1 2 3)
mean(mean(table))
2
1 |
2 |
3 |
mean(table)
DONE org bug/request: prevent certain org behaviour within code blocks
E.g. gets recognised as a link (when there's text inside the brackets). This is bad for R code at least, and more generally could be argued to be inappropriate. Is it difficult to get org to ignore text in code blocks? [DED]
I believe Carsten addressed this recently on the mailing list with the comment that it was indeed a difficult issue. I believe this may be one area where we could wait for an upstream (org-mode) fix.
[Dan] Carsten has fixed this now in the core.
DONE with :results replace, non-table output doesn't replace table output
And vice versa. E.g. Try this first with table and then with len(table) [DED]
table
1 | 2 | 3 |
4 | "schulte" | 6 |
2
Yes, this is certainly a problem. I fear that if we begin replacing anything immediately following a source block (regardless of whether it matches the type of our current results) we may accidentally delete hand written portions of the user's org-mode buffer.
I think that the best solution here would be to actually start labeling results with a line that looks something like…
This would have a couple of benefits…
- we wouldn't have to worry about possibly deleting non-results (which is currently an issue)
- we could reliably replace results even if there are different types
- we could reference the results of a source-code block in variable definitions, which would be useful if for example we don't wish to re-run a source-block every time because it is long-running.
Thoughts? If no-one objects, I believe I will implement the labeling of results.
DONE extra quotes for nested string
Well R appears to be reading the tables without issue…
these should be quoted
ls
"COPYING" |
"README.markdown" |
"block" |
"examples.org" |
"existing_tools" |
"intro.org" |
"org-babel" |
"rorg.org" |
"test-export.html" |
"test-export.org" |
tab[1][0]
README.markdown
as.matrix(tab[2,])
README.markdown
DONE simple ruby arrays not working
As an example eval the following. Adding a line to test
3 | 4 | 5 |
ar.first.first
DONE space trailing language name
fix regexp so it works when there's a space trailing the language name
:schulte
DONE Args out of range error
The following block resulted in the error below [DED]. It ran without error directly in the shell.
cd ~/work/genopca
for platf in ill aff ; do
for pop in CEU YRI ASI ; do
rm -f $platf/hapmap-genos-$pop-all $platf/hapmap-rs-all
cat $platf/hapmap-genos-$pop-* > $platf/hapmap-genos-$pop-all
cat $platf/hapmap-rs-* > $platf/hapmap-rs-all
done
done
executing source block with sh… finished executing source block string-equal: Args out of range: "", -1, 0
the error string-equal: Args out of range: "", -1, 0
looks like what
used to be output when the block returned an empty results string.
This should be fixed in the current version, you should now see the
following message no result returned by source block
.
DONE ruby arrays not recognized as such
Something is wrong in /mirrors/org-mode/src/commit/c23e62cd09049b993b32454837c472fc754f4d49/lisp/org-babel-script.el related to the recognition of ruby arrays as such.
[1, 2, 3, 4]
1 | 2 | 3 | 4 |
[1, 2, 3, 4]
1 | 2 | 3 | 4 |
REJECTED elisp reference fails for literal number
That's a bug in Dan's elisp, not in org-babel.
(message a)
30
Tests
Evaluate all the cells in this table for a comprehensive test of the org-babel functionality.
Note: if you have customized org-babel-default-header-args
then some
of these tests may fail.
functionality | block | arg | expected | results | pass |
---|---|---|---|---|---|
basic evaluation | pass | ||||
emacs lisp | basic-elisp | 5 | 5 | pass | |
shell | basic-shell | 6 | 6 | pass | |
ruby | basic-ruby | org-babel | org-babel | pass | |
python | basic-python | hello world | hello world | pass | |
R | basic-R | 13 | 13 | pass | |
tables | pass | ||||
emacs lisp | table-elisp | 3 | 3 | pass | |
ruby | table-ruby | 1-2-3 | 1-2-3 | pass | |
python | table-python | 5 | 5 | pass | |
R | table-R | 3.5 | 3.5 | pass | |
R: col names in R | table-R-colnames | -3 | -3 | pass | |
R: col names in org | table-R-colnames-org | 169 | 169 | pass | |
source block references | pass | ||||
all languages | chained-ref-last | Array | Array | pass | |
source block functions | pass | ||||
emacs lisp | defun-fibb | fibbd | fibbd | pass | |
run over | Fibonacci | 0 | 1 | 1 | pass |
a | Fibonacci | 1 | 1 | 1 | pass |
variety | Fibonacci | 2 | 2 | 2 | pass |
of | Fibonacci | 3 | 3 | 3 | pass |
different | Fibonacci | 4 | 5 | 5 | pass |
arguments | Fibonacci | 5 | 8 | 8 | pass |
bugs and tasks | pass | ||||
simple ruby arrays | ruby-array-test | 3 | 3 | pass | |
R number evaluation | bug-R-number-evaluation | 2 | 2 | pass | |
multi-line ruby blocks | multi-line-ruby-test | 2 | 2 | pass | |
forcing vector results | test-forced-vector-results | Array | Array | pass | |
deeply nested arguments | deeply-nested-args-bug | 4 | 4 | pass | |
sessions | pass | ||||
set ruby session | set-ruby-session-var | :set | :set | pass | |
get from ruby session | get-ruby-session-var | 3 | 3 | pass | |
set python session | set-python-session-var | set | set | pass | |
get from python session | get-python-session-var | 4 | 4 | pass | |
set R session | set-R-session-var | set | set | pass | |
get from R session | get-R-session-var | 5 | 5 | pass |
The second TBLFM line (followed by replacing '[]' with '') can be used to blank out the table results, in the absence of a better method.
basic tests
(+ 1 4)
expr 1 + 5
date
"org-babel"
'hello world'
b <- 9
b + 4
read tables
1 | 2 | 3 |
4 | 5 | 6 |
var1 | var2 | var3 |
---|---|---|
1 | 22 | 13 |
41 | 55 | 67 |
(length (car table))
table.first.join("-")
table[1][1]
mean(mean(table))
sum(table$var2 - table$var3)
x^2
This should return 169. The fact that R is able to use the column name to index the data frame (x$var3) proves that a table with column names (a header row) has been recognised as input for the R-square function block, and that the R-square block has output an elisp table with column names, and that the colnames have again been recognised when creating the R variables in this block.
x$var3[1]
references
Lets pass a references through all of our languages…
Lets start by reversing the table from the previous examples
table.reverse()
table
4 | 5 | 6 |
1 | 2 | 3 |
Take the first part of the list
table[1]
4 |
1 |
Turn the numbers into string
(mapcar (lambda (el) (format "%S" el)) table)
"(4)" | "(1)" |
and Check that it is still a list
table.class.name
source blocks as functions
(defun fibbd (n) (if (< n 2) 1 (+ (fibbd (- n 1)) (fibbd (- n 2)))))
(fibbd n)
sbe tests (these don't seem to be working…)
Testing the insertion of results into org-mode tables.
"the first line ends here
and this is the second one
even a third"
the first line ends here\n\n\n and this is the second one\n\neven a third
raise "oh nooooooooooo"
oh nooooooooooo
the first line ends here… | -:5: warning: parenthesize argument(s) for future version… |
forcing results types tests
8
triv.class.name
sessions
var = [1, 2, 3]
:set
var.size
var=4
'set'
var
a <- 5
'set'
a
Sandbox
To run these examples evaluate org-babel-init.el
org-babel.el beginning functionality
date
Sun Jul 5 18:54:39 EDT 2009
Time.now
Sun Jul 05 18:54:35 -0400 2009
"Hello World"
Hello World
org-babel-R
a <- 9
b <- 16
a + b
25
hist(rgamma(20,3,3))
org-babel plays with tables
Alright, this should demonstrate both the ability of org-babel to read tables into a lisp source code block, and to then convert the results of the source code block into an org table. It's using the classic "lisp is elegant" demonstration transpose function. To try this out…
- evaluate /mirrors/org-mode/src/commit/c23e62cd09049b993b32454837c472fc754f4d49/lisp/org-babel-init.el to load org-babel and friends
- evaluate the transpose definition
\C-c\\C-c
on the beginning of the source block - evaluate the next source code block, this should read in the table
because of the
:var table=previous
, then transpose the table, and finally it should insert the transposed table into the buffer immediately following the block
Emacs lisp
(defun transpose (table)
(apply #'mapcar* #'list table))
1 | 2 | 3 |
4 | schulte | 6 |
(transpose table)
'(1 2 3 4 5)
1 | 2 | 3 | 4 | 5 |
Ruby and Python
table.first.join(" - ")
1 - 2 - 3
table[0]
table
[[1, 2, 3], [4, "schulte", 6]]
1 | 2 | 3 |
4 | "schulte" | 6 |
len(table)
2
"add" | "class" | "contains" | "delattr" | "delitem" | "delslice" | "doc" | "eq" | "format" | "ge" | "getattribute" | "getitem" | "getslice" | "gt" | "hash" | "iadd" | "imul" | "init" | "iter" | "le" | "len" | "lt" | "mul" | "ne" | "new" | "reduce" | "reduce_ex" | "repr" | "reversed" | "rmul" | "setattr" | "setitem" | "setslice" | "sizeof" | "str" | "subclasshook" | "append" | "count" | "extend" | "index" | "insert" | "pop" | "remove" | "reverse" | "sort" |
(sandbox table) R
1 | 2 | 3 |
4 | schulte | 6 |
x <- c(rnorm(10, mean=-3, sd=1), rnorm(10, mean=3, sd=1))
x
-3.35473133869346 |
-2.45714878661 |
-3.32819924928633 |
-2.97310212756194 |
-2.09640758369576 |
-5.06054014378736 |
-2.20713700711221 |
-1.37618039712037 |
-1.95839385821742 |
-3.90407396475502 |
2.51168071590226 |
3.96753011570494 |
3.31793212627865 |
1.99829753972341 |
4.00403686419829 |
4.63723764452927 |
3.94636744261313 |
3.58355906547775 |
3.01563442274226 |
1.7634976849927 |
tabel
1 | 2 | 3 |
4 | "schulte" | 6 |
shell
Now shell commands are converted to tables using org-table-import
and if these tables are non-trivial (i.e. have multiple elements) then
they are imported as org-mode tables…
ls -l
"total" | 208 | "" | "" | "" | "" | "" | "" |
"-rw-r–r–" | 1 | "dan" | "dan" | 57 | 2009 | 15 | "block" |
"-rw-r–r–" | 1 | "dan" | "dan" | 35147 | 2009 | 15 | "COPYING" |
"-rw-r–r–" | 1 | "dan" | "dan" | 722 | 2009 | 18 | "examples.org" |
"drwxr-xr-x" | 4 | "dan" | "dan" | 4096 | 2009 | 19 | "existing_tools" |
"-rw-r–r–" | 1 | "dan" | "dan" | 2207 | 2009 | 14 | "intro.org" |
"drwxr-xr-x" | 2 | "dan" | "dan" | 4096 | 2009 | 18 | "org-babel" |
"-rw-r–r–" | 1 | "dan" | "dan" | 277 | 2009 | 20 | "README.markdown" |
"-rw-r–r–" | 1 | "dan" | "dan" | 11837 | 2009 | 18 | "rorg.html" |
"-rw-r–r–" | 1 | "dan" | "dan" | 61829 | 2009 | 19 | "#rorg.org#" |
"-rw-r–r–" | 1 | "dan" | "dan" | 60190 | 2009 | 19 | "rorg.org" |
"-rw-r–r–" | 1 | "dan" | "dan" | 972 | 2009 | 11 | "test-export.org" |
silent evaluation
:im_the_results
:im_the_results
:im_the_results
:im_the_results_
:im_the_results_
(sandbox) referencing other source blocks
Doing this in emacs-lisp first because it's trivial to convert emacs-lisp results to and from emacs-lisp.
emacs lisp source reference
This first example performs a calculation in the first source block
named top
, the results of this calculation are then saved into the
variable first
by the header argument :var first=top
, and it is
used in the calculations of the second source block.
(+ 4 2)
(* first 3)
18
This example is the same as the previous only the variable being passed through is a table rather than a number.
(defun transpose (table)
(apply #'mapcar* #'list table))
1 | 2 | 3 |
4 | schulte | 6 |
(transpose table)
(transpose table)
1 | 2 | 3 |
4 | "schulte" | 6 |
ruby python
Now working for ruby
89
2 * other
and for python
98
another*3
mixed languages
Since all variables are converted into Emacs Lisp it is no problem to reference variables specified in another language.
2
(* ruby-variable 8)
lisp_var + 4
20
R
a <- 9
a
9
other + 2
11
(sandbox) selective export
For exportation tests and examples see (including exportation of inline source code blocks) /mirrors/org-mode/src/commit/c23e62cd09049b993b32454837c472fc754f4d49/test-export.org
(sandbox) source blocks as functions
5
(* 3 n)
15
result
294
The following just demonstrates the ability to assign variables to literal values, which was not implemented until recently.
num+" schulte "
"eric schulte "
(sandbox) inline source blocks
This is an inline source code block
1 + 6
This is an inline source code block with header arguments.
n
(sandbox) integration w/org tables
(defun fibbd (n) (if (< n 2) 1 (+ (fibbd (- n 1)) (fibbd (- n 2)))))
(fibbd n)
(mapcar #'fibbd '(0 1 2 3 4 5 6 7 8))
Something is not working here. The function `sbe ' works fine when called from outside of the table (see the source block below), but produces an error when called from inside the table. I think there must be some narrowing going on during intra-table emacs-lisp evaluation.
original | fibbd |
---|---|
0 | 1 |
1 | 1 |
2 | 2 |
3 | 3 |
4 | 5 |
5 | 8 |
6 | 13 |
7 | 21 |
8 | 34 |
9 | 55 |
silent-result
(sbe 'fibbd (n "8"))
Buffer Dictionary
LocalWords: DBlocks dblocks org-babel el eric fontification