rx macro and eager macro-expansion failure in Emacs - emacs

(defun my-blah (str)
(rx (+ (eval (substring str 1)))))
Evaluation of that code puts the following message in the Messages buffer:
Eager macro-expansion failure: (void-variable str)
What does it mean and is it something to worry about?
The function seems to work fine. On the other hand, evaluation of the following code doesn't generate such a message:
(defvar my-str "oink")
(rx (+ (eval (substring my-str 1))))
Appendix: alternative ways to define the function without generating such messages
(defun my-nah (str)
(rx-to-string `(+ ,(substring str 1))))
(defun my-llama (str)
(eval `(rx (+ ,(substring str 1)))))

rx is a macro, which implies that it can be expanded pretty much at any time before actually evaluating the code. Typically, it means it will be expanded during compilation. Compilation happens before str is known, so the expansion of rx cannot depend on the value of str.
If you test it without compiling it, then macroexpansion happens late (just before/during evaluation) so the problem is hidden (tho if you use lexical-binding the problem will re-appear because the eval will not have access to the lexical context in which str is defined).
Eager macroexpansion (new in 24.4), is used to expand macros when loading a non-compiled file, which provides a macro-expansion behavior similar to the one for compiled code. For backward-compatibility, when an error occurs during eager macro-expansion, Emacs falls back on expanding macros late (after emitting the warning you saw).

If you want to easily test your rx expressions, use this :
(defun test-rx (test-file test-reg)
(if (get-buffer (setq test-buffer "RX"))
(kill-buffer test-buffer))
(with-temp-buffer
(insert-file-contents-literally test-file)
(goto-char (point-min))
(while
(re-search-forward test-reg nil t)
(when (match-string 0)
(let ((one (match-string 1))
(two (match-string 2)))
(if one
(progn
(pop-to-buffer test-buffer)
(goto-char (point-max))
(insert (format "Found %s and possibly %s" one two)))))))))
(defun test-rx-now (file)
"plop"
(interactive
(let (file)
(setq file (read-file-name "File to search: "))
(list file)))
(test-rx file
(rx
"$(function"
(*? anything)
(or "'" "\"")
(group
(* (not (any "'" "\"")))
".php"
)
(or "'" "\"")
(*? anything)
"})"
)))

Related

Haskell with emacs org-mode: Variable not in scope

After wandering off in frustration from before, I've decided to try Haskell in Emacs org-mode again. I'm using Haskell stack-ghci (8.6.3), Emacs 26.2, org-mode 9.2.3 set up with intero. This code block
#+begin_src haskell :results raw :session *haskell*
pyth2 :: Int -> [(Int, Int, Int)]
pyth2 n =
[ (x, y, z)
| x <- [1 .. n]
, y <- [x .. n]
, z <- [y .. n]
, x ^ 2 + y ^ 2 == z ^ 2
]
#+end_src
produces this RESULTS:
*Main| *Main| *Main| *Main| *Main|
<interactive>:59:16: error: Variable not in scope: n
<interactive>:60:16: error: Variable not in scope: n
<interactive>:61:16: error: Variable not in scope: n
However, this
#+begin_src haskell :results raw
tripleMe x = x + x + x
#+end_src
works fine. I've added the :set +m to both ghci.conf and the individual code block to no effect. This code works fine in a separate hs file run in a separate REPL. The pyth2 code in a separate file also can be called from the org-mode started REPL and run just fine as well. Not sure how to proceed. Can include Emacs init info if necessary.
Over on the org-mode mailing list I got an answer that basically is saying the same as you, D. Gillis. He had a similar work-around that actually is more org-mode-centric. Under a heading where your code blocks will be put this "drawer"
:PROPERTIES:
:header-args:haskell: :prologue ":{\n" :epilogue ":}\n"
:END:
and then (possibly in a local variable) run
#+begin_src haskell :results output
:set prompt-cont ""
#+end_src
For reasons unknown I've had to include the :results output otherwise a cryptic error of "expecting a string" happens.
On a few other notes, haskell babel doesn't respond/care about the :session option, i.e., when you run a code block, a REPL *haskell* starts and that will be the sole REPL. Also, a haskell-mode started REPL doesn't play well with an existing org-mode initiated REPL, i.e., if you start a REPL from haskell-mode, it kills the original org-mode *haskkell*REPL, and any new attempt to run org-mode code blocks can't see this new, non-*haskell*REPL. Then if you kill the haskell-mode REPL and try to run org-mode blocks, you get
executing Haskell code block...
inferior-haskell-start-process: List contains a loop: ("--no-build" "--no-load" "--ghci-options=-ferror-spans" "--no-build" "--no-load" . #2)
... you're hosed -- and nothing seems to shake it, not any restart/refresh, nor killing, reloading the file, i.e., a complete restart of Emacs is necessary. Anyone knowing a better solution, please tells usses.
This is a GHCi issue.
The same error occurs when your code is copied directly into GHCi, which also gives a parse error when it encounters the new line after the equal sign. This first error isn't showing up here because org-babel only shows the value of the last expression (in this case, the error caused by the list comprehension).
I'm not entirely familiar with how Haskell-mode sends the code to GHCi, but it looks like it involves loading in the buffer into GHCi as a file, which may be why you didn't have this problem working from the hs file.
There are a few options to fix this, none of which are completely ideal:
Move some portion of the list into the first line (e.g. the first line could be pyth2 n = [).
Wrap the entire function definition with :{ and :}.
Write an Elisp function to modify what is being sent to GHCi and then changes it back after it is evaluated.
The first two options require you to format your code in a form that the GHCi will accept. In your example case, the first option may not be too bad, but this won't always be so trivial for all multi-line declarations (e.g. pattern-matching function declarations). The downside to the second option is that it requires adding brackets to the code that shouldn't be there in real source code.
To fix the issue of extraneous brackets being added, I've written an Elisp command (my-org-babel-execute-haskell-blocks) that places these brackets around code blocks that it finds, evaluates the region, and then deletes the brackets. Note that this function requires that blocks be separated from all other code with at least one empty line.
Calling my-org-babel-execute-haskell-blocks on your example declares the function without any errors.
EDIT: The previous function I gave failed to work on pattern matching declarations. I've rewritten the function to fix this issue as well as to be comment aware. This new function should be significantly more useful. However, it's worth noting that I didn't handle multi-line comments in a sophisticated manner, so code blocks with multi-line comments may not be wrapped properly.
(defun my-org-babel-execute-haskell-blocks ()
"Wraps :{ and :} around all multi-line blocks and then evaluates the source block.
Multi-line blocks are those where all non-indented, non-comment lines are declarations using the same token."
(interactive)
(save-excursion
;; jump to top of source block
(my-org-jump-to-top-of-block)
(forward-line)
;; get valid blocks
(let ((valid-block-start-ends (seq-filter #'my-haskell-block-valid-p (my-get-babel-blocks))))
(mapcar #'my-insert-haskell-braces valid-block-start-ends)
(org-babel-execute-src-block)
(mapcar #'my-delete-inserted-haskell-braces (reverse valid-block-start-ends)))))
(defun my-get-blocks-until (until-string)
(let ((block-start nil)
(block-list nil))
(while (not (looking-at until-string))
(if (looking-at "[[:space:]]*\n")
(when (not (null block-start))
(setq block-list (cons (cons block-start (- (point) 1))
block-list)
block-start nil))
(when (null block-start)
(setq block-start (point))))
(forward-line))
(when (not (null block-start))
(setq block-list (cons (cons block-start (- (point) 1))
block-list)))))
(defun my-get-babel-blocks ()
(my-get-blocks-until "#\\+end_src"))
(defun my-org-jump-to-top-of-block ()
(forward-line)
(org-previous-block 1))
(defun my-empty-line-p ()
(beginning-of-line)
(= (char-after) 10))
(defun my-haskell-type-declaration-line-p ()
(beginning-of-line)
(and (not (looking-at "--"))
(looking-at "^.*::.*$")))
(defun my-insert-haskell-braces (block-start-end)
(let ((block-start (car block-start-end))
(block-end (cdr block-start-end)))
(goto-char block-end)
(insert "\n:}")
(goto-char block-start)
(insert ":{\n")))
(defun my-delete-inserted-haskell-braces (block-start-end)
(let ((block-start (car block-start-end))
(block-end (cdr block-start-end)))
(goto-char block-start)
(delete-char 3)
(goto-char block-end)
(delete-char 3)))
(defun my-get-first-haskell-token ()
"Gets all consecutive non-whitespace text until first whitespace"
(save-excursion
(beginning-of-line)
(let ((starting-point (point)))
(re-search-forward ".*?[[:blank:]\n]")
(goto-char (- (point) 1))
(buffer-substring-no-properties starting-point (point)))))
(defun my-haskell-declaration-line-p ()
(beginning-of-line)
(or (looking-at "^.*=.*$") ;; has equals sign
(looking-at "^.*\n[[:blank:]]*|")
(looking-at "^.*where[[:blank:]]*$")))
(defun my-haskell-block-valid-p (block-start-end)
(let ((block-start (car block-start-end))
(block-end (cdr block-start-end))
(line-count 0))
(save-excursion
(goto-char block-start)
(let ((token 'nil)
(is-valid t))
;; eat top comments
(while (or (looking-at "--")
(looking-at "{-"))
(forward-line))
(when (my-haskell-type-declaration-line-p)
(progn
(setq token (my-get-first-haskell-token)
line-count 1)
(forward-line)))
(while (<= (point) block-end)
(let ((current-token (my-get-first-haskell-token)))
(cond ((string= current-token "") ; line with indentation
(when (null token) (setq is-valid nil))
(setq line-count (+ 1 line-count)))
((or (string= (substring current-token 0 2) "--") ;; skip comments
(string= (substring current-token 0 2) "{-"))
'())
((and (my-haskell-declaration-line-p)
(or (null token) (string= token current-token)))
(setq token current-token
line-count (+ 1 line-count)))
(t (setq is-valid nil)
(goto-char (+ 1 block-end))))
(forward-line)))
(and is-valid (> line-count 1))))))

Convert command line arguments to list in Emacs Lisp

In an automated Emacs Lisp --batch/--script script I need to process the command line arguments given to the script.
I've gotten as far as getting the arguments into a list of the the form:
("--aaa=bbb" "--ccc=ddd=eee" "--blah")
Now, I need to convert them to a list of the form:
(("aaa" "bbb") ("ccc" "ddd=eee") ("blah"))
In Python I'd write something like;
output = []
for v in input:
output.append(v[2:].split("=", 1))
But have been unable to convert that code to Emacs Lisp. I found Elisp split-string function to split a string by . character but wasn't able to figure out how to make it only split on the first equals.
I was heading down a route of using (substring "abcdefg" x x) with (search) from the cl package but it felt like there should be a better way? I think also want to use (mapc '<function> input) where function does the v[2:].split("=",1) part.
You can use split-string. See the following code example.
(setq cmd-line '("--aaa=bbb" "--ccc=ddd=eee" "--blah"))
(setq cmd-line (mapcar (lambda (argstr)
(when (string-match "^--" argstr)
(split-string (substring argstr 2) "=")))
cmd-line))
The output is (("aaa" "bbb") ("ccc" "ddd" "eee") ("blah")).
That is not exactly what you want because of "eee". Maybe you can use that and just neglect "eee".
If the "eee" is really a problem a small modification helps:
(setq cmd-line '("--aaa=bbb" "--ccc=ddd=eee" "--blah"))
(setq cmd-line (mapcar (lambda (arg)
(when (string-match "^--" arg)
(setq arg (split-string (substring arg 2) "="))
(if (cdr arg)
(setcdr (cdr arg) nil))
arg))
cmd-line))
The output is:
(("aaa" "bbb") ("ccc" "ddd") ("blah"))
Variant for the new requirement in the question:
(setq cmd-line '("--aaa=bbb" "--ccc=ddd=eee" "--blah"))
(setq cmd-line (mapcar (lambda (arg)
(when (string-match "^--\\([^=]*\\)\\(?:=\\(.*\\)\\)?" arg)
(let ((opt (match-string 1 arg))
(val (match-string 2 arg)))
(if val
(list opt val)
(list opt)))))
cmd-line))
The output is:
(("aaa" "bbb") ("ccc" "ddd=eee") ("blah"))

Customize Elisp plist indentation

I don't like how plists are indented in Elisp.
;; current desired Python (for comparison)
;; '(a 1 '(a 1 {'a': 1,
;; b 2 b 2 'b': 2,
;; c 3) c 3) 'c': 3}
Tried on M-x emacs-version 24.3.1, ran emacs -Q, typed the plist and pressed C-x h C-M-\.
This indentation makes sense when it isn't a list:
(mapcar (lambda (x) (x + 1))
'(1 2 3 4))
How do I change formatting settings so that only plists (or, if that's impossible, all quoted lists) have the desired rectangular indentation, but indentation of everything else stays the same? I need this stored locally in an .el file, so that when I edit this file, it is indented as desired, but this behavior doesn't end up anywhere else.
Found it:
(setq lisp-indent-function 'common-lisp-indent-function)
Here's a sample file:
(setq x '(a 1
b 2
c 3))
;;; Local Variables:
;;; lisp-indent-function: common-lisp-indent-function
;;; End:
I'll just dump my whole indentation config here:
(setq lisp-indent-function 'common-lisp-indent-function)
(put 'cl-flet 'common-lisp-indent-function
(get 'flet 'common-lisp-indent-function))
(put 'cl-labels 'common-lisp-indent-function
(get 'labels 'common-lisp-indent-function))
(put 'if 'common-lisp-indent-function 2)
(put 'dotimes-protect 'common-lisp-indent-function
(get 'when 'common-lisp-indent-function))
You can fix this (in my opinion) bug by overriding lisp-indent-function. The original source of the hack was this Github Gist, which was referenced with some more explanation from this Emacs Stack Exchange answer.
However, I was very uncomfortable overriding a core function like this. For one, it's very opaque—how is a reader supposed to tell what is changed? And worse—what if the official definition of lisp-indent-function changed in the future? How would I know that I needed to update my hack?
As a response, I created the library el-patch, which is specifically designed to address this problem. After installing the package, you can override lisp-indent-function as follows:
(el-patch-defun lisp-indent-function (indent-point state)
"This function is the normal value of the variable `lisp-indent-function'.
The function `calculate-lisp-indent' calls this to determine
if the arguments of a Lisp function call should be indented specially.
INDENT-POINT is the position at which the line being indented begins.
Point is located at the point to indent under (for default indentation);
STATE is the `parse-partial-sexp' state for that position.
If the current line is in a call to a Lisp function that has a non-nil
property `lisp-indent-function' (or the deprecated `lisp-indent-hook'),
it specifies how to indent. The property value can be:
* `defun', meaning indent `defun'-style
(this is also the case if there is no property and the function
has a name that begins with \"def\", and three or more arguments);
* an integer N, meaning indent the first N arguments specially
(like ordinary function arguments), and then indent any further
arguments like a body;
* a function to call that returns the indentation (or nil).
`lisp-indent-function' calls this function with the same two arguments
that it itself received.
This function returns either the indentation to use, or nil if the
Lisp function does not specify a special indentation."
(el-patch-let (($cond (and (elt state 2)
(el-patch-wrap 1 1
(or (not (looking-at "\\sw\\|\\s_"))
(looking-at ":")))))
($then (progn
(if (not (> (save-excursion (forward-line 1) (point))
calculate-lisp-indent-last-sexp))
(progn (goto-char calculate-lisp-indent-last-sexp)
(beginning-of-line)
(parse-partial-sexp (point)
calculate-lisp-indent-last-sexp 0 t)))
;; Indent under the list or under the first sexp on the same
;; line as calculate-lisp-indent-last-sexp. Note that first
;; thing on that line has to be complete sexp since we are
;; inside the innermost containing sexp.
(backward-prefix-chars)
(current-column)))
($else (let ((function (buffer-substring (point)
(progn (forward-sexp 1) (point))))
method)
(setq method (or (function-get (intern-soft function)
'lisp-indent-function)
(get (intern-soft function) 'lisp-indent-hook)))
(cond ((or (eq method 'defun)
(and (null method)
(> (length function) 3)
(string-match "\\`def" function)))
(lisp-indent-defform state indent-point))
((integerp method)
(lisp-indent-specform method state
indent-point normal-indent))
(method
(funcall method indent-point state))))))
(let ((normal-indent (current-column))
(el-patch-add
(orig-point (point))))
(goto-char (1+ (elt state 1)))
(parse-partial-sexp (point) calculate-lisp-indent-last-sexp 0 t)
(el-patch-swap
(if $cond
;; car of form doesn't seem to be a symbol
$then
$else)
(cond
;; car of form doesn't seem to be a symbol, or is a keyword
($cond $then)
((and (save-excursion
(goto-char indent-point)
(skip-syntax-forward " ")
(not (looking-at ":")))
(save-excursion
(goto-char orig-point)
(looking-at ":")))
(save-excursion
(goto-char (+ 2 (elt state 1)))
(current-column)))
(t $else))))))
Here is another less heavyweight solution, based on emacsql-fix-vector-indentation. An advice around calculate-lisp-indent is sufficient.
This only works for plists that use keywords as keys, but that covers a majority of plists. To make this work on quoted lists instead, you could change the looking-at regexp to detect the ' or "`", but that will not cover, say, a nested list.
This can further be packaged up into a minor mode if there is a need to turn it off.
(defun my/inside-plist? ()
"Is point situated inside a plist?
We determine a plist to be a list that starts with a keyword."
(let ((start (point)))
(save-excursion
(beginning-of-defun)
(let ((sexp (nth 1 (parse-partial-sexp (point) start))))
(when sexp
(setf (point) sexp)
(looking-at (rx "(" (* (syntax whitespace)) ":")))))))
(define-advice calculate-lisp-indent (:around (func &rest args)
plist)
(if (save-excursion
(beginning-of-line)
(my/inside-plist?))
(let ((lisp-indent-offset 1))
(apply func args))
(apply func args)))

Emacs cond, possible to have things happen between clauses?

I programmed some months ago some code with a lot of if statements. If region-active-p, if beginning-of-line, those kind of things.
Having learned about the cond lisp, I was wondering if I could improve my code a lot.
The problem is that this cond is only doing things when "true" as far as I see it, while I actually need the move back-to-indentation in between these checks.
In order to properly skip the last clause, I even have to set variable values.
(defun uncomment-mode-specific ()
"Uncomment region OR uncomment beginning of line comment OR uncomment end"
(interactive)
(let ((scvar 0) (scskipvar 0))
(save-excursion
(if (region-active-p)
(progn (uncomment-region (region-beginning) (region-end))
(setq scskipvar 1))
(back-to-indentation)) ; this is that "else" part that doesn't fit in cond
(while (string= (byte-to-string (following-char)) comment-start)
(delete-char 1)
(setq scskipvar 1))
(indent-for-tab-command)
(when (= scskipvar 0)
(search-forward comment-start nil t)
(backward-char 1)
(kill-line))
)))
)
So basically my question is, I would kind of like to have some consequences of not giving "true" to a clause, before the check of another clause. Is this possible? If not, what would be the best thing to do?
EDIT: Since we are using this as the example case for a solution, I wrote it down so it is easier to understand.
If region is active, remove comments from region. If not, move point to intendation.
For as long as the following character is a comment character, delete it. Afterwards, indent this line.
If it didn't do any of the above, search forward for a comment character, and kill that line.
(defun delete-on-this-line (regex)
(replace-regexp regex "" nil (line-beginning-position) (line-end-position)))
(defun delete-leading-comment-chars ()
(delete-on-this-line (eval `(rx bol (* space) (group (+ ,comment-start))))))
(defun delete-trailing-comment-chars ()
(delete-on-this-line (eval `(rx (group (+ ,comment-end)) (* space) eol))))
(defun delete-trailing-comment ()
(delete-on-this-line (eval `(rx (group (+ ,comment-start) (* anything) eol)))))
(defun uncomment-dwim ()
(interactive)
(save-excursion
(if (region-active-p)
(uncomment-region (region-beginning) (region-end))
(or (delete-leading-comment-chars)
(delete-trailing-comment-chars)
(delete-trailing-comment)))))
Edit: A little explanation:
It's a lot easier to do regex replacements than manage loops to do deletion, so that gets rid of the state. And the steps are all mutually exclusive, so you can just use or for each option.
The rx macro is a little DSL that compiles down to valid regexes, and it's also amenable to lispy syntax transforms, so I can dynamically build a regex using the comment chars for the current mode.
(defmacro fcond (&rest body)
(labels ((%substitute-last-or-fail
(new old seq)
(loop for elt on seq
nconc
(if (eql (car elt) old)
(when (cdr elt)
(error "`%S' must be the last experssion in the clause"
(car elt)))
(list new)
(list (car elt))))))
(loop with matched = (gensym)
with catcher = (gensym)
for (head . rest) in body
collect
`(when (or ,head ,matched)
(setq ,matched t)
,#(%substitute-last-or-fail `(throw ',catcher nil) 'return rest))
into clauses
finally
(return `(let (,matched) (catch ',catcher ,#clauses))))))
(macroexpand '(fcond
((= 1 2) (message "1 = 2"))
((= 1 1) (message "1 = 1"))
((= 1 3) (message "1 = 3") return)
((= 1 4) (message "1 = 4"))))
(let (G36434)
(catch (quote G36435)
(when (or (= 1 2) G36434)
(setq G36434 t)
(message "1 = 2"))
(when (or (= 1 1) G36434)
(setq G36434 t)
(message "1 = 1"))
(when (or (= 1 3) G36434)
(setq G36434 t)
(message "1 = 3")
(throw (quote G36435) nil))
(when (or (= 1 4) G36434)
(setq G36434 t)
(message "1 = 4"))))
Here's something quick to do, what I think you may be after, i.e. something that would mimic the behaviour switch in C.
The idea is that all clauses are tested sequentially for equality, and if one matches, then all following clauses are executed, until the return keyword (it would be break in C, but Lisp uses return for the similar purpose in the loop, so I thought that return would be better). The code above thus will print:
1 = 1
1 = 3
Technically, this is not how switch works in C, but it will produce the same effect.
One thing I did here for simplicity, which you want to avoid / solve differently - the use of return keyword, you probably want to impose stricter rules on how it should be searched for.
cond
Cond evaluates a series of conditions in a list, each item in a list can be a condition, and then executable instructions.
The example in the Emacs Lisp manual is adequate to demonstrate how it works, I've annotated it here to help you understand how it works.
(cond ((numberp x) x) ;; is x a number? return x
((stringp x) x) ;; is x a string? return x
((bufferp x) ;; is x a buffer?
(setq temporary-hack x) ;; set temporary-hack to buffer x
(buffer-name x)) ;; return the buffer-name for buffer x
((symbolp x) (symbol-value x))) ;; is x a symbol? return the value of x
Each part of the condition can be evaluated any way you like, the fact x above is in each condition is coincidental.
For example:
(cond ((eq 1 2) "Omg equality borked!") ;; Will never be true
(t "default")) ;; always true
So comparisons with switch are a bit limited, it's essentially a list of if statements, that executes/returns the first true condition's body list.
Hopefully this helps you understand cond a bit better.
(cond (condition body ... ) ;; execute body of 1st passing
(condition body ... ) ;; condition and return result
(condition body ... ) ;; of the final evaluation.
;; etc
)
OR
You can do things similar to switch with OR, depending on how you structure the code.
This isn't functional style, because it relies on side-effects to do what you want, then returns a boolean value for flow control, here's an example in pseudo lisp.
(or)
(or
(lambda() (do something)
(evaluate t or nil) ; nil to continue; t to quit.
)
(lambda() (do something)
(evaluate t or nil) ; nil to continue; t to quit.
)
(lambda() (do something)
(evaluate t or nil) ; nil to continue; t to quit.
)
(lambda() (do something)
(evaluate t or nil) ; nil to continue; t to quit.
)
)
Here's working example of a switch like structure using or
(or
(when (= 1 1)
(progn
(insert "hello\n")
nil))
(when (= 1 2) ;; condition fails.
(progn
(insert "hello\n")
nil)) ;; returns false (nil)
(when (= 1 1)
(progn
(insert "hello\n")
t)) ;; returns true, so we bail.
(when (= 1 1)
(progn
(insert "hello\n")
nil))
)
Inserts :
hello
hello
(and)
The and operator (not just in Lisp) is also very useful, instead of evaluating everything until true, it evaluates conditions that are true, until a false is evaluated.
Both or & and can be used to build useful logic trees.
This is how I did it now according to Chris' idea that breaking it down into seperate functions would make it easier.
EDIT: Now also applied the or knowledge gained in this thread gained from Slomojo (no more variables!)
(defun sc-uncomment ()
(interactive)
(or
(if (region-active-p)
(uncomment-region (region-beginning) (region-end))
(back-to-indentation)
nil)
(if (string= (byte-to-string (following-char)) comment-start)
(sc-check-start)
(sc-end))))
(defun sc-check-start ()
(interactive)
(while (string= (byte-to-string (following-char)) comment-start)
(delete-char 1))
)
(defun sc-end ()
(interactive)
(search-forward comment-start nil t)
(backward-char 1)
(kill-line))
)

In elisp, how to evaluate a string of "var=value\n..." into lisp variables of the same name?

An mplayer tool (midentify) outputs "shell-ready" lines intended to be evaluated by a bash/sh/whatever interpreter.
How can I assign these var-names to their corresponding values as elisp var-names in emacs?
The data is in a string (via shell-command-to-string)
Here is the data
ID_AUDIO_ID=0
ID_FILENAME=/home/axiom/abc.wav
ID_DEMUXER=audio
ID_AUDIO_FORMAT=1
ID_AUDIO_BITRATE=512000
ID_AUDIO_RATE=0
ID_AUDIO_NCH=1
ID_LENGTH=3207.00
ID_SEEKABLE=1
ID_CHAPTERS=0
ID_AUDIO_BITRATE=512000
ID_AUDIO_RATE=32000
ID_AUDIO_NCH=1
ID_AUDIO_CODEC=pcm
ID_EXIT=EOF
Here's a routine that takes a string containing midentify output, and returns an association list of the key-value pairs (which is safer than setting Emacs variables willy-nilly). It also has the advantage that it parses numeric values into actual numbers:
(require 'cl) ; for "loop"
(defun midentify-output-to-alist (str)
(setq str (replace-regexp-in-string "\n+" "\n" str))
(setq str (replace-regexp-in-string "\n+\\'" "" str))
(loop for index = 0 then (match-end 0)
while (string-match "^\\(?:\\([A-Z_]+\\)=\\(?:\\([0-9]+\\(?:\\.[0-9]+\\)?\\)\\|\\(.*\\)\\)\\|\\(.*\\)\\)\n?" str index)
if (match-string 4 str)
do (error "Invalid line: %s" (match-string 4 str))
collect (cons (match-string 1 str)
(if (match-string 2 str)
(string-to-number (match-string 2 str))
(match-string 3 str)))))
You'd use this function like so:
(setq alist (midentify-output-to-alist my-output))
(if (assoc "ID_LENGTH" alist)
(setq id-length (cdr (assoc "ID_LENGTH" alist)))
(error "Didn't find an ID_LENGTH!"))
EDIT: Modified function to handle blank lines and trailing newlines correctly.
The regexp is indeed a beast; Emacs regexps are not known for their easiness on the eyes. To break it down a bit:
The outermost pattern is ^(?:valid-line)|(.*). It tries to match a valid line, or else matches the entire line (the .*) in match-group 4. If (match-group 4 str) is not nil, that indicates that an invalid line was encountered, and an error is raised.
valid-line is (word)=(?:(number)|(.*)). If this matches, then the name part of the name-value pair is in match-string 1, and if the rest of the line matches a number, then the number is in match-string 2, otherwise the entire rest of the line is in match-string 3.
There's probably a better way but this should do it:
(require 'cl)
(let ((s "ID_AUDIO_ID=0
ID_FILENAME=/home/axiom/abc.wav
ID_DEMUXER=audio
ID_AUDIO_FORMAT=1
ID_AUDIO_BITRATE=512000
ID_AUDIO_RATE=0
ID_AUDIO_NCH=1
ID_LENGTH=3207.00
ID_SEEKABLE=1
ID_CHAPTERS=0
ID_AUDIO_BITRATE=512000
ID_AUDIO_RATE=32000
ID_AUDIO_NCH=1
ID_AUDIO_CODEC=pcm
ID_EXIT=EOF"))
(loop for p in (split-string s "\n")
do
(let* ((elements (split-string p "="))
(key (elt elements 0))
(value (elt elements 1)))
(set (intern key) value))))
Here's a function you can run on the output buffer:
(defun set-variables-from-shell-assignments ()
(goto-char (point-min))
(while (< (point) (point-max))
(and (looking-at "\\([A-Z_]+\\)=\\(.*\\)$")
(set (intern (match-string 1)) (match-string 2)))
(forward-line 1)))
I don't think regexp is what really need. You need to split your string by \n and =, so you just say exactly the same to interpreter.
I think you can also use intern to get symbol from string(and set variables). I use it for the first time, so comment here if i am wrong. Anyways, if list is what you want, just remove top-level mapcar.
(defun set=(str)
(mapcar (lambda(arg)
(set
(intern (car arg))
(cadr arg)))
(mapcar (lambda(arg)
(split-string arg "=" t))
(split-string
str
"\n" t))))
(set=
"ID_AUDIO_ID=0
ID_FILENAME=/home/axiom/abc.wav
ID_DEMUXER=audio
ID_AUDIO_FORMAT=1
ID_AUDIO_BITRATE=512000
ID_AUDIO_RATE=0
ID_AUDIO_NCH=1
ID_LENGTH=3207.00
ID_SEEKABLE=1
ID_CHAPTERS=0
ID_AUDIO_BITRATE=512000
ID_AUDIO_RATE=32000
ID_AUDIO_NCH=1
ID_AUDIO_CODEC=pcm
ID_EXIT=EOF")