Re: [PATCH 4/5] emacs: Streaming S-expression parser

Subject: Re: [PATCH 4/5] emacs: Streaming S-expression parser

Date: Tue, 21 May 2013 20:03:40 -0400

To: Mark Walters, notmuch@notmuchmail.org

Cc:

From: Austin Clements


On Tue, 21 May 2013, Mark Walters <markwalters1009@gmail.com> wrote:
> Hi
>
> This patch looks good to me. Some minor comments below.

Some minor replies below.

In building some other code on top of this, I found an interesting (but
easy to fix) interface bug.  Currently, the interface is designed as if
it doesn't matter what buffer these functions are called from, however,
because they move point and expect this point motion to persist, it's
actually not safe to call this interface unless the caller is in the
right buffer anyway.  For example, if the buffer is selected in a
window, the with-current-buffer in the parser functions will actually
move a *temporary* point, meaning that the only way the caller can
discover the new point is to first select the buffer for itself.  I can
think of two solutions: 1) maintain our own mark for the parser's
current position or 2) tweak the doc strings and code so that it reads
from the current buffer.  1 keeps the interface the way it's currently
documented, but complicates the parser implementation and interface and
doesn't simplify the caller.  2 simplifies the parser and it turns out
all callers already satisfy the requirement.

> Best wishes
>
> Mark
>
>
> On Sat, 18 May 2013, Austin Clements <amdragon@MIT.EDU> wrote:
>> This provides the same interface as the streaming JSON parser, but
>> reads S-expressions incrementally.  The only difference is that the
>> `notmuch-sexp-parse-partial-list' helper does not handle interleaved
>> error messages (since we now have the ability to separate these out at
>> the invocation level), so it no longer takes an error function and
>> does not need to do the horrible resynchronization that the JSON
>> parser had to.
>>
>> Some implementation improvements have been made over the JSON parser.
>> This uses a vector instead of a list for the parser data structure,
>> since this allows faster access to elements (and modern versions of
>> Emacs handle storage of small vectors efficiently).  Private functions
>> follow the "prefix--name" convention.  And the implementation is much
>> simpler overall because S-expressions are much easier to parse.
>> ---
>>  emacs/Makefile.local    |    1 +
>>  emacs/notmuch-parser.el |  212 +++++++++++++++++++++++++++++++++++++++++++++++
>>  2 files changed, 213 insertions(+)
>>  create mode 100644 emacs/notmuch-parser.el
>>
>> diff --git a/emacs/Makefile.local b/emacs/Makefile.local
>> index 456700a..a910aff 100644
>> --- a/emacs/Makefile.local
>> +++ b/emacs/Makefile.local
>> @@ -3,6 +3,7 @@
>>  dir := emacs
>>  emacs_sources := \
>>  	$(dir)/notmuch-lib.el \
>> +	$(dir)/notmuch-parser.el \
>>  	$(dir)/notmuch.el \
>>  	$(dir)/notmuch-query.el \
>>  	$(dir)/notmuch-show.el \
>> diff --git a/emacs/notmuch-parser.el b/emacs/notmuch-parser.el
>> new file mode 100644
>> index 0000000..1b7cf64
>> --- /dev/null
>> +++ b/emacs/notmuch-parser.el
>> @@ -0,0 +1,212 @@
>> +;; notmuch-parser.el --- streaming S-expression parser
>> +;;
>> +;; Copyright © Austin Clements
>> +;;
>> +;; This file is part of Notmuch.
>> +;;
>> +;; Notmuch is free software: you can redistribute it and/or modify it
>> +;; under the terms of the GNU General Public License as published by
>> +;; the Free Software Foundation, either version 3 of the License, or
>> +;; (at your option) any later version.
>> +;;
>> +;; Notmuch is distributed in the hope that it will be useful, but
>> +;; WITHOUT ANY WARRANTY; without even the implied warranty of
>> +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>> +;; General Public License for more details.
>> +;;
>> +;; You should have received a copy of the GNU General Public License
>> +;; along with Notmuch.  If not, see <http://www.gnu.org/licenses/>.
>> +;;
>> +;; Authors: Austin Clements <aclements@csail.mit.edu>
>> +
>> +(require 'cl)
>> +
>> +(defun notmuch-sexp-create-parser (buffer)
>> +  "Return a streaming S-expression parser that reads from BUFFER.
>> +
>> +This parser is designed to incrementally read an S-expression
>> +whose structure is known to the caller.  Like a typical
>> +S-expression parsing interface, it provides a function to read a
>> +complete S-expression from the input.  However, it extends this
>> +with an additional function that requires the next value in the
>> +input to be a list and descends into it, allowing its elements to
>> +be read one at a time or further descended into.  Both functions
>> +can return 'retry to indicate that not enough input is available.
>> +
>> +The parser always consumes input from BUFFER's point.  Hence, the
>> +caller is allowed to delete any data before point and may
>> +resynchronize after an error by moving point."
>> +
>> +  (vector 'notmuch-sexp-parser
>> +	  buffer
>> +	  ;; List depth
>> +	  0
>> +	  ;; Partial parse position marker
>> +	  nil
>> +	  ;; Partial parse state
>> +	  nil))
>> +
>> +(defmacro notmuch-sexp--buffer (sp)        `(aref ,sp 1))
>> +(defmacro notmuch-sexp--depth (sp)         `(aref ,sp 2))
>> +(defmacro notmuch-sexp--partial-pos (sp)   `(aref ,sp 3))
>> +(defmacro notmuch-sexp--partial-state (sp) `(aref ,sp 4))
>
> Why the double hyphen --? Is it a name-space or some convention?

More specifically, this seems to be the most common Elisp convention for
indicating private symbols.

>> +(defun notmuch-sexp-read (sp)
>> +  "Consume and return the value at point in SP's buffer.
>> +
>> +Returns 'retry if there is insufficient input to parse a complete
>> +value (though it may still move point over whitespace).  If the
>> +parser is currently inside a list and the next token ends the
>> +list, this moves point just past the terminator and returns 'end.
>> +Otherwise, this moves point to just past the end of the value and
>> +returns the value."
>> +
>> +  (with-current-buffer (notmuch-sexp--buffer sp)
>> +    (skip-chars-forward " \n\r\t")
>> +    (cond ((eobp) 'retry)
>> +	  ((= (char-after) ?\))
>> +	   ;; We've reached the end of a list
>> +	   (if (= (notmuch-sexp--depth sp) 0)
>> +	       ;; .. but we weren't in a list.  Let read signal the
>> +	       ;; error.
>> +	       (read (current-buffer))
>
> Why is good for read to signal the error rather than us doing it?

This ensures the syntax error handling and signal behavior of
notmuch-sexp-read is identical in every way to a regular read call.
Maybe the comment should read "Let read signal the error like we do in
all other code paths."?

>> +	     ;; Go up a level and return an end token
>> +	     (decf (notmuch-sexp--depth sp))
>> +	     (forward-char)
>> +	     'end))
>> +	  ((= (char-after) ?\()
>> +	   ;; We're at the beginning of a list.  If we haven't started
>> +	   ;; a partial parse yet, attempt to read the list in its
>> +	   ;; entirety.  If this fails, or we've started a partial
>> +	   ;; parse, extend the partial parse to figure out when we
>> +	   ;; have a complete list.
>> +	   (catch 'return
>> +	     (when (null (notmuch-sexp--partial-state sp))
>> +	       (let ((start (point)))
>> +		 (condition-case nil
>> +		     (throw 'return (read (current-buffer)))
>> +		   (end-of-file (goto-char start)))))
>> +	     ;; Extend the partial parse
>> +	     (let (is-complete)
>> +	       (save-excursion
>> +		 (let* ((new-state (parse-partial-sexp
>> +				    (or (notmuch-sexp--partial-pos sp) (point))
>> +				    (point-max) 0 nil
>> +				    (notmuch-sexp--partial-state sp)))
>> +			;; A complete value is available if we've
>> +			;; reached depth 0.
>> +			(depth (first new-state)))
>> +		   (assert (>= depth 0))
>> +		   (if (= depth 0)
>> +		       ;; Reset partial parse state
>> +		       (setf (notmuch-sexp--partial-state sp) nil
>> +			     (notmuch-sexp--partial-pos sp) nil
>> +			     is-complete t)
>> +		     ;; Update partial parse state
>> +		     (setf (notmuch-sexp--partial-state sp) new-state
>> +			   (notmuch-sexp--partial-pos sp) (point-marker)))))
>> +	       (if is-complete
>> +		   (read (current-buffer))
>> +		 'retry))))
>> +	  (t
>> +	   ;; Attempt to read a non-compound value
>> +	   (let ((start (point)))
>> +	     (condition-case nil
>> +		 (let ((val (read (current-buffer))))
>> +		   ;; We got what looks like a complete read, but if
>> +		   ;; we reached the end of the buffer in the process,
>> +		   ;; we may not actually have all of the input we
>> +		   ;; need (unless it's a string, which is delimited).
>> +		   (if (or (stringp val) (not (eobp)))
>> +		       val
>> +		     ;; We can't be sure the input was complete
>> +		     (goto-char start)
>> +		     'retry))
>> +	       (end-of-file
>> +		(goto-char start)
>> +		'retry)))))))
>> +
>> +(defun notmuch-sexp-begin-list (sp)
>> +  "Parse the beginning of a list value and enter the list.
>> +
>> +Returns 'retry if there is insufficient input to parse the
>> +beginning of the list.  If this is able to parse the beginning of
>> +a list, it moves point past the token that opens the list and
>> +returns t.  Later calls to `notmuch-sexp-read' will return the
>> +elements inside the list.  If the input in buffer is not the
>> +beginning of a list, throw invalid-read-syntax."
>> +
>> +  (with-current-buffer (notmuch-sexp--buffer sp)
>> +    (skip-chars-forward " \n\r\t")
>> +    (cond ((eobp) 'retry)
>> +	  ((= (char-after) ?\()
>> +	   (forward-char)
>> +	   (incf (notmuch-sexp--depth sp))
>> +	   t)
>> +	  (t
>> +	   ;; Skip over the bad character like `read' does
>> +	   (forward-char)
>> +	   (signal 'invalid-read-syntax (list (string (char-before))))))))
>> +
>> +(defun notmuch-sexp-eof (sp)
>> +  "Signal an error if there is more data in SP's buffer.
>> +
>> +Moves point to the beginning of any trailing data or to the end
>> +of the buffer if there is only trailing whitespace."
>> +
>> +  (with-current-buffer (notmuch-sexp--buffer sp)
>> +    (skip-chars-forward " \n\r\t")
>> +    (unless (eobp)
>> +      (error "Trailing garbage following expression"))))
>> +
>> +(defvar notmuch-sexp--parser nil
>> +  "The buffer-local notmuch-sexp-parser instance.
>> +
>> +Used by `notmuch-sexp-parse-partial-list'.")
>> +
>> +(defvar notmuch-sexp--state nil
>> +  "The buffer-local `notmuch-sexp-parse-partial-list' state.")
>> +
>> +(defun notmuch-sexp-parse-partial-list (result-function result-buffer)
>> +  "Incrementally parse an S-expression list from the current buffer.
>> +
>> +This function consume an S-expression list from the current
>
> consumes

Oops, yes.

>> +buffer, applying RESULT-FUNCTION in RESULT-BUFFER to each
>> +complete value in the list.  It operates incrementally and should
>> +be called whenever the input buffer has been extended with
>> +additional data.  The caller just needs to ensure it does not
>> +move point in the input buffer."
>> +
>> +  ;; Set up the initial state
>> +  (unless (local-variable-p 'notmuch-sexp--parser)
>> +    (set (make-local-variable 'notmuch-sexp--parser)
>> +	 (notmuch-sexp-create-parser (current-buffer)))
>> +    (set (make-local-variable 'notmuch-sexp--state) 'begin))
>> +  (let (done)
>> +    (while (not done)
>> +      (case notmuch-sexp--state
>> +	(begin
>> +	 ;; Enter the list
>> +	 (if (eq (notmuch-sexp-begin-list notmuch-sexp--parser) 'retry)
>> +	     (setq done t)
>> +	   (setq notmuch-sexp--state 'result)))
>> +	(result
>> +	 ;; Parse a result
>> +	 (let ((result (notmuch-sexp-read notmuch-sexp--parser)))
>> +	   (case result
>> +	     (retry (setq done t))
>> +	     (end   (setq notmuch-sexp--state 'end))
>> +	     (t     (with-current-buffer result-buffer
>> +		      (funcall result-function result))))))
>> +	(end
>> +	 ;; Any trailing data is unexpected
>> +	 (notmuch-sexp-eof notmuch-sexp--parser)
>> +	 (setq done t)))))
>> +  ;; Clear out what we've parsed
>> +  (delete-region (point-min) (point)))
>> +
>> +(provide 'notmuch-parser)
>> +
>> +;; Local Variables:
>> +;; byte-compile-warnings: (not cl-functions)
>> +;; End:
>> -- 
>> 1.7.10.4

Thread: