This commit is contained in:
Adam Porter 2020-12-20 08:01:17 -06:00
parent f597aa714f
commit cd8014ae41

157
NOTES.org
View file

@ -487,6 +487,9 @@ Rather than searching for generic timestamp regexps, we could build a regexp bas
**** TODO Tests
*** TODO ~scheduled~
:LOGBOOK:
CLOCK: [2020-12-20 Sun 07:01]--[2020-12-20 Sun 07:58] => 0:57
:END:
**** TODO Tests
@ -496,6 +499,160 @@ Rather than searching for generic timestamp regexps, we could build a regexp bas
**** TODO Tests
**** DONE Benchmark
[2020-12-20 Sun 07:51] This is a great improvement. I tested it on the ~scheduled~ predicate too, but it doesn't make nearly as big a difference, because just searching for the ~SCHEDULED:~ prefix avoids testing most timestamps.
#+BEGIN_SRC elisp
(bench-multi-lexical :times 1 :ensure-equal t
:forms (("unoptimized"
(progn
(org-ql-defpred (ts ts-active ts-a ts-inactive ts-i)
(&key from to _on regexp (match-group 0) (limit (org-entry-end-position)))
;; NOTE: Arguments to this predicate are pre-processed in `org-ql--normalize-query'.
;; The underscore before `on' prevents "unused lexical variable" warnings due to the
;; pre-processing converting that argument to FROM and TO. The `regexp' argument is
;; also provided by the pre-processing and is not to be given by the user. FROM and
;; TO are actually expected to be `ts' structs. The docstring is written for users.
"Return non-nil if current entry has a timestamp in given period.
If no arguments are specified, return non-nil if entry has any
timestamp.
If FROM, return non-nil if entry has a timestamp on or after
FROM.
If TO, return non-nil if entry has a timestamp on or before TO.
If ON, return non-nil if entry has a timestamp on date ON.
FROM, TO, and ON should be either `ts' structs, or strings
parseable by `parse-time-string' which may omit the time value.
TYPE may be `active' to match active timestamps, `inactive' to
match inactive ones, or `both' / nil to match both types.
LIMIT bounds the search for the timestamp REGEXP. It defaults to
the end of the entry, i.e. the position returned by
`org-entry-end-position', but for certain searches it should be
bound to a different positiion, e.g. for planning lines, the end
of the line after the heading."
;; MAYBE: Define active/inactive ones separately?
:normalizers ((`(,(or 'ts-active 'ts-a) . ,rest) `(ts :type active ,@rest))
(`(,(or 'ts-inactive 'ts-i) . ,rest) `(ts :type inactive ,@rest)))
:preambles ((`(,predicate-names . ,rest)
(list :regexp (pcase (plist-get rest :type)
((or 'nil 'both) org-tsr-regexp-both)
('active org-tsr-regexp)
('inactive org-ql-tsr-regexp-inactive))
;; Predicate needs testing only when args are present.
:query (-let (((&keys :from :to :on) rest))
;; FIXME: This used to be (when (or from to on) query), but that doesn't seem right, so I
;; changed it to this if, and the tests pass either way. Might deserve a little scrutiny.
(if (or from to on)
query
t)))))
;; TODO: DRY this with the clocked predicate.
:body
(cl-macrolet ((next-timestamp ()
`(when (re-search-forward regexp limit t)
(ts-parse-org (match-string match-group))))
(test-timestamps (pred-form)
`(cl-loop for next-ts = (next-timestamp)
while next-ts
thereis ,pred-form)))
(save-excursion
(cond ((not (or from to)) (re-search-forward regexp limit t))
((and from to) (test-timestamps (ts-in from to next-ts)))
(from (test-timestamps (ts<= from next-ts)))
(to (test-timestamps (ts<= next-ts to)))))))
(setf org-ql-cache (make-hash-table :weakness 'key))
(org-ql-select (org-agenda-files)
'(ts :from "2020-01-01" :to "2020-12-31")
:action '(substring-no-properties (org-get-heading t t)))))
("optimized"
(progn
(org-ql-defpred (ts ts-active ts-a ts-inactive ts-i)
(&key from to _on regexp (match-group 0) (limit (org-entry-end-position)))
;; NOTE: Arguments to this predicate are pre-processed in `org-ql--normalize-query'.
;; The underscore before `on' prevents "unused lexical variable" warnings due to the
;; pre-processing converting that argument to FROM and TO. The `regexp' argument is
;; also provided by the pre-processing and is not to be given by the user. FROM and
;; TO are actually expected to be `ts' structs. The docstring is written for users.
"Return non-nil if current entry has a timestamp in given period.
If no arguments are specified, return non-nil if entry has any
timestamp.
If FROM, return non-nil if entry has a timestamp on or after
FROM.
If TO, return non-nil if entry has a timestamp on or before TO.
If ON, return non-nil if entry has a timestamp on date ON.
FROM, TO, and ON should be either `ts' structs, or strings
parseable by `parse-time-string' which may omit the time value.
TYPE may be `active' to match active timestamps, `inactive' to
match inactive ones, or `both' / nil to match both types.
LIMIT bounds the search for the timestamp REGEXP. It defaults to
the end of the entry, i.e. the position returned by
`org-entry-end-position', but for certain searches it should be
bound to a different positiion, e.g. for planning lines, the end
of the line after the heading."
;; MAYBE: Define active/inactive ones separately?
:normalizers
((`(,(or 'ts-active 'ts-a) . ,rest) `(ts :type active ,@rest))
(`(,(or 'ts-inactive 'ts-i) . ,rest) `(ts :type inactive ,@rest)))
:preambles
((`(,predicate-names . ,(and rest (guard (or (plist-get rest :from)
(plist-get rest :to)
(plist-get rest :on)))))
(-let (((&plist :from :to :on :type) rest))
(org-ql--from-to-on)
(list :regexp (-let* ((from (or from (ts-adjust 'day (- org-ql-ts-days-from-default) (ts-now))))
(to (or to (ts-adjust 'day org-ql-ts-days-to-default (ts-now)))))
(org-ql--ts-range-to-regexp from to))
:query query)))
(`(,predicate-names . ,rest)
(list :regexp (pcase (plist-get rest :type)
((or 'nil 'both) org-tsr-regexp-both)
('active org-tsr-regexp)
('inactive org-ql-tsr-regexp-inactive))
;; Predicate needs testing only when args are present.
:query (-let (((&keys :from :to :on) rest))
;; FIXME: This used to be (when (or from to on) query), but that doesn't seem right, so I
;; changed it to this if, and the tests pass either way. Might deserve a little scrutiny.
(if (or from to on)
query
t)))))
;; TODO: DRY this with the clocked predicate.
:body
(cl-macrolet ((next-timestamp ()
`(when (re-search-forward regexp limit t)
(ts-parse-org (match-string match-group))))
(test-timestamps (pred-form)
`(cl-loop for next-ts = (next-timestamp)
while next-ts
thereis ,pred-form)))
(save-excursion
(cond ((not (or from to)) (re-search-forward regexp limit t))
((and from to) (test-timestamps (ts-in from to next-ts)))
(from (test-timestamps (ts<= from next-ts)))
(to (test-timestamps (ts<= next-ts to)))))))
(setf org-ql-cache (make-hash-table :weakness 'key))
(org-ql-select (org-agenda-files)
'(ts :from "2020-01-01" :to "2020-12-31")
:action '(substring-no-properties (org-get-heading t t)))))))
#+END_SRC
#+RESULTS:
| Form | x faster than next | Total runtime | # of GCs | Total GC runtime |
|-------------+--------------------+---------------+----------+------------------|
| optimized | 3.78 | 1.243185 | 0 | 0 |
| unoptimized | slowest | 4.700824 | 0 | 0 |
** PROJECT Multi-pass query normalization
It would allow, e.g. one query to be normalized into another, and then into another. It would be helpful for timestamp-related ones, I think.