Reviewing my attempt to speedup collecting of refile targets
https://orgmode.org/list/s209r8$16en$1...@ciao.gmane.io/
I have realized that refile cache is unreliable.
With specific customization, cache content and thus refile targets
depend on the function called first: `org-refile' ([C-u] C-c C-w) or
`org-goto' (C-u C-c C-w).
I decided to try to provide minimal example and steps to reproduce in
the form of a test marked as expected failure.
Certainly cache should be enabled
(org-refile-use-cache t)
`org-refile-targets' should have (nil . (:maxlevel . 5)) for the
affected buffer. The value of the following setting is intentionally set
to alternative value in comparison to `org-goto' code
(org-refile-use-outline-path nil)
To see the issue interactively, you likely should set
(org-outline-path-complete-in-steps nil)
Clean cache (C-u C-u C-u C-c C-w) and try jumping using C-u C-c C-w and
C-u C-c C-j in various order. Use TAB completion to see targets.
To make difference more apparent, define
`org-refile-target-verify-function' to filter-out some headings.
My expectation that each command has list of targets formatted
accordingly to user setting or `org-goto' internal overrides. Actually
lists of target are the same since they share cache entry.
I suppose, cache keys should include values of all parameters affecting
filtering and formatting, not only regexp for heading selection.
However I have no idea how to derive some value suitable for cache key
from `org-refile-target-verify-function'.
>From d23b10d658539a4646ef015ac2660e1f8c8e7e1b Mon Sep 17 00:00:00 2001
From: Max Nikulin <maniku...@gmail.com>
Date: Wed, 28 Apr 2021 22:30:04 +0700
Subject: [PATCH] testing/lisp/test-org.el: Refile cache failure
testing/lisp/test-org.el (test-org/org-refile-cache): Add expected
failure for cache content discrepancy when it is populated
by `org-refile' or by `org-goto' (C-u C-c C-j).
Usage of just regexp for heading filtering as cache key is unreliable
since other settings affects selected entries
(`org-refile-target-verify-function') or content of entries
(`org-refile-use-outline-path'). However most of users have no chance
to notice collision since they do not call both functions
interchangeably or have `org-refile-targets' different from
the value in `org-goto' code.
---
testing/lisp/test-org.el | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/testing/lisp/test-org.el b/testing/lisp/test-org.el
index 9f0ede1b9..f44cd76a6 100644
--- a/testing/lisp/test-org.el
+++ b/testing/lisp/test-org.el
@@ -6394,6 +6394,30 @@ Paragraph<point>"
(org-refile-targets `((nil :level . 1))))
(member (buffer-name) (mapcar #'car (org-refile-get-targets)))))))
+(ert-deftest test-org/org-refile-cache ()
+ "Demostrate a cache conflict of `org-goto' and `org-refile'.
+
+Notice that `org-refile-target-verify-function' can cause more severe
+divergence of cache content generated by these functions."
+ :expected-result :failed
+ (org-test-with-temp-text "* H1\n** H2\n"
+ (let ((org-refile-use-cache t)
+ ;; Regexp serves as cache key, so targets should be the same
+ ;; as in `org-goto' code.
+ (org-refile-targets '((nil . (:maxlevel . 5))))
+ ;; This value is opposite to the one in `org-goto' code.
+ (org-refile-use-outline-path nil)
+ (targets-refile))
+ ;; If jumping using `org-refile' (C-u C-c C-w)...
+ (setq targets-refile (mapcar #'car (org-refile-get-targets)))
+ (org-refile-cache-clear)
+ ;; Tune settings to simulate `org-goto' (C-u C-c C-j).
+ (let ((org-refile-use-outline-path t))
+ ;; Value is discarded, call just to populate the cache.
+ (org-refile-get-targets))
+ ;; Targets got by `org-refile'.
+ (let ((targets-goto (mapcar #'car (org-refile-get-targets))))
+ (should (equal targets-refile targets-goto))))))
;;; Sparse trees
--
2.25.1
