[PATCH] org-macs.el: Do not compare wall time and file modification time

[Max Nikulin <manikulin@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2013 bytes --]

Mark Barton to emacs-orgmode, emacs-devel. master 4a1f69ebca 2/2: Use 
(TICKS . HZ) for current-time etc. Tue, 26 Apr 2022 23:37:50 -0700. 
https://list.orgmode.org/BF5B9308-3FEF-4DC6-98C9-BFF36F19D36C@gmail.com
 >
> The change also breaks org-file-newer-than-p function that triggered the
> debugger while loading my init that uses org babel.

I think, it should be fixed in the bugfix Org branch.  The attached 
patch is a compromise to some degree, but I do not see a robust solution.

I do not consider current behavior as reliable, however if you would 
prefer to keep it, the following patch may be used instead:

Paul Eggert to emacs-orgmode. Re: master 4a1f69ebca 2/2: Use (TICKS . 
HZ) for current-time etc. Wed, 27 Apr 2022 00:39:01 -0700. 
https://list.orgmode.org/f200c9ab-d1d4-d5a8-24cf-4e1082528fe7@cs.ucla.edu

The changes are not covered by unit tests at least when most babel 
languages are disabled.

On 30/04/2022 17:56, Max Nikulin wrote:
> 
>      (and mtime (not (and time (time-less-p mtime time))))

Treating equality as "newer" would break `org-compile-file', so I 
changed the condition. Previously it was not a case since file 
modification time is usually in the past in comparison to current time.

> On 30/04/2022 01:10, Paul Eggert wrote: 
>> +  (when-let ((mtime (file-attribute-modification-time (file-attributes file))))
>> +    (time-less-p time mtime))) 

> `file-attribute-modification-time' makes code clearer, but it causes
> some complications. Formally compatibility with Emacs-25 (e.g.
> ubuntu-18.04 LTS bionic) is not required for the "main" branch. Emacs
> sources have the "bugfix" Org branch of the stable release though. The
> latter still supports Emacs-25, so either the Emacs source tree and the
> Org bugfix branch will diverge at this point or it is safer to avoid
> `file-attribute-modification-time' till the next major Org release.
> Maybe Org maintainers and developers will correct me.

I have found `file-attribute-modification-time' in org-compat.el.

[-- Attachment #2: 0001-org-macs.el-Do-not-compare-wall-time-and-file-modifi.patch --]
[-- Type: text/x-patch, Size: 3649 bytes --]

From d37b5bb295c69572d1615e7fb2c0bcce05cb2b58 Mon Sep 17 00:00:00 2001
From: Max Nikulin <manikulin@gmail.com>
Date: Fri, 6 May 2022 23:34:52 +0700
Subject: [PATCH] org-macs.el: Do not compare wall time and file modification
 time

* lisp/org-macs.el (org-file-newer-than-p): Fix Emacs-29 problem with
changed representation of system clock timestamp.  Recommend passing
file modification time and do not truncate its precision.
(org-compile-file): Store file modification time instead of system clock
for later comparison by `org-file-newer-than-p'.

It changes behavior of `org-babel-load-file' for the case of equal
modification time of source and tangled files that may happen on
filesystems with coarse timestamps, for example HFS+.  The file will be
tangled again.  Treating equal timestamps as up to date state would
break `org-compile-file' however.  Anyway time comparison is not really
reliable since precision of filesystem timestamps is unknown and it
differs from system clock precision.

Reported by Mark Barton <mbarton98@gmail.com>
https://list.orgmode.org/BF5B9308-3FEF-4DC6-98C9-BFF36F19D36C@gmail.com

During discussion of the issue Paul Eggert <eggert@cs.ucla.edu>
suggested over variants of the changes in the same thread.
---
 lisp/org-macs.el | 32 +++++++++++++++++++++-----------
 1 file changed, 21 insertions(+), 11 deletions(-)

diff --git a/lisp/org-macs.el b/lisp/org-macs.el
index b10725bd5..556bf658d 100644
--- a/lisp/org-macs.el
+++ b/lisp/org-macs.el
@@ -256,16 +256,26 @@ ignored in this case."
 ;;; File
 
 (defun org-file-newer-than-p (file time)
-  "Non-nil if FILE is newer than TIME.
-FILE is a filename, as a string, TIME is a list of integers, as
-returned by, e.g., `current-time'."
-  (and (file-exists-p file)
-       ;; Only compare times up to whole seconds as some file-systems
-       ;; (e.g. HFS+) do not retain any finer granularity.  As
-       ;; a consequence, make sure we return non-nil when the two
-       ;; times are equal.
-       (not (time-less-p (cl-subseq (nth 5 (file-attributes file)) 0 2)
-			 (cl-subseq time 0 2)))))
+  "Non-nil if FILE modification time is greater than TIME.
+TIME should be obtained earlier for the same FILE name using
+
+  (file-attribute-modification-time (file-attributes file))
+
+If TIME is nil (file did not exist) then any existing FILE
+is considered as a newer one.  Some file systems have coarse
+timestamp resolution, for example 1 second on HFS+ or 2 seconds on FAT,
+so nil may be returned when file is updated twice within a short period
+of time.  File timestamp and system clock `current-time' may have
+different resolution, so attempts to compare them may give unexpected
+results.
+
+Attempt to check whether a derived file has been updated in
+response to modification of its source file may give unreliable
+result.  Equal timestamps in such case may mean that the derived
+file is up to date however this function returns nil assuming
+that the FILE is not modified."
+  (let ((mtime (file-attribute-modification-time (file-attributes file))))
+    (and mtime (or (not time) (time-less-p time mtime)))))
 
 (defun org-compile-file (source process ext &optional err-msg log-buf spec)
   "Compile a SOURCE file using PROCESS.
@@ -299,7 +309,7 @@ it for output."
 	 (full-name (file-truename source))
 	 (out-dir (or (file-name-directory source) "./"))
 	 (output (expand-file-name (concat base-name "." ext) out-dir))
-	 (time (current-time))
+	 (time (file-attribute-modification-time (file-attributes output)))
 	 (err-msg (if (stringp err-msg) (concat ".  " err-msg) "")))
     (save-window-excursion
       (pcase process
-- 
2.25.1