->>> OK, I've found the bug. As internal pages, comments are not preprocessed
->>> (and so their date attributes do not have a chance to take effect) until
->>> they are already being inlined, by which time they have already been
->>> sorted by the files' ctimes. Possible fixes include:
->>>
->>> * scan any modified/created internal pages (including preprocessing them
->>> in scan mode) before rendering any page, fixing all possible bugs of this
->>> form but causing a slower scan phase
->>>
->>> * list and scan the comments below a given page in its `pagetemplate`
->>> hook before inlining them, causing a slower build phase for pages with
->>> many comments
->>>
->>> I think the former is a more general fix, and I'll try to implement
->>> it. --[[smcv]]
+----
+
+[[!template id=gitbranch author="[[smcv]]" branch=smcv/comments-metadata]]
+
+I thought that, as internal pages, comments were not preprocessed
+(and so their date attributes did not have a chance to take effect) until
+they were already being inlined, by which time they have already been
+sorted by the files' ctimes. Actually, I was wrong about that - internal
+pages have a special case elsewhere - but they did skip the `scan` hook,
+which is also fixed in my branch.
+
+The real bug was that the preprocess hook for comments didn't run
+in the scan phase; my branch fixes that, streamlines that hook a bit
+when run in the scan phase (so it doesn't htmlize, and only runs nested
+directives in scan mode), and adds a regression test. --[[smcv]]
+
+[[!tag patch]]
+
+> Thanks.. I am not 100% sure if I just forgot to scan internal pages
+> or left it out as some kind of optimisation since none needed to be
+> scanned. Anyway, if it was an optimisation it was not much of one
+> since they were preprocessed. All applied, [[done]]. --[[Joey]]