Re: [External] : [emacs bookmark.el] Sorting by last set

[Karl Fogel <kfogel@red-bean.com>]

A few thoughts I just had while looking over the recent 
improvement to bookmark.el sorting (commit f461eb8fa -- not the 
later followup commit fccde52158, which isn't relevant for these 
thoughts):

1) We use the value `last-modified' (for `bookmark-sort-flag') to 
   represent sorting by the date the bookmark was most recently 
   updated to point to a particular target.

   But there are other ways to "modify" bookmarks: for example, 
   you could edit a bookmark's annotation without updating its 
   target.  One could make a reasonably good argument as to why an 
   annotation change should count as a "modification" for the 
   purposes of sorting... and, one could make a reasonably good 
   argument why it shouldn't.

   My purpose here is just to ask: is the name "last-modified" 
   really the most appropriate one for the behavior currently 
   implemented?

   The simple solution would be to just change the symbol to 
   `last-set-date'.  I think that would be my choice.  It would 
   reduce the potential for confusion and misunderstanding.

   (I won't go in to the more complex solutions here, as I'm not 
   sure that bookmark.el really needs its sorting capability to be 
   fully operational battle station.)

* Now that we're using a symbol for *one* possible value of 
  `bookmark-sort-flag', should we use symbols for *all* possible 
  values?  (And leave the treatments of `t' and nil as legacy 
  compatibility behaviors, documented as such but deprecated in 
  favor of using the corresponding new symbols instead when 
  writing new configurations.)

* Finally, perhaps `bookmark-sort-flag' is no longer the right 
  name for 
  this variable?  It's not a "flag" anymore -- a "flag" should be 
  either `t' or nil.

  This naming tradition mostly holds throughout Emacs, although 
  there is at least one other exception.  Of the 54 symbols whose 
  names end in "-flag" that I found via `M-x apropos', I did a 
  random sampling of about 15.  A few don't have real doc strings 
  (sigh), but their values at least were either 't' or nil.  Of 
  the documented ones in my sample, all but one were true Boolean 
  flags.  The single exception was `quit-flag', which I would now 
  also argue is misnamed, but I'm not advocating here that we 
  change that one :-).

  However, I think it would be good to deprecate 
  `bookmark-sort-flag' in favor of 'bookmark-sort-behavior' or 
  something, and do whatever the usual compatibility dance is for 
  such situations.  It's useful for the suffix "-flag" to actually 
  mean something, and I'd rather not have bookmark.el contribute 
  to the dilution of that linguistic tradition.

By the way, I still think the change as currently landed is an 
unambiguous improvement.  I would still want to keep it even if we 
didn't take any of my suggestions above.

Best regards,
-Karl