Re: [bug] org-agenda-write does not handle date stamps without day of week

[Nick Dokos <nicholas.dokos@hp.com>]

[-- Attachment #1: Type: text/plain, Size: 2854 bytes --]

Karl Voit <devnull@Karl-Voit.at> wrote:

> * Nick Dokos <nicholas.dokos@hp.com> wrote:
> > Karl Voit <devnull@Karl-Voit.at> wrote:
> >
> > For me, it was a "no time to work on org - stash it"...
> 
> OK. I just wanted to make sure that it *is* on someone's todo list
> :-)
> 
> >> * Karl Voit <devnull@Karl-Voit.at> wrote:
> >> >
> >> > * <2012-03-05 08:00-09:00> Wrong: ends up as full day event
> >
> > org-agenda-write calls org-export-icalendar which calls org-print-icalendar-entries
> > which loops over all the entries and parses them, decomposing them into timestamps.
> > Each timestamp is then passed to org-parse-time-string. It's this one that cannot
> > handle non-standard formats: it uses a regexp and assumes that all the matched parts
> > are going to be in fixed places:
> >
> > As to how to fix it, there are several possibilities:
> >
> > 1. fix your scripts that produce time stamps to include day-of-week.
> 
> Sorry, deriving DOW from an arbitrary timestamps from arbitrary data
> sources is either pretty time consuming (calendar calculations) or
> simply hard to calculate.
> 
> Outside Org-mode, DOW is seldom part of time-stamp data :-(
> 
> > 2. change the callers of org-parse-time-string to make sure that DOW is included.
> > There are roughly three dozen callers, so 2. is possible but a pain.
> 
> Ack.
> 
> > 3. change just one caller: org-print-icalendar-entries to make sure that DOW is included.
> > 3. is simple but ugly as sin, 
> 
> Ouch, ack :-)
> 
> > 4. change org-parse-time-string to handle a missing DOW.
> > 4. is the best way to handle it within org.
> 
> Full ack.
> 
> > I vote for 1. where *you* have to do all the work ;-)
> 
> YMMD :-)
> 
> If my brain would be compatible to ELISP, I'd send a patch.
> Promised.
> 
> But I'll take my chance and wait for someone else (you?)
> implementing 4. to resolve this issue for everybody. I really
> appreciate every second you guys invest in maintaining Org-mode!
> 

I don't know about you, but whenever I engage in hand-to-hand combat
with a complicated regexp, I come out bruised, muddied and a lot worse
for wear. In any case, I'm attaching an org file with my investigations.
It contains a description and a code block for testing.

I hope that the attachment will come through unscathed: it contains
regexps, and munging a regexp that looks like hen scratchings in the
first place through uncooperative mailers is not something to be
relished.

BTW, I'm not advocating a change: I'll leave it to Karl to do that if he
really wants to and to the maintainers to decide whether it's worth
doing. But it can be done (more or less). And maybe somebody will come
up with a better way than the proof-of-concept that I'm attaching here.

Nick

PS. BTW, if you look at the attachment, it helps if you have a wide window,
    something like 165 characters wide.


[-- Attachment #2: Karl's folly --]
[-- Type: text/plain, Size: 3016 bytes --]

* Modify org-ts-regexpr0 to satisfy Karl Voit :-)

This is an attempt to allow timestamps without a day-of-week.

Modifying complicated regexps is no fun: I had to resort to re-builder
to figure out what was going on. The problem was that the first " *" which
is supposed to eat all the spaces between the "2012-03-05" part and
the "Mon" part, would, in the absence of the "Mon" part, eat the space
before the time spec "08:00". But the regexp for the time part explicitly
requires that space (see the second sp in the explanatory line below).
So I tried making the DOW part optional by appending a ?, but that didn't
do the trick. re-builder showed the space problem: there was no space left
to match if DOW is absent, so I changed the literal space " " to 0 or more
spaces " *" and that did the trick (with some caveats noted below).

In the code block below, tslist is a list of test cases. The code
block sets org-ts-regexp0 first to its original value, then to the
changed value (the first one is there just for comparison between the
two). I've marked the changes (of which, the second is the significant
one). We map the function org-parse-time-string over the test
cases and get a list of results. The results look reasonable, but the
new regexp allows funny-looking input like the second and fourth test
case.

One additional caveat is that org-odt uses org-ts-regexp0, so changing
it without checking org-odt's use is fraught with peril, but that's what
I've done: I have no idea whether it causes problems in org-odt.

#+BEGIN_SRC elisp
  ;                           year_______    -   month______    -   day________    sp   DOW____________       sp  hour_________    :   minutes____
  (setq org-ts-regexp0 "\\(\\([0-9]\\{4\\}\\)-\\([0-9]\\{2\\}\\)-\\([0-9]\\{2\\}\\) *\\([^]+0-9>\r\n -]*\\)\\( \\([0-9]\\{1,2\\}\\):\\([0-9]\\{2\\}\\)\\)?\\)")
  ; changed                                                                                                v    v
  (setq org-ts-regexp0 "\\(\\([0-9]\\{4\\}\\)-\\([0-9]\\{2\\}\\)-\\([0-9]\\{2\\}\\) *\\([^]+0-9>\r\n -]*\\)?\\( *\\([0-9]\\{1,2\\}\\):\\([0-9]\\{2\\}\\)\\)?\\)")
  (setq tslist '(
          "<2012-03-05 Mon 08:00>"
          "<2012-03-05 Mon08:00>"
          "<2012-03-05 Mon 8:00>"
          "<2012-03-05 Mon8:00>"
          "<2012-03-05 Mon     8:00>"
          "<2012-03-05    Mon     8:00>"
          "<2012-03-05 Mon >"
          "<2012-03-05 Mon>"
          "<2012-03-05 08:00>"
          "<2012-03-05 8:00>"
          "<2012-03-05       8:00>"
          "<2012-03-05>"))
  
  (pp (mapcar 'org-parse-time-string tslist))
  
  
#+END_SRC

#+RESULTS:
#+begin_example
((0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 0 5 3 2012 nil nil nil)
 (0 0 0 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 8 5 3 2012 nil nil nil)
 (0 0 0 5 3 2012 nil nil nil))
#+end_example