From: Paul Haldane (Paul.Haldane@newcastle.ac.uk)
Date: Fri May 14 1999 - 08:45:30 CDT
> 2. date/thread indexes are not properly ordered. Messages arriving
> close together with hypermail in single message update mode result in
> misordering which sometimes gets fixed by new messages coming in, and
> sometimes gets worse with new messages coming in.
Hmmm - I've not noticed this. I'm _reasonably_ confident that the date
index will be properly sorted (according to the date in the 'From ' line).
I've not been able to reproduce this (with a21 but I don't thing this has
changed between 20 and 21).
You can get _apparent_ misordering because the dates displayed are from
the Date: headers whilst the messages are actually sorted by the date in
the 'From ' line. Are you sure the messages are mis-sorted in this sense?
There's a little perl script tacked onto the end of this message that
checks the ordering of an archive. Takes two parameters, first is the
directory containing the archive, second is 0 or 1 - 1 if the reverse
config option was used to build the archive - defaults to 0.
> 3. I'd prefer a consistent, local format for the date string in the
> index pages. Parseing the date out of messages coming from multiple
> locations and multiple MUAs creates a confusing mess.
Try 2a21 which has most of the functionality you describe for presentation
of dates - some more comments below.
Paul
...
> 3. The date that shows in the indexes is what's parsed out of the sender's
> email. Perhaps this got debated and settled earlier? While it's nice to
> see the order things were sent in, having timezones from around the world
> showing up can make it pretty confusing (especially with the unreliability
> described in item 2).
At the moment we store the dates from the 'From ' line and the Date: line
separately - if one is missing or unparseable then we use the other for
both. If they differ by more than 70 minutes then we take the time from
the 'From ' line (hah - just checked the code and I'm _not_ doing that yet
- only prints a warning and that's only if debug is turned on. I'll look
at making this happen all the time).
You can see how hypermail has parsed the two dates in the message by
looking at the comments in the generated html files. isosent and
isoreceived are the times from the Date and From lines respectively
formatted as YYYYYMMDDHHmmSS (as per ISO 8601:1988). These are UTC times.
> I think it makes more sense to use the archiving system's date,
The archiving system's date as stored in the 'From ' line has the
advantage that it should at least be consistent (and it's more likely to
be under your control).
> formatted
> in as terse a format as possible; for example, date -u +"%e %b %Y %H:%M:%S"
> This allows easy comparison, should be reliably parseable out of the
> message header, and minimizes the index page space required because TZ
> and/or GMT offsets are not needed.
a21 allows you to set the the presentation format for date/time using the
dateformat option.
> The "show_header" option including "Date" should provide the means to
> preserve the sender's "Date:" field, if that's specifically desired.
I've not allowed for this - thought about it, but then decided not to
bother at the moment.
> But
> not in the index, where it can do more harm than good!
Agreed.
=======
Script for checking ordering....
#!/usr/local/bin/perl
$archive = shift; $reverse = shift;
$dateindex = $archive.'/date.html';
open(DATE, "< $dateindex") || die;
while (<DATE>) {
next unless /^<LI><A HREF="(\d\d\d\d).*/;
$file = $1;
push @files, $file;
}
close(DATE);
chdir($archive) || die;
$lastdate = '00000000000000';
$debug = 0;
while ($file = shift(@files)) {
print "$file\n";
open(MESSAGE, "< $file.html") || die;
while (<MESSAGE>) {
next unless /isoreceived/;
/.*isoreceived="(\d*)".*/;
$date = $1;
print "$file $date\n" if $debug > 1;
if (($reverse && ($date < $lastdate)) ||
(!$reverse && ($date > $lastdate))) {
print "out of order $file $date $lastdate\n";
}
}
close(MESSAGE);
$lastdate = $date;
}