2007-08-09 14:36 Ben 'Cerbera' Millard would you help analyse a data table I've found?
2007-08-09 14:36 zcorpan sure
2007-08-09 14:36 Ben 'Cerbera' Millard http://aa.usno.navy.mil/data/docs/EarthSeasons.html -
it's done as ASCII text; I'm making an HTML version of
it
2007-08-09 14:36 Ben 'Cerbera' Millard to simulate what they might do if they did so
2007-08-09 14:37 Ben 'Cerbera' Millard http://sitesurgeon.co.uk/tables/astro/06-seasons/ - my
progress so far. it seems too irregular to express
without headers+id
2007-08-09 14:38 Ben 'Cerbera' Millard I thought scope="colgroup rowgroup" or something might
work, but the second equinox and second soltice in each
year are hard to associate the right headers to
2007-08-09 14:38 Ben 'Cerbera' Millard (I realise that's currently non-conforming, btw)
2007-08-09 14:38 zcorpan i don't quite get the table
2007-08-09 14:39 zcorpan why are the years mentioned twice
2007-08-09 14:39 zcorpan ?
2007-08-09 14:39 Ben 'Cerbera' Millard it's an irregular layout, presumable to save vertical
space
2007-08-09 14:40 Ben 'Cerbera' Millard the Perhihelion and Aphelion events are in the left
third. the first and second Equinox and Solstice are the
in final two thirds
2007-08-09 14:40 Ben 'Cerbera' Millard guess they wanted a year with each type of event
2007-08-09 14:40 zcorpan yeah, ok
2007-08-09 14:42 zcorpan i think this table should be able to work with just
scope=row on the 5th column if the auto algorithm is
improved
2007-08-09 14:42 Ben 'Cerbera' Millard scope=row would subtract the headers to the left of it?
2007-08-09 14:43 zcorpan no
2007-08-09 14:43 zcorpan but you'd want it to cover the last three columns too
2007-08-09 14:44 zcorpan Equinoxes is associated with 20, 08, 48 and 22, 18, 43,
right
2007-08-09 14:44 Ben 'Cerbera' Millard yes
2007-08-09 14:46 zcorpan so if we start with a data cell and go leftwards until
we hit a TH, associate it with the cell, then continue
to the left and associate each TH with the cell, until
we hit a TD again, in which case we only associate
further THs on the left that have scope=row
2007-08-09 14:46 zcorpan unless those THs have scope=col or something
2007-08-09 14:46 zcorpan pretty much like html4 suggests
2007-08-09 14:46 zcorpan and then similarly upwards
2007-08-09 14:47 Ben 'Cerbera' Millard I see
2007-08-09 14:48 Ben 'Cerbera' Millard so you think of the table more in terms of cell to
headers than header to cells?
2007-08-09 14:50 zcorpan dunno, it can probably be described equivalently the
other way around
2007-08-09 14:50 Ben 'Cerbera' Millard hmm, I think you're right actually. Perihelion gets
associated with all cells to its right until you hit | .
2007-08-09 14:51 Ben 'Cerbera' Millard | (Equinoxes) gets associated will all
cells to its right
2007-08-09 14:51 zcorpan yeah
2007-08-09 14:52 Ben 'Cerbera' Millard the "Mar" would be a | and would get associated with
all | s to its right, stop at the next |
2007-08-09 14:52 zcorpan yeah
2007-08-09 14:53 Ben 'Cerbera' Millard and then Sept gets the last three (adding to Equinoxes)
2007-08-09 14:53 zcorpan yep
2007-08-09 14:54 Ben 'Cerbera' Millard 1992 gets associated with the | s below it, stopping
at the next |
2007-08-09 14:54 zcorpan then the algorithm needs to integrate with colgroup,
rowgroup and headers="" as well
2007-08-09 14:54 zcorpan and that needs to get specced
2007-08-09 14:58 Ben 'Cerbera' Millard http://www.guidebookgallery.org/icons/components -
heirarchical row headers (aka nested row headers) have
given me some headaches
2007-08-09 14:58 Ben 'Cerbera' Millard if scope="rowgroup" is used on "Applications", that
would stop it at "Settings"
2007-08-09 14:59 Ben 'Cerbera' Millard but what about "Office", "Multimedia", "Networking" and
"Internet"
2007-08-09 14:59 Ben 'Cerbera' Millard (Settings would also use scope="rowgroup")
2007-08-09 15:00 zcorpan yeah... html doesn't support that case
2007-08-09 15:00 zcorpan you'd need to put "Office" as a colspanning cell to the
left of which is supposed to apply
2007-08-09 15:01 Ben 'Cerbera' Millard rowspanning?
2007-08-09 15:01 zcorpan yeah
2007-08-09 15:02 Ben 'Cerbera' Millard yeah, that's what I thought. you need an empty | on
the first row of a group like that, so it stays
alongside that row of data
2007-08-09 15:02 Ben 'Cerbera' Millard | | Office | | data
|
|---|
| Notepad | data
2007-08-09 15:02 Ben 'Cerbera' Millard I think
2007-08-09 15:03 zcorpan yeah, probably
[...]
2007-08-10 11:19 Ben 'Cerbera' Millard when we were talking tables yesterday, | | | add
together for the final | . but in | | | | ,
the second | replaces the first | for the final
| ?
2007-08-10 11:19 Ben 'Cerbera' Millard (and similar for the vertical axis)
2007-08-10 11:19 zcorpan yes
2007-08-10 11:20 Ben 'Cerbera' Millard http://sitesurgeon.co.uk/tables/astro/06-seasons/scope-auto.html
- so what happens with the "1992" and "1993" headers?
2007-08-10 11:20 Ben 'Cerbera' Millard won't 1993 events get a 1992 header as well?
2007-08-10 11:22 zcorpan no, you don't associate headers with headers
2007-08-10 11:22 Ben 'Cerbera' Millard oh
2007-08-10 11:23 zcorpan which is a bit of an issue because sometimes headers
have headers
2007-08-10 11:23 Ben 'Cerbera' Millard yeah
2007-08-10 11:23 Ben 'Cerbera' Millard in the example table, each data cell has 4 headers, by
my count:
2007-08-10 11:24 Ben 'Cerbera' Millard 1. Unit of time (d, h or m)
2. Month
3. Event type (Perihelion, Aphelion, Equinox, Solstice)
4. Year
2007-08-10 11:24 zcorpan yeah
2007-08-10 11:25 zcorpan with that table the year isn't associated with the
relevant cells
2007-08-10 11:25 zcorpan you need it to rowspan probably
2007-08-10 11:26 Ben 'Cerbera' Millard what if it had colspan to be above the relevant cells?
2007-08-10 11:26 Ben 'Cerbera' Millard oh, then the unit of time gets blocked?
2007-08-10 11:26 zcorpan i mean colspan
2007-08-10 11:27 zcorpan i always mix them up :(
2007-08-10 11:27 Ben 'Cerbera' Millard no worries :)
2007-08-10 11:27 Ben 'Cerbera' Millard won't that block the unit of time from getting all the
way down the table?
2007-08-10 11:29 zcorpan good point. either | s in need to imply
scope=col or you need to declare it explicitly
2007-08-10 11:29 zcorpan thinking about it, implying scope=col for | s in
wouldn't be a bad idea
2007-08-10 11:30 Ben 'Cerbera' Millard sometimes the main headers are repeated in the table,
lower down
2007-08-10 11:30 Ben 'Cerbera' Millard I'm not sure if allowing multiple is feasible?
authors aren't doing it, afaict
2007-08-10 11:30 zcorpan no, and it won't have the right presentation
2007-08-10 11:31 zcorpan 'table-header-row's are always at the top
2007-08-10 11:31 Ben 'Cerbera' Millard ah
2007-08-10 11:32 zcorpan i think repeating headers in a big table is a
presentation issue, actually
2007-08-10 11:32 Ben 'Cerbera' Millard http://www.moneyextra.com/stocks/ftse100/ - here's a
good example of the complete headers being repeated
lower down
2007-08-10 11:32 Ben 'Cerbera' Millard we should probably try to support existing content (as
well), though
2007-08-10 11:32 zcorpan yeah, i don't see any reason why that table wouldn't
work
2007-08-10 11:34 Ben 'Cerbera' Millard it's not doing this, but if the main headers were in
and you implied scope=col for them, it wouldn't
work?
2007-08-10 11:34 zcorpan hmm, then perhaps you would hear the same header twice,
indeed
2007-08-10 11:35 Ben 'Cerbera' Millard yeah...I'm not sure if authors are doing that, to be
fair :)
2007-08-10 11:35 zcorpan this would need some research
2007-08-10 11:35 zcorpan which approach would have the intended result in most
cases
2007-08-10 11:36 Ben 'Cerbera' Millard if the screen reader people get involved, they might
have useful experience of attempts to infer heading
relationships in the absence of explicit association
2007-08-10 11:37 zcorpan i'm not holding my breath
2007-08-10 11:39 Ben 'Cerbera' Millard in the Yahoo! presentation with Doug Geoffrey of
Window-Eyes, he said they were involved with Microsoft
to develop access standards in Windows
2007-08-10 11:39 Ben 'Cerbera' Millard do they hate W3C or something?
2007-08-10 11:39 zcorpan dunno
2007-08-10 11:39 zcorpan the web is not their top priority
2007-08-10 11:39 Ben 'Cerbera' Millard well, that I can understand
2007-08-10 11:39 zcorpan yeah
2007-08-10 11:40 Ben 'Cerbera' Millard then again, they all seem to care enough to have
websites :)
2007-08-10 11:40 zcorpan who doesn't have a website? :)
2007-08-10 11:41 Ben 'Cerbera' Millard true :P
2007-08-10 11:41 Ben 'Cerbera' Millard I guess it's WAI's responsibility to get ATs on board,
not HTMLWG Participants
2007-08-10 11:42 zcorpan yeah, or the w3c staff or something
2007-08-10 11:47 Ben 'Cerbera' Millard http://sitesurgeon.co.uk/tables/astro/06-seasons/scope-colspan-auto.html
- this looks like it would work now
2007-08-10 11:49 zcorpan yeah
2007-08-10 11:49 zcorpan now we need to forward the algorithm we have in our
heads to the working group
2007-08-10 11:49 zcorpan :)
2007-08-10 11:50 Ben 'Cerbera' Millard is it just us, or have you been developing this with
other people?
2007-08-10 11:51 zcorpan it's actually in the html4 spec (but very vague)
2007-08-10 11:52 Ben 'Cerbera' Millard yeah, I know that one but I think what we're discussing
differs from it
2007-08-10 11:52 Ben 'Cerbera' Millard in the way scope works
2007-08-10 11:53 zcorpan true
2007-08-10 11:53 zcorpan no, i haven't really discussed it much with others
2007-08-10 11:53 Ben 'Cerbera' Millard ok, let me refresh my memory of it
2007-08-10 11:54 Ben 'Cerbera' Millard "The search in a given direction stops when the edge of
the table is reached or when a data cell is found after
a header cell."
that's a key part of what we've discussed
2007-08-10 11:54 Ben 'Cerbera' Millard that lets you imply groups in a table without | |
|---|
or
2007-08-10 11:55 zcorpan yup
2007-08-10 11:55 Ben 'Cerbera' Millard "If a header cell has the headers attribute set [...]
the search stops for the current direction."
that's similar to how we want scope to work, but
value-sensitively?
2007-08-10 11:57 zcorpan not sure we want precense of scope to stop the search
2007-08-10 11:57 zcorpan presence
2007-08-10 12:02 Ben 'Cerbera' Millard if it has a certain value and the search is in a certain
direction, I thought we wanted it to stop that
direction? namely, the value "row" when the search is
going left?
2007-08-10 12:03 zcorpan why?
2007-08-10 12:04 zcorpan in the tables we were looking at, the next cell to the
left was a TD anyway
2007-08-10 12:04 Ben 'Cerbera' Millard oh yeah, so it was!
2007-08-10 12:05 Ben 'Cerbera' Millard in that case, I think there's a different problem
2007-08-10 12:06 Ben 'Cerbera' Millard http://sitesurgeon.local/tables/astro/06-seasons/scope-colspan-auto.html
- in the Sept Equinox for 1992, the search needs to
continue left after it reaches "48" after finding
" | Sept" until it finds " | Equinoxes?
2007-08-10 12:06 Ben 'Cerbera' Millard so the search stopped for plain | but continued for
| ?
2007-08-10 12:07 Ben 'Cerbera' Millard * " | Equinoxes"
2007-08-10 12:08 zcorpan the search needs to continue, but we need a state or
flag to not associate the "Mar" TH with the "22" TD
after we have gone past "Sept" (we still need to find
"Equinoxes" because it has scope=row)
2007-08-10 12:09 Ben 'Cerbera' Millard exactly
2007-08-10 12:10 Ben 'Cerbera' Millard and something similar for vertical
2007-08-10 12:10 zcorpan yeah
2007-08-10 12:11 Ben 'Cerbera' Millard so 1992 doesn't get associated with the "4" below "1993"
but the "d" way above it does, since it has scope=col
2007-08-10 12:12 zcorpan yes
2007-08-13 15:47 Ben 'Cerbera' Millard http://sitesurgeon.co.uk/tables/thatcher/01-water/minimal.html
- a case where | blocks the previous | but | must still get through
2007-08-13 15:49 zcorpan yep... implying scope=col for | s in would
make that table work as intended
2007-08-13 15:50 zcorpan semantically, i think using means that the
header cells in it are intended to apply to all tbodies
and thus as if scope=col was specified
2007-08-13 15:51 zcorpan (but still we might need some data about the tradeoff
for doing so)
2007-08-13 15:52 Ben 'Cerbera' Millard is logically expanded as "table header"
2007-08-13 15:53 Ben 'Cerbera' Millard so "the headers for the whole table" seems a logical
definition
2007-08-13 15:53 Ben 'Cerbera' Millard as there can only be 1 section, that further
implies that anything in it will apply to the whole
thing
2007-08-13 15:56 zcorpan yeah
2007-08-13 15:56 Ben 'Cerbera' Millard should the apply to the | ?
2007-08-13 15:57 zcorpan yes
2007-08-13 15:59 Ben 'Cerbera' Millard http://projectcerbera.com/misc/food/2004-04 - that would
work for tables using for column totals
2007-08-13 16:00 zcorpan indeed
2007-08-13 16:00 Ben 'Cerbera' Millard I've seen other uses of , not sure if they'll
work if applies to them. will try and find them
again
2007-08-13 16:00 zcorpan ok
2007-08-13 16:14 zcorpan it might be a good idea to sort out our requirements --
which table structures need to Just Work
2007-08-13 16:15 Ben 'Cerbera' Millard yeah, I'd say conventional cell arrangements have the
highest priority because they are more feasible
2007-08-13 16:15 Ben 'Cerbera' Millard and probably more common
2007-08-13 16:15 Ben 'Cerbera' Millard of the 33 tables I've worked on so far, nearly all have
their headers above or to the left of the data
2007-08-13 16:16 zcorpan makes sense
2007-08-13 16:18 zcorpan my point is that the proposal to the list, when we have
something structured, should be
1. point to data about the most common table structures
2. explain why the current algorithm is insufficient
3. propose changes to the algorithm to make it cover
more cases
2007-08-13 16:18 zcorpan or some such
2007-08-13 16:18 zcorpan :)
2007-08-13 16:18 Ben 'Cerbera' Millard I see
2007-08-13 16:19 Ben 'Cerbera' Millard well, my table collections are deliberately biassed
towards the more interesting tables
2007-08-13 16:19 Ben 'Cerbera' Millard but even those are still pretty regular
2007-08-13 16:19 Ben 'Cerbera' Millard headers in one or more rows along the top is very common
2007-08-13 16:20 Ben 'Cerbera' Millard surprisingly :)
2007-08-13 16:20 zcorpan :)
2007-08-13 16:20 Ben 'Cerbera' Millard what was surprising is how row headers are done
2007-08-13 16:20 Ben 'Cerbera' Millard some authors are doing them as regular below a |
2007-08-13 16:20 Ben 'Cerbera' Millard some leave an empty | in the headers row and do them
as |
2007-08-13 16:21 Ben 'Cerbera' Millard some have both | at the top and | down the side
2007-08-13 16:21 zcorpan http://keryx.se/resources/html-elements.xhtml is another
table that might be interesting
2007-08-13 16:22 Ben 'Cerbera' Millard | but no scope on the column headers
2007-08-13 16:22 Ben 'Cerbera' Millard that's interesting
2007-08-13 16:23 Ben 'Cerbera' Millard it seems authors regularly assume that | will
overwrite a previous | of the same width, if
there are | cells in between
2007-08-13 16:23 Ben 'Cerbera' Millard but won't block | or | with a different
width
2007-08-13 16:24 zcorpan interesting observation
2007-08-13 16:25 Ben 'Cerbera' Millard wewt :)
2007-08-13 16:25 Ben 'Cerbera' Millard it's fairly conventional in print media, at least in the
UK. sometimes see it in price lists and TV guides
2007-08-13 16:27 zcorpan wonder if that's the case more often than not, and if it
would be feasable to define in the algorithm
2007-08-13 16:28 Ben 'Cerbera' Millard I think Henri Sivonen said in a message to HTMLWG that
the algorithm should recreate the way humans "walk" a
table mentally
2007-08-13 16:29 Ben 'Cerbera' Millard in cases where that requires an understanding of the
subject, I don't think it's feasible
2007-08-13 16:29 Ben 'Cerbera' Millard but there is some system to it, largely what the HTML4
algorithm hints at (look left, look up)
2007-08-13 16:32 zcorpan when i think about it it should be trivial. when walking
from TH, just continue to associate TDs after passing
through a TH that has a different colspan. If it's the
same colspan, don't associate further TDs
2007-08-13 16:32 zcorpan (this would be augmenting the algorithm we discussed
before)
2007-08-13 16:33 Ben 'Cerbera' Millard yeah, that makes sense
2007-08-13 16:36 Ben 'Cerbera' Millard I havn't seen | in the middle of tables to
indicate seperate sections within each row, so I don't
think we'll need this augmentation horizontally
2007-08-13 16:36 Ben 'Cerbera' Millard *separate
2007-08-13 16:36 Ben 'Cerbera' Millard but they may well exist
2007-08-13 16:37 zcorpan otoh, having the same heuristics in both directions is
more natural and might perhaps be easier to implement
2007-08-13 16:38 zcorpan not that i have an opinion either way :)
2007-08-13 16:39 Ben 'Cerbera' Millard when you say "propose changes to the algorithm", is that
the HTML5 one?
2007-08-13 16:40 zcorpan yes
2007-08-13 16:40 zcorpan which other could i be referring to? :)
2007-08-13 16:41 Ben 'Cerbera' Millard the HTML4 one is the current Recommendation, vague or
not. but as HTMLWG is working on HTML5, it makes sense
to make changes to that one
2007-08-13 16:42 zcorpan html4 is not updated... html5 is the spec that can be
changed :)
2007-08-13 16:42 zcorpan html4 being a Rec is not really relevant
2007-08-13 16:42 Ben 'Cerbera' Millard that's true
2007-08-13 16:43 Ben 'Cerbera' Millard I've got work coming in later today or early tomorrow
2007-08-13 16:43 Ben 'Cerbera' Millard so my work on this will stop for a while, probably a
week
2007-08-13 16:43 zcorpan ok
2007-08-13 16:45 Ben 'Cerbera' Millard tables seem to have footnotes quite frequently
2007-08-13 16:46 Ben 'Cerbera' Millard usually the table has some sort of punctuation character
like * or ** which is then explained in a or just
text separated by
after or sometimes before the
table
2007-08-13 16:47 zcorpan yup... though footnotes are orthogonal to tables and can
be researched separately i think
2007-08-13 16:47 Ben 'Cerbera' Millard yeah, footnotes that work with anything would be cool
2007-08-13 16:48 Ben 'Cerbera' Millard the inconsistencies I find with with the way authors
currently do this are too big to work around, on the
whole
2007-08-13 16:48 Ben 'Cerbera' Millard but you can normally figure it out as a human
2007-08-13 16:58 Ben 'Cerbera' Millard would it be worthwhile for me to send some initial ideas
to HTMLWG to help focus table discussion around solving
problems in real tables and figuring out parts of an
algorithm which could make it into HTML5?
2007-08-13 16:59 Ben 'Cerbera' Millard basically the sort of conversation we have over MSN
2007-08-13 16:59 zcorpan sure
2007-08-13 16:59 Ben 'Cerbera' Millard okies
2007-08-13 16:59 zcorpan at least point out the problems
2007-08-13 16:59 zcorpan as a start
2007-08-13 17:01 Ben 'Cerbera' Millard I'll start it casually so people can participate easily
2007-08-13 17:02 zcorpan sounds good
2007-08-13 17:03 Ben 'Cerbera' Millard the site Stephen Ferg works at is a goldmine of
statistical data tables
2007-08-13 17:03 Ben 'Cerbera' Millard http://stats.bls.gov/ro5/ecimid.htm
2007-08-13 17:04 Ben 'Cerbera' Millard they mix headers+id with implied | + | association
2007-08-13 17:05 zcorpan |
2007-08-13 17:06 Ben 'Cerbera' Millard I meant the nested row headers use headers="" while the
top-level row headers and all column headers are plain
|
2007-08-13 17:07 zcorpan right
2007-08-13 17:08 Ben 'Cerbera' Millard that's smart, imho: you achieve the association even
through the cells are arranged irregularly, but you keep
the markup small by only using it where needed
2007-08-13 17:08 Ben 'Cerbera' Millard *though
[...]
2007-08-28 21:57 Ben 'Cerbera' Millard oh, I had an idea about heirarchical row headers
2007-08-28 21:58 Ben 'Cerbera' Millard the algorithm to apply a | to other cells won't go
back up the table, will it?
2007-08-28 21:58 Ben 'Cerbera' Millard it only apples the header downwards and leftwards
2007-08-28 21:58 Ben 'Cerbera' Millard *rightwards
2007-08-28 21:59 zcorpan right
2007-08-28 22:00 Ben 'Cerbera' Millard in that case, you could use scope=rowgroup to fix it in
the simpler cases
2007-08-28 22:01 Ben 'Cerbera' Millard - header 1
- header 2
- header 3
- header 4
- header 5
...
2007-08-28 22:01 Ben 'Cerbera' Millard header 1 and 2 would get scope=rowgroup. header 1 and
header 5 would have a | before them
2007-08-28 22:02 zcorpan aha!
2007-08-28 22:02 zcorpan clever
2007-08-28 22:02 Ben 'Cerbera' Millard :)
2007-08-28 22:02 zcorpan although doesn't scale
2007-08-28 22:02 Ben 'Cerbera' Millard doesn't work if there is a second set at header 2's
level
2007-08-28 22:02 zcorpan indeed
2007-08-28 22:03 zcorpan we won't be able to express all possible tables
2007-08-28 22:03 zcorpan we'll have to draw the line somewhere
2007-08-28 22:03 Ben 'Cerbera' Millard indeed
2007-08-28 22:04 zcorpan although headers="" covers many funky tables
2007-08-28 22:04 zcorpan if someone would actually use it correctly :)
2007-08-28 22:04 Ben 'Cerbera' Millard the Gorleston Tide Table does
2007-08-28 22:06 Ben 'Cerbera' Millard I think Stephen Ferg's heirarchical row header tables
are using it sanely and would work if we handle the
various techniques in a single algorithm
2007-08-28 22:06 Ben 'Cerbera' Millard but that might not be feasible
2007-08-28 22:07 zcorpan oh it's probably well possible to come up with an
algorithm that integrats with headers
2007-08-28 22:07 zcorpan however we first need to figure out if it helps or hurts
to support headers
2007-08-28 22:08 Ben 'Cerbera' Millard yeah, I've seen some headers+id which are imperfect
2007-08-28 22:08 Ben 'Cerbera' Millard normally they are using header+id which we can ignore