loge.hixie.ch

Hixie's Natural Log

2003-12-02 17:14 UTC The mystery of why only four properties apply to table columns

Have you ever wondered why columns in CSS can only have four properties applied to them?

On the face of it, it seems strange — if you wanted to make all the cells in a column of prices be red and right-aligned, the following snippet of CSS would seem to make sense:

col.prices { color: red; text-align: right; }

...and indeed, IE6 supports this. So why does the spec say it shouldn't?

The answer to this question lies in another question: How would you implement it?

The colour of text is dependent on the 'color' property of its element. Unless specified, the 'color' property (basically) defaults to 'inherit', which means "take the value of the parent element".

So for some text in a cell, the colour is determined by the 'color' property of the cell, which is taken from the row, which is taken from the table, which is taken from the table's parent, and so on.

What about the column? Well, the column isn't one of the cell's ancestors, so it never gets a look-in! And therein lies the problem.

Now, while that explains why it doesn't apply in the current model, it doesn't actually explain why the model couldn't be changed a little. For example you could say that if a cell has no value set, it should inherit from the column instead of the row.

Unfortunately, you then run into two problems. The first is that in CSS, everything is always set. That isn't a huge problem, because you could always say that the initial value of certain properties was 'auto' or 'normal', and have that value Do The Right Thing when inherited. It wouldn't be pretty, but it could be done. (This is the approach I used to solve some similar problems in the CSS3 Generated and Replaced Content module.)

The second problem is rather more fundamental, and to explain it we'll take a look at the overall processing model for CSS.

As you read this, remember that what we want to do is compute the value of 'color' on cells from the appropriate table column.

Here is how CSS works, at a very high level:

  1. Parse the stylesheets and the document.
  2. For each element in the document:
    1. Decide which CSS rules apply.
    2. Perform the CSS cascade with those rules.
    3. Perform inheritance of properties if the result of the cascade is the keyword 'inherit' (or no specified value for inherited properties).
    4. Perform computations (turn 'em's into 'px's, etc). According to CSS2.1, the getComputedStyle() DOM method returns these values.
    At this point, every element has a value for every property ('display', 'color', etc).
  3. Lay out the document.
  4. Paint the document.

Now, columns are only columns because their 'display' property is set to 'table-column', and a cell is only a cell because it's 'display' is 'table-cell', and the exact relationship between a cell and a column can only be calculated when laying out the document, since you have to take into account what cells span several columns, etc.

So you know which cell is in which column during stage 3, and not earlier. But the stage where you would work out what the color property of a cell is step 2! And at this stage, you don't yet know which column applies to which cell, so you can't go and ask the column for the value.

Hence why the spec limits the properties that apply.

So why does it work in IE6? Well, it can do this despite what I said above because it doesn't support explicit CSS inheritance (the 'inherit' keyword), it doesn't support getComputedStyle(), and it doesn't support 'display: table-column' and the other table display types. In IE6, the model is probably more like:

  1. Parse the stylesheets and the document.
  2. For each element in the document:
    1. If it's a table, map out its structure.
    2. Decide which CSS rules apply.
    3. Perform the CSS cascade.
    4. Perform inheritance of properties, inheriting magically from table columns if the column isn't inheriting its style itself.
  3. Lay out the document.
  4. Paint the document.

As you can see, this is rather different from what the specs say, and is only possible because the model has been quite radically changed, at the expense of a number of useful features.

By the way, if anyone has a way to solve this problem, the working group would probably be very interested in hearing it. The www-style@w3.org mailing list would be the place to make such suggestions.

Pingbacks: 1 2