Re: the *precise* definition of 1em

   From: "David Perrell" <davidp@earthlink.net>
   Cc: <www-style@w3.org>
   Date: Sat, 19 Jul 1997 17:22:19 -0700
   X-MSMail-Priority: Normal
   X-Priority: 3
   X-Mailer: Microsoft Internet Mail 4.70.1161
   MIME-Version: 1.0
   Content-Type: text/plain; charset=ISO-8859-1
   Content-Transfer-Encoding: 7bit

   Hakon Lie wrote:
   > Whenever 'numbers' contain the glyph you are looking for, it should
   > used. Othewise 'someotherfont' should be used. This way, you may end
   > up using both fonts in the same element. This raises an issue: which
   > of the values should be inherited? [1] specifies that it is the
   > approximated value that should be inherited, but does not specify
   > happens when there are more than one approximated value. Suggestions?

   Average the 'approximated' values: (10 + 12)/2 = 11pt.

If you do that, there will still be obscure cases where you will have
1em still giving you a different font, I think.

How about this example:

BODY { font-family: one two three; font-size: 11pt }
P { font-size: 1em }

Assume that the font named `one' is available only as a 50 point font.

Assume that two is available as 12 and 24 points.

Assume that three is availible as 10 and 20.

So we end up using one as 50 points, two as 12 points, and three as
10 points.

The average will be 72/3 = 24 points.

Thus 1em will give you some larger fonts.

But the above example may not be worth considering, since the 50 point
font might be too far away to ever be considered.

So, try another situation:

BODY { font-family: one two three four five six; font-size: 11pt }
P { font-size: 1em }

one is availible as 9 points and 12 points.  Thus we use 12 points.

two is available as 10 and 13 points.  The same is true for three, four
five, and six.  So for these 5 fonts, we use 10 points.

The average will be 62/6 = 10 points.

Thus, the 1em will cause the font named one to appear smaller, since
10 (the average) is closer to 9 points than 12 points.

These cases are obscure, but if we need to make sure that 1em
will give us the same size as he parent element, the average rule
won't always work reliably.