24005 - font size 9px rounded down to 8px {font} {ll}

Status: VERIFIED FIXED

(Core :: Layout, defect, P1)

Description

[Erik van der Poel]

There is a piece of code in the Windows GFX that rounds the font size in the
same way as Netscape 4.X. The problem is that this code ends up rounding 9px
down to 8px, which is illegible.

Comment 1

[Erik van der Poel]

Removed the rounding code after discussing with Troy and testing on various
"Top Sites" (such as Yahoo).

Comment 2

[Chris Petersen]

Fixed in the Jan 18th build.

Comment 3

[Erik van der Poel]

I have put the rounding code back in, since removing it caused problems. (HTML
font sizes 1-7 were too large, and it became difficult to see the effect of
other changes like buster's min line height fix.)

Comment 4

[Hixie (not reading bugmail)]

Erik: is the code which makes this dependant on a pref checked in or not?

Comment 5

[Erik van der Poel]

Yes, the pref code has been checked in. The default is to round. To turn it off:

  user_pref("font.size.nav4rounding", false);

By the way, I did some more analysis of Peter's changes last night, and I found
that the original check-in was indeed intended to round font sizes in the same
way as Nav4. Then he added some other code that was supposedly intended to
round the font size for "proper Windows font scaling", but it was commented out
and the Nav4 rounding was still in effect. Then he commented out the Nav4
rounding, and turned on the "proper Windows" rounding, but even that is wrong
in my opinion, because that one also rounds 9px down to 8px. I've sent some
email to Peter to ask about this history, twice, but no response so far. Cc'ing
RickG and SteveC, in case they are still in touch with Peter.

Comment 6

[Hixie (not reading bugmail)]

Here is a test case test case which demonstrates this issue:
   http://www.bath.ac.uk/%7Epy8ieh/m/font-size-rounding-code.html

I will attach an image which shows Mozilla's current (2000021916) behaviour with
the pref turned off, the pref turned on (default), Nav4's behaviour on Windows, 
and MSIE5's behaviour on Windows.

With the pref turned off, we render: 
   font-size: 9px    as    9px text
   font-size: 8px    as    8px text (antialiased)
   <font size=1>     as    10px text

With the pref turned on (rounding), we render: 
   font-size: 9px    as    8px text
   font-size: 8px    as    8px text (antialiased)
   <font size=1>     as    10px text

Navigator 4 on Windows renders:
   font-size: 9px    as    8px text
   font-size: 8px    as    8px text (antialiased)
   <font size=1>     as    9px text (but spaced out a little)

Internet Explorer 5 renders:
   font-size: 9px    as    9px text 
   font-size: 8px    as    8px text (antialiased)
   <font size=1>     as    9px text (but spaced out a little)

I'm *guessing*, but I may be wrong, that The Todd Fahrner Algorithm (as given
in bug 18136) will fix the problem with <font size=1> displaying at a bigger
size than in legacy browsers. However, it should be noted that the rounding code
itself is NOT the cause of this discrepancy! I seem to recall it was being 
blamed for larger fonts at some point.

Other bugs which say <font size=1> is too big or suchlike are: 
   bug 14033, Extra whitespace between <FONT size=1> lines
   bug 21920, <font size=1> rendered too big 
   bug 18756, Tables being drawn incorrectly, maybe <font size=1> too big

After verifying that those bugs are indeed all caused by the same root cause,
I will mark them as duplicated of some other bug... but what is the master bug
for tracking the problem of <font size=1> being too small? 

Comment 7

[Hixie (not reading bugmail)]

Attachment: How the test case looks in Moz without and with font rounding, in WinNav4, and in WinIE5.

Comment 8

[Erik van der Poel]

I just received 2 long emails from Peter, regarding font size rounding on Win.
I need some time to read and understand them, but I thought I'd let the Cc'ees
on this bug all know about it. I will paste the info into this bug report later.

Comment 9

[Erik van der Poel]

Peter explained that his "proper Windows rounding" code was intended to round
the point size down to the nearest integer, since he believed that all Windows
apps did this. The Windows API documentation says something about using a
routine called MulDiv to compute the font size (lfHeight), and that MulDiv takes
integers as arguments. He also said that IE4 and Word rounded point sizes down.

However, I just tried it in WinIE5, and it did not round point sizes down to
the nearest integer.

So I think we need to reinstate Peter's original Nav4-compatible rounding code,
by uncommenting it, and commenting out the "proper Windows rounding" line. This
would give our pref font.size.nav4rounding its intended meaning. This would also
solve all of our current <font size="1"> problems.

RickG, I hereby ask for permission to check the fix into M14.

Later, we need to get Pierre to check in his code based on Todd's proposal.
That is a separate issue.

Comment 10

[Erik van der Poel]

The Top 100 sites look good with the Nav4 rounding code reinstated, but the UI's
fonts (menus, URL bar) look too small...

I tried bumping up the UI font from 3mm to 9pt, and that was better, but that
would mean making a cross-platform change to compensate for a Windows change...

The UI fonts' 3mm would look better if we didn't round the font sizes at all,
but to make that the default, we would need Pierre's new code based on Todd's
proposal...

Maybe we should just wait for Pierre's code to land.

Comment 11

[Hixie (not reading bugmail)]

Erik wrote:
> So I think we need to reinstate Peter's original Nav4-compatible rounding 
> code, by uncommenting it, and commenting out the "proper Windows rounding" 
> line. This would ... solve all of our current <font size="1"> problems.

Would it? i.e., Have you checked or is this a guess?
Could you provide a screenshot of:
   http://www.bath.ac.uk/%7Epy8ieh/m/font-size-rounding-code.html
...with the True Nav4 Rounding Code enabled?

Comment 12

[Erik van der Poel]

I'm guessing that it would fix the <font size="1"> problems. I'll attach a
screenshot...

Comment 13

[Erik van der Poel]

Attachment: screenshot of WinMoz with Nav4 rounding

Comment 14

[Erik van der Poel]

I'm running NT4, not Win98, so no anti-aliasing, by the way.

Comment 15

[Hixie (not reading bugmail)]

It does indeed fix the problem.

So we have a problem... either we wait for Pierre's code, which should
presumably get <font size=1> to work the same as legacy browsers, while at the
same time working without the rounding code, or we have <font size=1> bigger
than legacy browsers, or we have inconsistent UI sizes.

My vote is with waiting for Pierre's code and having larger <font size=1> in
the meantime. (i.e., status quo) When we check in Pierre's code, I suggest we
should also disable rounding by default at the same time (right?).

Bug 18136 is the one which covers Pierre's coding of Todd's algorithm.

Comment 16

[Erik van der Poel]

No, initially, Pierre will check in his code with the pref defaulting to true,
i.e. the old HTML 1-7 system. Then we will test it with the pref set to false,
to see if the new code is ready to be turned on by default:

  user_pref("font.size.nav4algorithm", false);

Comment 17

[leger]

ccing - hyatt, hangas, ben, matt, german, syd

We will take this change for beta - putting on PDT+ radar for beta1.  We want 
all to review it when it goes in ASAP though to make sure UI looks good.

Comment 18

[Hixie (not reading bugmail)]

Erik: Right -- so Pierre's initial check-in of Todd's algorithm should have no 
effect except if you change the prefs, right?

So if I understand this correctly the options are as follows:

           |    Pierre's Coding of Todd's Algorithm   |
           | Not checked |          Checked in        |
           | in (as now) | but disabled | and enabled |
  ---------+----------------------------+-------------+
  Peter's  | *       <font> bad         | <font> good |
  Rounding |           UI good          | UI good     |
  (as now) |           9px bad          | 9px bad     |
  ---------+----------------------------+-------------+
    Nav4   |         <font> good        | <font> good |
  Rounding |           UI bad           | UI bad      |
           |           9px bad          | 9px bad     |
  ---------+----------------------------+-------------+
     No    |         <font> bad         | <font> good |
  Rounding |           UI good          | UI good     |
           |           9px good         | 9px good  # |
  ---------+----------------------------+-------------+
          where: * = where we are now
                 # = where we would like to go

Is that a correct summary of the situation?

Comment 19

[s.marshall]

Sorry for spam - I assume everyone is already aware that this bug affects many
px sizes (odd sizes up to 13px, then even sizes thereafter), not just 9px. 18px
arial is particularly bad. I made up a <a
href="http://www.leafdigital.com/mozilla/pxsize.html">test case</a> with the
actual current behaviour compared to a gif of IE5 output. (from Win NT)

Comment 20

[leger]

Must fix by 03/03 to make the beta1 train.

Comment 21

[Erik van der Poel]

Pierre flipped the pref, so that we don't round by default now.

Comment 22

[Chris Petersen]

Text looks fine to me in the latest build. Fixed in the Feb 29th build.