types, and mColGroups. Besides being cleaner and easier to maintain, this should speed some things up slightly
because I don't have to check display types every time I iterate through colgroup frames. I *know*
mColGroups contains only colgroups. I might do the same for rowgroups (that is, keep unknown frame types
in their own list as well.) But that's optional.
the inner table frame used to incorrectly create actual content objects for anonymous colgroups and cols. Now,
it just creates frames as appropriate, not content. To support this, I added some pseudo style contexts and
related atoms.
C++ compiler is strict about getting 'const' right. Most of the changes
are to add 'const' to the type casts for the second argument of the
GetStyleData method, which wants a const reference.
1. width=0 is treated as width=auto, not width=min
2. empty cells <td></td> are assigned a width of (2*borderWidth + 2*cellspacing + 3) pixels
The "3" is a fudge factor added by Nav and IE.
the problem was the way table layout apportioned height from a rowspanning cell.
Now, we do 2 things correctly:
1) we handle rows with rowspanning cells and no other cells.
2) we divide the height of a rowspanning cell proportionately to the rows spanned, where the proportion is defined by
the max height of a cell in that row that has no effective rowspan.
the partial fix for bug 1431 has to do with dividing width between percent-width and fixed-width columns in the same
table. You won't see any better layout for this page (ww.boston.com) as a result of the code I added. The rest of this
fix is a bit risky, so I'm waiting until after stability push.
and/or to the width attribute on the table itself), the rule is to try to give the column the largest value
assigned to it by any cell intersecting that column.
Separate storage is maintained for a column's natural desired width (width from any source other than a cell
with a colspan) and the maximum contribution to the column's width from any cell with a colspan.
of cells with rowspans. Basically, I just override HandleEvent() in nsTableRowGroupFrame
to ask nsTableRowFrame::Contains(), rather than the default action which is to
just use the child's rect.
not like specified width (which you might logically expect since all column widths are specified, but sadly that's
not the way Nav or IE interprets this case. )
nsTableFrame's use of IsPseudoFrame has been removed.
ables were not correctly applying the min width of the content of cells
that had col spans in some cases. In this case, the nested table was specified
as being too narrow for the content. There were 4 columns each with width=25%
in the first row, and in the next row a single cell with colspan=4 and whose
content min size was wider than the specified table width.
now autowidth columns as assigned a width starting from their minimum. The remaining
space is proportionately divided among them, never allowing a column to go over its max (for
autowidth tables only.)
the layout strategy can now query the outer frame for the caption min width,
and use it to expand the table if necessary
did some related cleanup while I was in there.
table frames no longer assume they are n levels below the inner table frame
table frames no longer assume their child frames are always the expected type of
frame. Now they check the display type and use that as sufficient proof that the frame
is of the right type. This implies a heavy reliance on the frame construction code to do
the right thing and always stamp out the right kind of frame for a given display type.
from their respective InitalReflow() methods.
table-layout=fixed now completely skips pass1 reflow and doesn't require maxElementSize
information, saving a tremendous amount of time during initial reflow.
moved *IsSpannedInto and *HasSpanningCells into cellmap, giving us the oppurtunity
to easily cache results if we want. These methods are still accessed through nsTableFrame.
changed nsTableFrame::GetCellAt to nsTableFrame::GetCellFrameAt to make it more clear what
the method returns: a frame, not content.
nsHTMLAtoms.cpp,h added some HTML 4 strings for tables
nsHTMLParts.h added constructors for new table content classes
nsGenericHTMLElement.cpp,h added table frame creation, common table attribute parsing
nsHTMLTable*Element.cpp these are the new table content classes
nsHTMLContentSink.cpp enabled the new content and removed the old
nsHTMLStyleSheet.cpp fixed the table style optimizations to work with new content
nsTable*Frame.cpp these work with the new content
these fixes are for nested tables with percent widths, for tables including both rowspan and colspan,
and for tables with only a single column.
some rounding errors were eliminated.
more debugging output, and some new assertions.
1) sometimes we were adding 1 too many pixels in constrained tables
2) sometimes we were using the wrong value for the max width as an input to constrained tables
aFilter is useful for determining if a frame should output itself during
a List(). In any event, the children of the frame are always processed.
added nsIListFilter, an interface for any object that wants to act like a filter for frame dumps
added static NS_LAYOUT nsIListFilter * GetFilter(nsString *aFilterName);
currently, only TableListFilter implements nsIListFilter.
in response to List(...TableListFilter), the frame model is dumped in a way that is useful
for table layout regression testing. Currently this is controlled by command line switches on the viewer app.
cells with colspans and specified widths divide their width provisionally between the columns they span. This provisional division is ammended by any cell in any of the effected columns that:
1) has a colspan=1, and
2) has a width specified
the widths of all other columns must be adjusted for cells like these. Further complicating matters is that it seems <faith> that if multiple colspanning cells intersect a column and each has a width
specified such that the column would have a different width depending on which is used, only the first such cell is used </faith>. I can neither confirm nor deny this reading the mozilla code.
I also taught the cell map how to deal with this situation:
colspan=1 | colspan=2 | colspan=1
colspan=1 | colspan=2 | colspan=1
This table "really" has 4 columns, but the middle cells are treated as if they have no colspan.
Trust me, it matters. The original colspan attributes can't simply be thrown away because Mr. DOM
could come along at any time and add/remove a cell that would make them important.
* fixes sample6, yahoo resizing
* added BasicTableLayoutStrategy::ColIsSpecifiedAsMinimumWidth helper function,
which tells us if the given column has a width attribute that means "make me as small as possible"
* nsTableFrame methods now no longer use mCellMap directly, they ask for it from
GetCellMap which gets the cell map from the first-in-flow. Only the first-in-flow has a
cell map because the cellmap refers to the table as a single object, not to each table frame.
* Likewise, continuing table frames do not call BalanceColumnWidths or SetTableWidth
which are "global" operations. Responsibility for these lies with the first-in-flow.
buster%netscape.com
harishd%netscape.com
wtc%netscape.com
troy%netscape.com
alecf%netscape.com
kipp%netscape.com
joki%netscape.com
michaelp%netscape.com
peterl%netscape.com
spider%netscape.com
dcone%netscape.com
pierre%netscape.com