[reportlab-users] Table sizing algorithms - comments welcome

[Andy Robinson]

I'm working on a change to table sizing algorithms,
driven by some urgent requirements at a customer site.

The current algorithm for determinng column widths i
like this:

1. if they are all given, use them
2. if any column has a width of None, or if you
did not give any widths, try to 'size' each
table cell. Strings (multiline or single line)
can always be sized).
3. if a column has a width of None, and any cell
in it contains a flowable or list of flowables
(e.g. a Paragraph), raise an exception.

The problem with paragraphs is that the height and
width cannot be known; one is a function of the
other.  Some much smarter algorithm would be needed
to work out the sizes.  Do you pick widths to make


I suggest to fix (3) by doing this:
1. as before
2. as before

3. classify all columns inro 3 groups:
   - size given by user
   - size computable 
   - size not computable
   Take the width of the containing frame,
   subtract the widths we know, and divide 
   the rest evenly between the noncumptable
   columns.
   
A second wrinkle is to correctly work out widths
within spanned row and column ranges.

This is fairly dumb but it means you will
at least get a table out and not a traceback,
and it will not be a significant slowdown.
IMHO it's reasonable to say that if you don't
specify the space, we'll divide it up evenly.

Note that it introduces a dependency betweenm tables
and their ontainers.  Until now tables have had
a "known size" which is nothing to do with the
frame - they make themselves big enough for the
content and that's all.  If you have a table
with paragraphs and no column widths, it will
now attempt to "fill the frame" - except for the
(user-settable) 6 point margin down the sides.


I won't check in for a day or two as I want
to be very careful that existing stuff is not
affected and write more test cases.

But if anyone has thoughts on a 'smart' way
to assign a width to a paragraph cheaply, 
please let me know!

- Andy

p.s. the "guiding principle" on tables is
basically still the same and can be summed
up as follows:
1. if you possibly can, specify all the widths
  explicitly
2. only set the row heights if the standard
  algorithm does not do what you want.