[cl-typesetting-devel] Page n of m?

[Bob Hutchison]

On Oct 1, 2004, at 3:18 PM, Peter Seibel wrote:
>
>> Checking if the x and y coordonates have changed is probably too
>> strict and will give many redefined refs IMO.
>
> Yes. I did have to bump up *max-number-of-passes* to get it to
> actually settle down. (Took three passes instead of two.)
>
>

I hate to bring this up, but it would be very good to minimise the 
number of passes. I've attached the output of extended-time (lispworks 
on a mac) for a 173 page test. The elapsed time was almost 4 minutes 
(more like 3.5 minutes of actual work though). This is already a lot of 
time, I think, and a third pass would take another couple of minutes. 
More importantly is that half the execution time is in the garbage 
collector with 1.3 *billion* bytes allocated to generate a 1.5 million 
byte pdf -- a 1000x factor. (and I've done some things to try to reduce 
the amount of memory used)

user time    =    204.700
system time  =      3.500
Elapsed time =   0:03:52
Allocation   = 1,288,097,632 bytes
0 Page faults

                                        total    /   user     /   system
  total gc activity              =    101.130000 /  99.300000 /   
1.830000
  main promote (   43 calls)     =     49.860000 /  48.670000 /   
1.190000
  mark and sweep (  250 calls)   =     50.230000 /  49.600000 /   
0.630000
  internal promote (   16 calls) =     16.640000 /  15.750000 /   
0.890000
  promote (    0 calls)          =      0.000000 /   0.000000 /   
0.000000
  fixup (  128 calls)            =     24.740000 /  24.560000 /   
0.180000
  compact (   19 calls)          =      1.040000 /   1.030000 /   
0.010000