In article <[EMAIL PROTECTED]>, Rick Myers <[EMAIL PROTECTED]> writes: > On Apr 13, 2002 at 11:45:33 +0000, Ton Hospel wrote: >> >> The really interesting entry is in fact the one by Rick Myers >> because it can be trivially improved to 54: >> >> -l for$.(++$_..pop){s!.!$.*"$&.$'"%($`?10:$^T)!eg}print > > Hmm.. Well, the ++$_ would've been good for 60, but the twist > of moving $& inside the double quotes breaks for me with > input values >= 45. > > The problem is that as $& grows in length, it pushes the > trailing digits of $' past were perl will "see" them when it > converts the value to float (for the *). The conversion back > to int (by the %) then causes an off-by-one type rounding > error. > > In other words, at 45, for the last non-zero digit... > > $.*$&+$.*".$'" == 126 > > $.*"$&.$'" == 125.9999999999999857891452847979962825775146484375 >
Interesting, so it in fact depends critically on the number of significant digits in a double. So we have two 60's that you can really run through the test program. Still, is there a way to fudge it and work around the float mess...
