Re: [RFC] logic change in m2-valprint.c

[msnyder at sonic dot net]

> On Sun, Jul 01, 2007 at 08:54:19AM -0700, Michael Snyder wrote:
>> > It can't be backwards; this is trying to print "1" or "1, 2" but your
>> > change would make it print ", 1" and ", 12".
>> >
>> > I think it's also trying to shorten ranges to "{1..3, 6..7}".  The
>> > bug's got to be in there somewhere.  Why's the code dead?  I'm not
>> > seeing it... element_seen can be reset by the bit clear case, and then
>> > we'll get into the test you're changing again.
>>
>> 'Cause we're not in a loop, and it's not a static variable.
>> The code is serial.  At entry we set empty_set to one, and
>> then we test to see if it's zero.  It can't be zero.
>
> How sure are you sure we're not in a loop? :-)

I'm sorry -- too many balls in the air.  Try this:

91   	  int empty_set        = 1;
92   	  int element_seen     = 0;
...
123  	      for (i = low_bound; i <= high_bound; i++)
...
133  		      if (! element_seen)
134  			{
135  			  if (! empty_set)
136  			    fprintf_filtered (stream, ", ");
137  			  print_type_scalar (target, i, stream);
138  			  empty_set    = 0;
139  			  element_seen = 1;
140  			  previous_low = i;
141  			}

So, the reasoning is as follows: line 136 is dead code, because
line 135 can never test true, because the first time we enter
the block beginning at line 133, empty_set will be true (there
is nowhere else for it to be set false), and there is no second
time -- we will never enter this block again because we will set
"element_seen" to true (and there is nowhere else for it to be
set false again).

Was that clear?

So on pass "n", we enter the block.  Empty_set must be true.
We now set empty set false, and element_seen to true, and
therefore on pass "n + m" we cannot enter the block.  So there
is no point at which we can execute line 136.