From: Robert Klemme <shortcutter@...>
Date: 2008-04-26T21:45:03+09:00
Subject: Re: goto

On 26.04.2008 01:29, Simon Krahnke wrote:
> * Robert Klemme <shortcutter@googlemail.com> (08:32) schrieb:
> 
>> On 21.04.2008 23:28, Michael W. Ryder wrote:
>>> I chose this after reading about it in Knuth's 2nd volume.  I wanted
>>> to try to implement the binary greatest common divisor algorithm in
>>> Ruby. It took a little effort but I was able to implement it without
>>> any Go Tos.  It is hard to go from languages that require jumping
>>> around in the code to one that doesn't.  It's a totally different mind
>>> set.
>> And, does it feel better?  For me it does - for Knuth apparently not.  I
>> haven't felt the need for a goto in ages.  I also rarely use continue
>> and break in loops.  My impression is that quite a number of cases where
>> either is chosen can be greatly improved and actually simplified by
>> choosing a different loop condition.  My 0.02EUR...
> 
> Tell that Matz & Co.
> 
> ,----[ ruby-1.9$ grep -c goto *.c | grep -v 0 ]

Last time I checked C did not have exceptions.  Without looking at the 
exact locations that fact alone might attribute for quite a few of them:

File ruby/ext/openssl/ossl_ocsp.c
     5 100.00% err

File ruby/ext/openssl/ossl_pkcs12.c
     3 100.00% err

File ruby/ext/openssl/ossl_pkcs7.c
     3 100.00% err

File ruby/bignum.c
     8 72.73% bad
     1  9.09% bignum
     1  9.09% bigparse
     1  9.09% out_of_range

"goto", "break" et al aren't bad in itself like to tool is, but it is 
like a sharp knife which can do damage more easily when handled 
carelessly than an edgeless knife.  Also, often a sharp knife is not the 
right tool either, like, when you want to fasten a screw.

I do say though that a lot of code I see (and have to change) has 
"break" and "continue" in there and becomes easier to read and 
understand once you take them out.  I do have to concede that in the 
past of my company there was one guy who wrote exceptionally bad code.


Looking a bit further at label counts in goto.c:

File ruby/regex.c
    39 51.32% fail
     4  5.26% numeric_char
     4  5.26% invalid_escape
     4  5.26% nofinalize
     3  3.95% end_of_pattern
     3  3.95% restore_best_regs
     3  3.95% repeat
     3  3.95% normal_char
     2  2.63% too_big
     2  2.63% memory_exhausted
     2  2.63% invalid_pattern
     1  1.32% nested_meta
     1  1.32% on_failure
     1  1.32% unfetch_interval
     1  1.32% advance
     1  1.32% startpos_adjust
     1  1.32% begbuf_match
     1  1.32% range_retry

Of course, a state machine is special in a way.  Like a parser, which is 
typically generated:

File ruby/parse.c
     6  8.70% decode_num
     6  8.70% trailing_uc
     5  7.25% retry
     4  5.80% again
     4  5.80% eof
     3  4.35% yyerrlab
     3  4.35% yynewstate
     3  4.35% error
     3  4.35% yyoverflowlab
     2  2.90% yyerrorlab
     2  2.90% yyerrlab1
     2  2.90% id_regist
     2  2.90% gvar
     2  2.90% quoted
     2  2.90% ternary
     2  2.90% yyreduce
     2  2.90% escaped
     2  2.90% yybackup
     2  2.90% yyreturn
     2  2.90% id
     2  2.90% yydefault
     1  1.45% yysetstate
     1  1.45% new_id
     1  1.45% quotation
     1  1.45% yyacceptlab
     1  1.45% yyabortlab
     1  1.45% start_num
     1  1.45% octal_number
     1  1.45% newline

If you want to do it yourself, here's the quickly hacked gotolyzer(TM):

Robert@Babelfish2 /c/TEMP
$ cat gotolyzer.rb
#!/bin/env ruby

ARGV << "." if ARGV.empty?

ARGV.each do |dir|
   Dir[dir + "/**/*.{h,c}"].each do |file|
     cnt = Hash.new 0
     File.read(file).scan(%r{\bgoto\b\s+([^\s;]+)}) do |m|
       cnt[m] += 1
     end
     unless cnt.empty?
       puts "File #{file}"
       sum = 0
       cnt.sort_by {|k,v| sum += v; -v}.each do |label, count|
         printf "%5d %5.2f%% %s\n", count, count * 100.0 / sum, label
       end
       puts
     end
   end
end

Robert@Babelfish2 /c/TEMP
$

You couldn't do that as elegantly and shortly in C.  That's the kind of 
sacrifice Phlip mentioned.

Cheers

	robert