More on counting upper case characters

Question

(*This relates to a recent question I posted about counting upper case characters in a string. dev_sleidy responded with an applescript approach that works, but at 6 lines I was hoping for more compact code. So I played around with a one-line command line approach that so far seems reliable (tell me if you find it isn't). However, there are a couple peculiarities which I would like to understand. I know, I know, I should just shut up and use it. But I must be part cat and need to be killed.
The code that appears to work:*)
set T to "MY GraNd QuestioN A"
set UCcount to do shell script "echo " & quoted form of T & " | tr -d [[:space:]] | tr -d [:lower:] | wc -m"
(*The peculiarities:
1) If I substitute >>sed 's/ //g'<< for >>tr -d [[:space:]]<<, the result always seems to be one count high. (Same thing appears to happen when executed in Terminal.) Why?
2) When using regular expressions the format [[:class:]] has always worked for me, as it does for "space" in the code above. But for the second tr -d pipe (above code) those outside brackets mess everything up, and single brackets work fine. Same thing happens with both [[:lower:]] and [[:upper:]] but doesn't seem to happen when executed in Terminal. Why? Are there rules for bracketing character classes (I've seen different versions)?
Or is all this a UNIX forum question?*)

Answer 1

Hello
Honestly I don't understand why such result is returned by CODE1.
Its prototype [1] tested here works as expected.
But I noticed CODE2 may return wrong result, for its prototype [2] actually returns wrong result with MacPerl 5.6.1r2 under OS9.1 when input string is read from file insead of being given as string literal. I didn't test that possibility. Sorry for that. Please disregard CODE2, for I think CODE1 will be enough.
(The wrong result is caused by the fact that 'Ç' encoded in UTF-8 is counted as 2 chararcters by length function if it is read from file due to certain limitation of MacPerl 5.6.1r2. I don't know whether that applies to recent Perl under OSX.)
[1]
use utf8;
while (<>){
s/[^[:upper:]]//g;
} continue {
print;
[2]
use utf8;
local $/;
while (<>){
s/[^[:upper:]]//g;
print length;
Re. your original question about peculiarities.
As for 1), I think it is because 'echo' is adding trailing \n and "sed 's/ //g'" won't remove it. Try "sed 's/[[:space:]]//g'"  instead. Indeed "sed 's/[^[:upper:]]//g' will remove all non-upper case letters at once, I guess.
As for 2), I think it is because 'tr' actually does not process regexp. Its character class support is based upon POSIX but the usage is different from regexp, i.e. [:class:] in 'tr', while [[:class:]] in proper regexp. Meanwhile, I think, the regexp syntax, [[:class:]], in 'tr' argument works in Terminal because in Terminal, 'bash' processes the regexp in its parameter expansion (path name expansion, in this case) and 'tr' receives the expanded result as its argument.
cf.
http://developer.apple.com/documentation/Darwin/Reference/ManPages/man1/tr.1.htm l
http://developer.apple.com/documentation/Darwin/Reference/ManPages/man1/bash.1.h tml
Hope this may be of some help,
H