Why isn't my exception section catching my error?

I got an interesting email today from a reader of one of my PL/SQL 101" articles for Oracle Magazine, Building with Blocks.

Q. had taken the code from the article, made some changes, tried to run them, and got very confused. He wrote:

When I run this code, I see "Hello World".

DECLARE
  l_message VARCHAR2(100) := 'Hello World!';
BEGIN
  DBMS_OUTPUT.put_line (l_message);
EXCEPTION
  WHEN OTHERS
  THEN
    DBMS_OUTPUT.put_line ('Error='||SQLERRM);
END;
/

Hello World!

When I change the block to make the l_message variable too small for its string, I see the VALUE_ERROR error message.

DECLARE
  l_message VARCHAR2(10);
BEGIN
  l_message := 'Hello World!';
  DBMS_OUTPUT.put_line (l_message);
EXCEPTION
  WHEN OTHERS
  THEN
    DBMS_OUTPUT.put_line ('Error='||SQLERRM);
END;
/

Error=ORA-06502: PL/SQL: numeric or value error: character string buffer too small 

But when I change the name of the variables inside the call to DBMS_OUTPUT.PUT_LINE to "l_message1", the exception handler is not displaying the error message, instead I see the following:
When I change the block to make the l_message variable too small for its string, I see the VALUE_ERROR error message.

DECLARE
  l_message VARCHAR2(10) := 'Hello World!';
BEGIN
  DBMS_OUTPUT.put_line (l_message1);
EXCEPTION
  WHEN OTHERS
  THEN
    DBMS_OUTPUT.put_line ('Error='||SQLERRM);
END;
/

ORA-06550: line 5, column 25:
PLS-00201: identifier 'L_MESSAGE1' must be declared 

Why am I seeing this inconsistent behavior?  

I was confused about the source of his confusion, but after a couple of back-and-forth emails (what you read above is the cleaned-up version of that back-and-forth), the light bulb lit in my brain. Which enabled me to clear up his confusion, and inspire me to write a blog post, in case the same confusion was confusing anyone else.

Here's the most important thing to remember:

Exception sections handle exceptions raised when the block is executed. But first that block must be compiled!

When you are writing code within stored program units, like procedures and functions, this is clearly a two stage process:

1. Compile
2. Execute

But when you are working with an anonymous block, all you do is execute the block. So you can certainly be forgiven for thinking that anything that goes wrong will be caught in the exception handler.

The reality is different however. When you run an anonymous block, the PL/SQL engine will first of all parse and compile the block. If all goes well, then the PL/SQL runtime engine will execute the compiled code.

If, however, your code fails to compile, well....you will get a compile error, rather than an runtime exception.

How do you know you've gotten a compile error? Check the prefix. If it is "PLS", something went kablooey at compile time. Your code cannot be executed. If the prefix is "PLW", that's a compile-time warning, which means that the code compiled, but the PL/SQL engine has suggestions for improving it. And if the prefix is "ORA", ah, then you've got a runtime error - an exception.

Only in that last case with the exception handler, if present, be able to trap the exception.

I hope that makes thing nice and clear. Any comments or questions? :-)