The problem: runtime error handling in LSL is… well, it’s practically nonexistant. If something goes wrong, the script will usually shout some message to the world and then die. The obvious problem there is that you have to then manually restart the script and lose all data! Also, for the security-concious, it can potentially reveal implementation details of your script.
Idea: a new event in LSL, error, which would allow scripts to intercept and deal with error messages in a sane and consistent manner. Here’s an example:
default
{
state_entry()
{
integer a = 1/0;
}
error( integer type, string message )
{
if( type == ERROR_MATH )
{
llOwnerSay( message + "You should have paid more attention in math class." );
}
}
}
The two parameters for the error event are the type, which is one of a selection of integer constants, and the message, which is a human-readable message describing what went wrong. In this case, the type is ERROR_MATH (indicating a math error), and the message would be something like “Error: Division by zero.”. The output from this script would be a message to the owner, “Error: Division by zero. You should have paid more attention in math class.”
Some more useful (and less insulting) ways of handling errors might be to reset the script, send an IM to an assistant shopkeep, clearing some buffers to free up memory, giving a custom message on chat, or some other appropriate behavior for that particular application. If the script doesn’t define the event block, errors are reported in the default way, as they are now (e.g. shouting the error on the debug channel).
To complement the new error event, would be a new function for triggering errors: llError( integer type, string message ). When called, it triggers an error of the given type, using the given message. If an error event block is defined, the error is given to that block to be handled; if there’s no block, it’s handled in the default manner for that event type.
A second new function would be llNoHandleError, which behaves like llError, except that it always performs the default handling, bypassing the error event block if it exists. This could also be used within the error event block to “pass through” event types that it’s not intended to handle.
So, scripters: give me some feedback on this. Should we make a JIRA for this, or is it actually a really stupid idea?
