[Axiom-developer] (no subject)

[daly]

>>In general, I fee that instead of guessing what given construct
>> means you should just throw errors; once an unhandled construct
>> appears it is easy to catch errors in the debugger and find out
>> what it means. If you try to mask errors debugging becomes much
>> harder

> That's OK with me. ...

I'm a big fan of "exact debugging". That is, give enough unique information
to make it perfectly clear where and why the error occurs. 
For instance, you'll see
   
  @echo NNN this stanza does this

all over the Makefiles so I can instantly pinpoint where the problem occurs.
Once I know where and once I have a failing input it usually takes only a
few minutes to fix.

As Waldek says, if there is wrong-but-not-failing output then I have to
understand the whole architecture of how MathML support works to try to
find the failure.

So something like: 

  ERROR NNN: I got <this> and was trying to do <this> in routine <here>

really narrows the scope of the problem. I don't need a debugger anymore.

Tim