[Logiweb] pyk warning: * imply * mistakenly unfit?

[Klaus Ebbe Grue]

Hi Kasper,

> "[ math value define
>    var x imply var y
>  as
>    var x macro imply var y
> end define end math ]"
> ...
> * imply * mistakenly unfit?
> ...
> Does anyone know what the warning means?

It means that the definition is "unfit for optimization". In other words, 
it will be slow compared to a "fit" definition.

You can to three things:
(1) Ignore the message.
(2) Ignore the message + get rid of the message.
(3) Make the definition "fit".

Concerning (3). After macro expansion, "x imply y" has the property that 
if x is false then "x imply y" returns T without evaluating y. For that 
reason, "x imply y" is "non-strict" in y. Logiweb has a particularly 
efficient way of translating strict constructs. To make the definition 
above strict in y, replace "var x macro imply var y" by
   var y guard parenthesis var x macro imply var y end parenthesis
The "u guard v" constructs computes u, discards the value, computes v, and 
returns the value of v. It is convenient for making definitions strict.

Concerning (2). Very many constructs are "unfit". To avoid too many error 
messages, the compiler tries to guess whether or not a construct was 
intended to be "fit". To indicate its uncertainty, the compiler says
"construct xyzzy mistakenly unfit?".

At present, a function is "fit" if all constructs on the right hand side 
of the defintion are either "fit" themselves or parameters that occur on 
the left hand side. Furthermore, the construct being defined must be 
strict.

When the compiler deems a construct unfit, it looks at the root of the 
right hand side. If the root of the right hand side is unfit, the compiler 
thinks the construct is unfit on purpose and generates no error message. 
So the second approach is to put an unfit identity operator in the root, 
ie. to change "var x macro imply var y" to
   identity var x macro imply var y end identity
The "identity u end identity" construct is an unfit operation that just 
returns u (to see that it is unfit, you have to read the *full* story 
about fitness analysis in Sectin 3.6 of the base page).

If a page contains more than one unfit operation, the compiler just 
reports the number and refers you to look in a Lisp variable for the names 
of the unfit operations. That is unfortunate since, when the compiler is 
invoked from the Linux shell, the compiler quits Lisp before you have a 
chance to inspect that Lisp variable. I work on a solution.

-Klaus