Re: [PATCH] srfi-64: fix unused variable warnings

[Maxime Devos]

On Thu, 2021-04-01 at 23:12 -0700, Aleix Conchillo Flaqué wrote:
> Hi Maxime,
> 
> Thank you for your comments!
> 
> On Thu, Apr 1, 2021 at 4:37 AM Maxime Devos <maximedevos@telenet.be> wrote:
> 
> > For example, in:
> > 
> > >  (define (%test-comp2 comp x)
> > >      (syntax-case (list x (list (syntax quote) (%test-source-line2 x)) 
> > > comp) ()
> > >        (((mac tname expected expr) line comp)
> > >         (syntax
> > > -     (let* ((r (test-runner-get))
> > > -            (name tname))
> > > +     (let ((r (test-runner-get)))
> > >         (test-result-alist! r (cons (cons 'test-name tname) line))
> > >         (%test-comp2body r comp expected expr))))
> > 
> > I would keep the let* (but reverse the binding order), but change 'tname'
> > with 'name' in the call to 'test-result-alist!', such that 'test-X' macros
> > behave somewhat more like procedure calls (except for installing exeption
> > handlers and having access to the s-expression of the code that will be run,
> > of course).  It's largely a matter of taste, though.
> > 
> 
> I've done this change. One thing I don't understand is the "reverse
> the binding order", I've done it as suggested but is this change the
> one you refer to as "matter of taste"?

Yes, that's the change I was referring to.  As to why: a procedural
equivalent of 'test-assert would look more or less like

;; (possibly more arguments are required)
(define* (test-assert* name thunk expression)
  ;; THUNK: when called, return something that will be
  ;;   used as true/false.
  ;; EXPRESSION: S-expression representing the body
  ;;  of THUNK
  (let ((r (test-runner-get)))
    ;; evaluate (thunk) here within some exception
    ;; handlers and use r
    ...))

(Similar equivalents to test-equal, test-eq ... can be written
as well.)

Suppose  '(test-assert* NAME (lambda () EXP) 'EXP) is evaluated.

Then first NAME is evaluated, which can have side-effects.  The
lambda expression and (quote EXP) are evaluated as well, but no
side-effects are possible here (aside for allocating some memory
for the thunk, which can lead to a 'out-of-memory exception, but
that's usually simply ignored).

Only after the arguments are evaluated will '(test-runner-get)
be evaluated.

However, for the original test-assert macro, the evaluation
order is different.  From a REPL:

> (use-modules (srfi srfi-26))
> ,expand (test-assert NAME EXP)
;; output manually cleaned up
$6 = (let* ((r (test-runner-get))
            (name NAME))
       more code ....)

It should be clear that here 'NAME is evaluated *after*
'(test-runner-get) is evaluated, unlike for the 'test-assert*
procedure.

That said, SRFI-64 does not require NAME to be evaluated even
if trying to get the test runner fails for some reason, I
don't think anyone ever changes the test runner from within
a ‘call’ (not really a call as test-assert is a macro) to
test-assert, and in practice NAME is a constant, so in practice
it doesn't really matter in what order things are evaluated.

Also see next comment:

> > In any case, it is good that 'tname' is now evaluated only once, as per
> > SRFI-64 (notice ***It is evaluated only once.*** (markup mine)):
> > 
> > [...]
> Yes, this makes sense. Thanks again for pointing that out.

This is done correctly in the new patch (and the old patch IIRC).
Also, by reversing the binding order from

-       (let* ((r (test-runner-get))
-              (name tname))

to

+       (let* ((name tname)
+               (r (test-runner-get)))

the expression tname is also evaluated *at least* once,
thus TNAME is evaluated *exactly* once, which seems like
a nice property to have, though this is a bit stricter
than SRFI-64 demands IIUC.

Also, the formatting seems to have gone wrong.
Shouldn't this be

+       (let* ((name tname)
+              (r (test-runner-get)))

?  If in Emacs, I recommend scheme-mode, in which case pressing tab on the
second line would produce the desired formatting.  Alternatively, select
a region of text and presss tab.

> > As this patch does not ‘merely’ fix a warnings, but fixes a bug, could you 
> > change
> > the patch message accordingly?  Something like
> > 
> >   srfi-64: fix double evaluation of test-name.
> > 
> > perhaps?
> > 

The revised commit message looks good to me.

Greetings,
Maxime.
p.s: I'm not a guile maintainer so you will have to wait on
someone else to actually merge this.

signature.asc
Description: This is a digitally signed message part