Fix `test/error`, and fix a (long) sequence of bugs that were uncovered...

[jpellegrini]

@egallesio -- In tests/test.stk, the macro test/error was ignoring errors, as it should. However, it was also ignoring the code to be tested, completely!

So I wrote a new macro (using define-macro) that uses with-handler to capture errors. Then I ran the test suit, and... Well, there were a lot of new errors!

One of the problems I have already sent separately (PR #546 -- the bytevector problem), but then there were so many others that I thought I'd keep them in one single PR.

Besides the several simple error fixes, I have three comments, below.

I'll be away for some days, probably...

1. Shouldn't a+0.0i be considered complex?

STklos treats it as the real number a, but... What if that 0.0 imaginary part came from an inexact computation that underflowed? I see several Schemes do:

• a+0i = a (real)
• a+0.0i = a+0.0i (complex) So I changed this. If it's not OK, I can revert it. (tell me)

EDIT: Ok, I found it in R7RS (section 6.2.6):

(real? -2.5+0.0i) => #f

So it seems like my intuition was in accordance with the standard...

2. We cannot test syntax errors

with-handle, guard, etc will not catch syntax errors, or compile-time errors at all. So I commented out several of those.

3. Misterious bug

One test fails when I run the test suit, but works fine on the REPL. (It's the guard test adapted from Gauche). Do we comment it out?

[egallesio]

Hi @jpellegrini,

I'm working on this PR. Correcting all the problems gave you a lot of work. Thanks. BTW, I have added a test/compile-error for the tests you had to comment.

[egallesio]

I have finally finished it.

Shouldn't a+0.0i be considered complex?

Yes! I have taken as is your code.

We cannot test syntax errors.

Now, we can :wink: In fact I use eval to force a compilation and test if it yields an error. BTW, the initial version of test/error was almot OK. It was:

(define-syntax test/error
  (syntax-rules ()
    ((_ str code)
     (test str *test-failed* result))))

The last parameter should be code in place of result . Consquently, I have not used your fix.

Misterious bug

Not seen it. :smiley:

[jpellegrini]

Great @egallesio !