Module tests

Module tests

[Eli Zaretskii]

Commit 2b3f5de2b3 removed these 2 tests:

  (should (= (mod-test-sum 1 #x1fffffff)
             (1+  #x1fffffff)))
  (should (= (mod-test-sum -1 #x20000000)
             #x1fffffff))

They were written to test the code in emacs-module.c which switches
from scalar values to cons cells when the integer value is too large
for 32 bits and Emacs is compiled with wide ints.

I don't understand why the removal.  AFAICS, we still switch from
integers to cons cells, so the borderline values just above and below
the threshold are where be dragons, and we should keep them (they
helped me to find and fix a couple of bugs in the previous
implementation).  What am I missing?

Re: Module tests

[Paul Eggert]

Eli Zaretskii wrote:

> I don't understand why the removal.

Those tests failed on 32-bit Emacs configured --without-wide-int, as 1+ silently 
wraps around whereas mod-test-sum signals an integer overflow. Plus, on a 
32-bit-only Emacs #x20000000 evaluates to a floating-point number -- a flaw in 
Emacs, but is probably not something we can fix during a feature freeze.

Sorry, I hadn't realized the points of those tests, and thought my replacements 
were testing the same thing. I just now brought those tests back and ported them 
to 32-bit-only Emacs.

Re: Module tests

[Eli Zaretskii]

> Cc: emacs-devel@gnu.org
> From: Paul Eggert <eggert@cs.ucla.edu>
> Date: Sun, 6 Dec 2015 14:56:11 -0800
> 
> Eli Zaretskii wrote:
> 
> > I don't understand why the removal.
> 
> Those tests failed on 32-bit Emacs configured --without-wide-int, as 1+ silently 
> wraps around whereas mod-test-sum signals an integer overflow. Plus, on a 
> 32-bit-only Emacs #x20000000 evaluates to a floating-point number -- a flaw in 
> Emacs, but is probably not something we can fix during a feature freeze.
> 
> Sorry, I hadn't realized the points of those tests, and thought my replacements 
> were testing the same thing. I just now brought those tests back and ported them 
> to 32-bit-only Emacs.

Thanks!