Re: master 0da97d6: * src/keyboard.c (timer_check_2): Replace redundant test with assertion

Re: master 0da97d6: * src/keyboard.c (timer_check_2): Replace redundant test with assertion

[Eli Zaretskii]

IMNSHO, this change is dangerous: it replaces run-time conditions with
assertions, but IME most if not all of the active developers and
people who track the master branch don't compile with --enable-checking,
or don't run the trunk build on a regular basis.  Which means these
assertions, if they happen, will not be seen nor reported.

So basically you removed a safety net without any replacement.

I suggest to revert this change.  It gains us nothing of importance.

Re: master 0da97d6: * src/keyboard.c (timer_check_2): Replace redundant test with assertion

[Stefan Monnier]

Eli Zaretskii [2022-08-05 21:20:09] wrote:
> IMNSHO, this change is dangerous: it replaces run-time conditions with
> assertions, but IME most if not all of the active developers and
> people who track the master branch don't compile with --enable-checking,
> or don't run the trunk build on a regular basis.  Which means these
> assertions, if they happen, will not be seen nor reported.
>
> So basically you removed a safety net without any replacement.
>
> I suggest to revert this change.  It gains us nothing of importance.

I introduced a bug in my code which made the test fail, which made me
discover that:
- the test currently can never fail (because the code that leads to this
  point checks the condition earlier).
- if the test were to fail, Emacs would lock up in an inf-loop.
I kept an `eassert` basically to document the assumption.


        Stefan

Re: master 0da97d6: * src/keyboard.c (timer_check_2): Replace redundant test with assertion

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: emacs-devel@gnu.org
> Date: Fri, 05 Aug 2022 17:41:00 -0400
> 
> Eli Zaretskii [2022-08-05 21:20:09] wrote:
> > IMNSHO, this change is dangerous: it replaces run-time conditions with
> > assertions, but IME most if not all of the active developers and
> > people who track the master branch don't compile with --enable-checking,
> > or don't run the trunk build on a regular basis.  Which means these
> > assertions, if they happen, will not be seen nor reported.
> >
> > So basically you removed a safety net without any replacement.
> >
> > I suggest to revert this change.  It gains us nothing of importance.
> 
> I introduced a bug in my code which made the test fail, which made me
> discover that:
> - the test currently can never fail (because the code that leads to this
>   point checks the condition earlier).
> - if the test were to fail, Emacs would lock up in an inf-loop.

I'm not against the assertions, I'm against replacing run-time
conditions with assertions.  You can keep the assertion of you
reinstate the conditional, if it is important to you to flag when
these conditions happen.

> I kept an `eassert` basically to document the assumption.

There's no assumption here, not one that we want to bet our user's
session fate on.  We want to prevent bad, perhaps fatal, errors when
that condition is violated.