<antocuni>
the tests pass on my local machine, so it's already a progress compared to the current state
<ronan>
it's progress indeed. But I remain globally unhappy with the exception hack
<antocuni>
I agree it's sub optimal, but what do you propose instead?
<antocuni>
instead of the exception hack, I mean
<ronan>
doing it the cpyext way
<ronan>
and if we share the error indicator with cpyext, we get to support mixed code automatically
<antocuni>
I'm fine as long as it doesn't impact the performance (and I fear it will)
<ronan>
we need to properly benchmark the 2 approaches, otherwise we'll never resolve the issue
<antocuni>
agreed
<antocuni>
basically, my main concern is that we should try to avoid one of the cpyext mistakes, which was "let's make it working now and think about performance later"
<antocuni>
so, I am fine with any solution as long as we keep performance in mind, because cpyext proves that the "think later" approach doesn't work well
<ronan>
agreed, but I fear we're falling into the opposite trap which is premature optimisation