2014-12-26 23:07 GMT+01:00 Cl��ment Bera <bera.clement@gmail.com>:

2014-12-26 22:30 GMT+01:00 Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com>:

2014-12-26 19:02 GMT+01:00 Cl��ment Bera <bera.clement@gmail.com>:

Hello,

It does not happen with the Cog VM nor with older Pharo VM.

It's a VM bug and it's specific to the Pharo-VM fork.

Most probably it's because of the float refactoring for Smallfloat some stuff got incorrectly merged.

Please consider writing a test for that in FloatTest and make a slice for Pharo if you have some time.

Thanks Clement,

there's already a test that is failing (FloatTest>>testNaNCompare).

What I find amazing is that I did not find this failure on the CI server...

That's because the CI server does not use the latest VM, but the stable one...��

Ah OK, the CI server is single bleeding-edge rather than double (VM+image) which might be a reasonnable choice for separating concerns.
But that was misleading!
If downloading the artefact from the CI server serves you a different version than tested...

��

2014-12-26 18:35 GMT+01:00 Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com>:

---------- Forwarded message ----------
From: Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com>
Date: 2014-12-26 16:52 GMT+01:00
Subject: regression of NaN comparison
To: Pharo Development <Pharo-project@lists.gforge.inria.fr>

I downloaded latest Pharo4.0-mac.zip from https://ci.inria.fr/pharo/
and I experienced a resurgence of this old bug:

2<Float nan.

This is not one of the errors reported in the CI server.
Any clue?