|
|
Boost Testing : |
From: Stefan Slapeta (stefan_at_[hidden])
Date: 2006-06-23 07:23:50
Sean Huang wrote:
>
>> For one thing I am not sure that this is actually the right fix for the
>> issue but anyway.
> What do you think is the right fix? I know very little about BBv2 and bjam
> but my original message was posted about 1 week ago.
>
Maybe somebody from the boost.build group can tell us if this is the
right procedure?
>> The more important issue is that you changed the toolset name for a
>> required platform for the upcoming release. This is, well undesirable.
> Unless you decide IC9.1+vc8 is not a required platform, the toolset name
> needs to be changed. I think one of the major reasons people upgrade to
> IC9.1 is to get vc8 compatibility.
>
Sean is right here. We definitely need something to distinguish between
the VC base in the toolset. Though it's currently not easily possible
without manual intervention to change this base dynamically on a machine
(Intel asks for compatibility during the installation and "burns" this
into iclvars.bat), we really have to test Intel now based on both VC 7.1
and VC 8. And we need to make this difference visible in the toolset
name somehow.
I've always been testing with vc 7.1 only because changing between 7.1
and 8 is quite difficult and error prone (and it needs 2 completely
separate test runs).
>> There are two ways to resolve this either revert the change or fix all
>> references in status/explicit-failures-markup.xml.
> Let's do the right thing.
>
- use the vc71 name for errors related to intel on vc71
(I not aware of such)
- use the vc8 name for errors related to intel on vc8
(must be at least those caused by the well-known vc8
library bugs; they also occur on plain vc8 only)
- use wildcards for errors related to intel compiler
Stefan