Re: [Boost-users] [math] fpclassify.hpp fails to compile with Intel icc 11.1

Subject: Re: [Boost-users] [math] fpclassify.hpp fails to compile with Intel icc 11.1
From: Paul A. Bristow (pbristow_at_[hidden])
Date: 2011-01-05 06:49:55

> -----Original Message-----
> From: boost-users-bounces_at_[hidden] [mailto:boost-users-
> bounces_at_[hidden]] On Behalf Of Gabriel Redner
> Sent: Tuesday, January 04, 2011 3:16 AM
> To: boost-users_at_[hidden]
> Subject: Re: [Boost-users] [math] fpclassify.hpp fails to compile with
Intel icc
> 11.1
> 
> Hi Gabe
> 
> In this case, no reason at all.  1.42 is the version that comes
prepackaged for
> Ubuntu, so that's what's I had installed.  However, this is just a small
academic
> project, so there was really no barrier to downloading 1.45, dumping it
into our
> source tree, and pointing the build at it.
> 
> On the other hand, I have worked in a large development organization which
> was quite conservative about upgrading to the newest version of any
library or
> tool.  Upgrades take time, and if they temporarily break someone's setup
(or
> even a whole team's!) they can consume a huge number of man-hours.  Even
> worse if the new version brings a new and unknown bug with it.  So, those
> managing our projects tended to be cautious about upgrades.  We'd use old
> versions of libraries, old compilers, everything.  We were usually a
couple of
> years behind in our boost version, and our compilers were quite a bit
older than
> that.
>  As long as it worked, we wouldn't touch it until someone could make a
good
> case for the benefits of upgrading outweighing the risks.  In the case of
this bug,
> such questions as "Do you *have* to use the boost::math special functions?
> How much effort would it take to just roll our own?  Does the new version
bring
> in any known regressions?
> How long has the new version been 'in the wild,' and how well is the
library
> vetted in general?"
> 
> Of course I can only speak from my own experience - I have no idea if such
> practices are common, but hopefully this is the sort of information you
were
> looking for :)

Thanks for this reply.

I understand some of the pressures to keep with 'The-Devil-You-Know', but I
have a suspicion that these hyper-cautious users don't know how much testing
multi-platform Boost code gets, so that things that will bite them are
exposed by these tests, no to mention the squeals from early-adopters users
who find themselves bitten.

But we see a steady stream of users on this list who have been bitten by
bugs that have already been squashed in later version.

So I feel that even cautious users should be aware that (provided they don't
go for the released-today hot off Sourceforge and wait until the dust has
settled, perhaps for a few weeks) the balance of risk is much more in favour
of using recent releases than they think.

Paul

---
Paul A. Bristow,
Prizet Farmhouse, Kendal LA8 8AB  UK
[phone]  07714330204
pbristow_at_[hidden]