Re: Compiler Revisions Should go Out As Well as Going Up.

[Simon Wright <simon@pushface.org>]

Pascal Obry <pascal@obry.net> writes:

> Le 11/12/2013 13:11, Austin Obyrne a écrit :

>> Very often there is nothing to be gained except a whole load of
>> trouble in changing to newer compiler versions of the same language.
>
> That's plain wrong. And almost an insult to the people who have
> development the compiler since all those years!

For most people there's good reason to at least try the latest
version. Of course you need a good set of regression tests to be able to
do this!

But in some cases upgrading is a real problem. I'm thinking of systems
that have gone through extensive and expensive validation exercises,
including for example avionics flight tests or missile firings; even if
you trust the compiler team, you'll have an uphill task to persuade the
person who signs off (Design Authority, Chief Engineer, ...).

GNAT 5.02a1 had an off-the-cuff remark a long way down a release note
that for a certain architecture (the one we were using!) there was a
one-in-millions chance that a function result would be corrupted,
because the epilogue code wasn't interrupt-safe; you could use a switch
(-fno-something-or-other that I don't remember) to prevent
this. Needless to say we abandoned the upgrade.