Re: type based aliasing again -- a suggestion from a user

["George T. Talbot" <george at moberg dot com>]

As a user, not a maintainer of the compiler, who has been following this
discussion since the maintainers and Linus were discussing about this
issue, I have the following suggestion:

Even after following this issue for so long, I still have only a hazy
idea of the implications of -fstrict-aliasing, and I do not have a clear
idea, in real-world terms, of what kind of code breaks.  The examples
given so far have not been realistic ones that I'd see in actual working
software.

Everyone's been saying "the user wants this, the user wants that" but I
really do think that this issue hasn't been expressed clearly enough for
[most, some, a couple, just me] users to get a grasp on it.

I would suggest that people working on this issue pause to write the GNU
info page which describes the new behavior to the users and warns them
of what constructs are now invalid.  Not everyone has a copy of the ISO
standard (I don't)--maybe this would be a good way to get user feedback.

In other words, write the page in the user manual first.  Expressing the
issues in user-manual form, rather than compiler implementation and
mailing list discussions will force a level of rigor and clear
explanation that would go a long way to helping users of the compiler
understand this issue and give feedback.

Who knows?  After revising it a couple of times, it might shed some
light on how best to warn users, etc.?
--
George T. Talbot
<george@moberg.com>