Regression (of sorts) in 20021111 snapshot

[Paul Koning]

>>>>> "Jack" == Jack Lloyd <lloyd@acm.jhu.edu> writes:

 Jack> The behaviour of g++ when facing long long constants that don't
 Jack> have an LL suffix seems to be changing from release to
 Jack> release. Specifically, some versions print a warning, some an
 Jack> error, some ignore it.

 Jack> egcs 1.1.2 - fine with it gcc 2.95.x - Error unless
 Jack> -fpermissive/-pedantic is set, warning with gcc 3.[01].* - fine
 Jack> gcc 3.2 - haven't checked 20021111 - same as gcc 2.95.x

 Jack> This is annoying for me, because I know those constants are
 Jack> bigger than a long, so a warning is not needed (particularly
 Jack> when a single file has 1000s of such constants), and errors are
 Jack> even more annoying. I had thought this behaviour was limited to
 Jack> 2.95.x up until now, though (and have specific workarounds for
 Jack> those compilers in my build stuff).

 Jack> I would file a bug in GNATS, but the change seems entirely
 Jack> intentional. I am just wondering the reason for the recent
 Jack> change, and if there is a way to work around the problem
 Jack> cleanly (ie my makefiles will continue to work with egcs 1.1.2,
 Jack> etc), and preferably without a warning at compile time.

A recent (3.3-something) gcc spits out a warning.  I'm not sure if
there is a way to turn it off.

As far as I'm concerned, this is an important feature.  I would
strenuously object to it being removed.  Whether it should be on by
default, or whether a switch should exist to turn it off if it's on by
default, is another matter.

I agree it shouldn't be an error (unless -Werror is set of course).

    paul