> I think GNU gzip just silently produces archives it can't test.  Which
> is bad, but not as bad as causing scripts to abort and report spurious errors.

This thread, or one like it, went around a while ago, didn't it?  Wasn't the
upshot that the format has a 32 bit field that contains size%4G, and can
always be tested by comparing the recovered size%4G to the stored value?
Strictly that makes it a probabilistic test - it could be fooled if some
corruption changed the recovered size by an exact multiple of 4G (!) - I
think that's a probability I'm willing to accept.

In which case the message, if any, when the size wraps 4G should be
considered an info message, not a warning, simply noting that the size
exceeds 4G and will be stored and verified mod 4G, and not changing exit
status. As an info message, it seems harmless to me and probably worth
keeping.

-Chap