'&' operator doesn't work on leftvalue about orangec

[LADSoft]

Hi lionfaith!

well the argument of & is supposed to work on lvalues - and usually an
lvalue is taken as something that can be assigned to (e.g. it is the
left side of an assignment statement). Other compilers such as the
microsoft compiler, the borland compiler, and the watcom compiler all
seem to stick to that definition; gcc is the only one I can find that
has this other behavior.

However, that said, the GCC implementation doesn't seem to be out of
bounds, because the latest spec actually defines an lvalue as something
that can refer to an object, and I can see a case being made that the
specific examples you have quoted do indeed refer to objects. So I
will look into implementing it :)

David

lionfaith wrote:

Hello!
I'm trying to compile on windows some SDL based free software project
which AFAIK compiles on Linux (supposedly with gcc) and I envetually
step upon errors on this kind of expressions:
&--theEnemy->thinktime
&(theEnemy->dx -= 0.5)
&(++enemy[WC_BOSS].shield)
&(--engine.eventTimer)
&(star[i].x += (engine.ssx * star[i].speed))
And many others all alike. Although it is used in some kind
optimisation hack, it seems to be valid code. The address-of operator
is not limited to variable, but can be applied on any leftvalue. Or so
it seems.

As '&' operator is hard to search for on the internet, the only clear
reference to this behavior I could find is this:
https://msdn.microsoft.com/en-us/library/64sa8b1e.aspx

—
Reply to this email directly or view it on GitHub
#10.

[LADSoft]

hi lionfaith,

I released a new version which will allow the compiler to take the address of objects that are returned from an assignment, a preincrement, or a postincrement. Please let me know whether this is sufficient or if there are other types of expressions gcc lets get by as well.

David

[lionfaith]

Hi! Great thanks! So fast!
I've downloaded 5.25 and the faulty files now compiles without complains.