forums.silverfrost.com

[sparge]

I don't suppose this could be called a compiler bug; more a blind spot. The following code is a heavily-stripped down version of a routine I was hacking around yesterday.

program unspotted_no_op
integer trys, dsa, numdsa
logical test (4)
numdsa = 10
test = .true.
outer_loop: do dsa = 1, numdsa
trys = 0
inner_loop: do
trys = trys + 1
if (test (1)) then
if (test (2)) cycle outer_loop
if (test (3)) cycle inner_loop
exit outer_loop
end if
end do inner_loop
if (test (4)) exit outer_loop
end do outer_loop
stop
end program unspotted_no_op

When compiled with /RELEASE, FTN95 correctly issues:

warning 133 - This statement will never be executed

in respect of the line:

if (test (4)) exit outer_loop

When compiled with /DEBUG or /CHECKMATE, however, FTN95 fails to spot that anything is or amiss (or fails to mention the fact).

I'm glad I investigated this because my first thought was that the RELEASE build was issuing a false positive. In fact, it was quite correct in identifying something that was definitely not the design intent, and which I'd never have spotted in the original code.

[sparge]

Looks like I need to ask an explicit question!

Am I right in thinking as follows:
- there is no requirement for a compiler to identify dead code
- nonetheless, if a compiler elects to try and do so, build-dependent ability to do so is not the design intent, and could usefully be investigated

Having pondered this further, I've realised that dead code per se is not the issue; the real issue is the potentially infinite loop that dead code signals. Right?

Andy

[PaulLaidler]

Andy

I have seen your post but it may be a while before I can investigate its implications.

Paul

[PaulLaidler]

Andy

We have now fixed this problem for the next release.

Yes, I think you are right in what you say.

I was quite impressed that FTN95 was able to give a warning even though it only only showed up in release mode. Debug mode required a slightly different route through the compiler because do loops have to be handled in a different way.

[sparge]

I was very impressed! I hope I didn't give any impression to the contrary - not my intention at all. The more I think about how a compiler might possibly figure out that a given line can't be got to in any possible runtime scenario, the more impressed I am. I was just surprised that the warning came in RELEASE mode rather than CHECKMATE, which is where I would have expected it.