http://d.puremagic.com/issues/show_bug.cgi?id=573

           Summary: Segfault from version(release) statement
           Product: D
           Version: 0.174
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Keywords: wrong-code
          Severity: normal
          Priority: P2
         Component: DMD
        AssignedTo: bugzilla@digitalmars.com
        ReportedBy: daekharel@gmail.com


void main() { version (release) {} else { assert(0); } }

-------
The above throws an AssertError, as it should, when compiled & linked normally and run. However, when compiled with the "-release" flag, it segfaults when run instead of doing nothing, as it should. Strangely, if it is compiled with "-v=release", it runs without error.


-- 

<d-bugmail@puremagic.com> wrote in message news:bug-573-3@http.d.puremagic.com/issues/...
> http://d.puremagic.com/issues/show_bug.cgi?id=573
>
>           Summary: Segfault from version(release) statement
>           Product: D
>           Version: 0.174
>          Platform: PC
>        OS/Version: Linux
>            Status: NEW
>          Keywords: wrong-code
>          Severity: normal
>          Priority: P2
>         Component: DMD
>        AssignedTo: bugzilla@digitalmars.com
>        ReportedBy: daekharel@gmail.com
>
>
> void main() { version (release) {} else { assert(0); } }
>
> -------
> The above throws an AssertError, as it should, when compiled & linked
> normally
> and run. However, when compiled with the "-release" flag, it segfaults
> when run
> instead of doing nothing, as it should.

You sure about that?  assert(0) is treated as a special case, and when compiled in release mode, they are left in and execute an illegal instruction.

> Strangely, if it is compiled with
> "-v=release", it runs without error.

AFAIK there is no "release" version defined when you use the -release flag. Therefore, the code is doing exactly as it should -- it compiles in the "else" clause in release mode, causing an exception when the program is run. But when you manually define the "release" version, it skips the "else" and leaves the assert out.

http://d.puremagic.com/issues/show_bug.cgi?id=573


daekharel@gmail.com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID




------- Comment #1 from daekharel@gmail.com  2006-11-20 17:11 -------
Whoops. Looks like I misinterpreted what -release does: it doesn't switch the "release" version flag, and it removes assertions (or at least simple ones like assert(0)), replacing them by stuff that causes segfaults.

I'll assume there's a good reason for the segfault behavior (probably has
something to do with assert(0) signifying unreachable code), but in any case,
this bug is invalid.


-- 

http://d.puremagic.com/issues/show_bug.cgi?id=573





------- Comment #2 from jarrett.billingsley@gmail.com  2006-11-20 18:43 -------
Ah, sorry, I guess I should have posted my reply here instead of on the newsgroups.

assert(0) is treated as a special case, and when compiled in release mode, they are left in and execute an illegal instruction.


-- 

d-bugmail@puremagic.com wrote:
> assert(0) is treated as a special case, and when compiled in release mode, they are left in and execute an illegal instruction.  

Not precisely an illegal instruction, but a HLT (or equivalent) instruction, which works like a break point when running it under a debugger.

Walter Bright wrote:

>> assert(0) is treated as a special case, and when compiled in release mode, they are left in and execute an illegal instruction.  
> 
> Not precisely an illegal instruction, but a HLT (or equivalent) instruction, which works like a break point when running it under a debugger.

On PowerPC it does a "trap" instruction which also has that effect:

Program received signal SIGTRAP, Trace/breakpoint trap.
_Dmain () at halt.d:3
3         assert(0);
(gdb)

--anders