[Cfp-interest] Compound statement exit complications

[Ian McIntosh]

I agree that the new pragmas we're discussing should apply to compound
statements (C lexical blocks).  There are sometimes complications.


Straightforward:
   {
       #pragma  STDC  FENV_ROUND  FE_TOWARD_ZERO
       . . .
   }  /* pragma's effect ends here */


A little trickier:
   {
       #pragma  STDC  FENV_ROUND  FE_TOWARD_ZERO
       . . .
       if (condition) goto label1;  /* pragma's effect ends if the goto is
   executed */
       . . .
   }  /* pragma's effect ends here */
   . . .
   label1:

In that case the compiler has to observe that Label1 is outside the
compound statement affected by the pragma, and turn off the pragma's effect
both when exiting through the bottom of the compound statement and also
when exiting in the middle via a goto or equivalent.


A lot trickier, using gcc / IBM xlc label variables:
   void* label_variable;
   {
       #pragma  STDC  FENV_ROUND  FE_TOWARD_ZERO
       . . .
       if (condition) label_variable = label2;
       else label_variable = label3;
       . . .
       if (condition) goto label_variable;  /* pragma's effect MIGHT OR
   MIGHT NOT end if the goto is executed, depending on where it goes to */
       . . .
       label2:  /* if label_variable points here, goto leaves the pragma in
   effect */
       . . .
   }  /* pragma's effect ends here */
   . . .
   label3:  /* if label_variable points here, goto ends the pragma's effect
   */

- Ian McIntosh          IBM Canada Lab         Compiler Back End Support
and Development
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20140513/e5c7c928/attachment.html