Home Original page

[Python-Dev] RE: More informative error messages

Scott David Daniels Scott.Daniels at Acm.Org
Wed Oct 8 11:26:43 EDT 2003 
[Tim Peters]

>....  OTOH, almost
>every call to a C API function has to test+branch for an error-return value,
>and I've often wondered whether a setjmp/longjmp-based hack might allow for
>cleaner and more optimizable code (hand-rolled "real exception handling").
>
setjmp/longjmp are nightmares for compiler writers.   The writers tend 
to turn
off optimizations around them and/or get corner cases wrong.  If you 
read the
C standard, precious little is guaranteed around setjmp/longjmp.  The C 
code
using disciplined setjmp/longjmp, will read well, probably be be quite
optimizable, but ....

At least some of the C compilers will mis-optimize such code and others 
will be
painfully slow due to the interaction of two compiler coding strategies: 
first,
emit straightforward sloppy code easily cleaned up in the optimization 
passes,
and second, turn off optimization in the presence of setjmp/longjmp.

Maybe the general compiler world has changed, but I had nightmares 
supporting
a language which generated C including setjmp/longjmp calls, and ran it on
top of three C compilers.  Each compiler had nasty cases to avoid, and the
resulting least common denominator was painfully inept.

-Scott David Daniels
More information about the Python-Dev mailing list