https://gcc.gnu.org/bugzilla/show_bug.cgi?id=38658
Martin Sebor <msebor at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|RESOLVED |UNCONFIRMED
Resolution|INVALID |---
Summary|inefficient code on trivial |trivial try/catch statement
|try/catch statement |not eliminated
--- Comment #3 from Martin Sebor <msebor at gcc dot gnu.org> ---
There are situations when throwing an exception is required to result in a call
to std::terminate() but those require the exception to escape the function
where it's thrown from. Outside of those, it seems to me that as long as the
observable effect of the optimized code is indistinguishable from the original
(such as in the test case) removing try/catch statements should be a valid
optimization.
Allocating memory is not an observable effect of throwing an exception so it
doesn't count. std::uncaught_exception() has to be called along the way so
that is also out to the extent that GCC can see the code between the throw
expression and the catch block.
It's clear that not all cases can be optimized as I suggest but I believe at
least some of them can (the one in the test case in comment #0). I'm going to
reopen this request. If it turns out that there's something I'm missing we can
close it once it's made clear in a test case that the optimization would either
be invalid even in that subset of cases, or if it looks like it would not be
profitable enough to bother.
