| 1 |
Am 22.01.2014 18:42, schrieb Florian Philipp: |
| 2 |
> Hi list! |
| 3 |
> |
| 4 |
> I'm wondering if anyone tried setting -fnothrow-opt as a CXX_FLAG in |
| 5 |
> make.conf? |
| 6 |
> |
| 7 |
> It makes C++ throw() statements behave like C++11 nothrow. This could |
| 8 |
> have measurable performance benefits and will reduce code size. The only |
| 9 |
> downside is that, when a function violates its throw() guarantee, |
| 10 |
> terminate() will be called instead of unexpected(). |
| 11 |
> |
| 12 |
> However, neither function is allowed to do anything but terminate the |
| 13 |
> program. So all you possibly lose is a meaningful error message just |
| 14 |
> before the program crashes. |
| 15 |
> |
| 16 |
> So, what do you think? Safe to enable by default? |
| 17 |
> |
| 18 |
> Thanks in advance! |
| 19 |
> Florian Philipp |
| 20 |
> |
| 21 |
|
| 22 |
As a follow-up: |
| 23 |
I grepped all distfiles to identify candidates for testing this flag. |
| 24 |
The only package that I have installed (KDE desktop) that makes regular |
| 25 |
use of throw() statements is LibreOffice. However, since LO already |
| 26 |
enables the similar -fno-enforce-eh-specs flag, there seems to be no |
| 27 |
significant advantage in enabling -fnothrow-opt. |
| 28 |
|
| 29 |
Other packages only use these statements in the rare cases that it is |
| 30 |
required for STL compatibility (subclassing extensions, specializing |
| 31 |
templates). |
| 32 |
|
| 33 |
For now I add -fno-enforce-eh-specs and -fnothrow-opt to my CXXFLAGS but |
| 34 |
I doubt it will help much. |
| 35 |
|
| 36 |
Regards, |
| 37 |
Florian Philipp |
Archives