John Levon <[EMAIL PROTECTED]> writes:
| Doing lyx -x lyx-quit big.lyx, the attached patch
| gives the following results (warm cache, error around 4%) :
|
| user (s) | sys (s) | elapsed (s) |
| ----------------------------------------------------------------------
| 28.33 (0.00%) | 0.30 (0.00%) | 29.61 (0.00%) | newinsetsold
| 24.29 (-14.27%) | 0.32 (6.61%) | 25.92 (-12.48%) | newinsetsnew
|
| and the binary is only 14k bigger (with -g -O). This seems worth it to me.
I don't want this patch yet.
If you have lines like:
do_something(with_this.size());
and widh_this.size() is inlined and do_something segfaults, the
segfault can look like it happened inside with_this.size().
Also when having inlines in the header file, alway put the inlined
method after the class definithin with an inline, in the class inline
should not be used:
class Foo {
publi:
int size();
};
inline
Foo::size() {
return ...;
}
As you can see with this it is _very_ easy to move the inlined method
from the header file to the C file (and back).
I have been thinking of a scheme where we can choose at compile time
if we want to have methods like this inlined or not.
foo.h:
class Foo {
public:
int size();
};
#ifdef INLINE_SIMPLE_METHODS
#define INLINE inline
#include "foo_inlines.h"
#endif
foo_inlines.h:
INLINE
Foo::size() {
return ...;
}
foo.C:
#ifndef INLINE_SIMPLE_METHODS
#define INLINE
#include "foo_inlines.h"
#endif
Or another scheme that gives the same effect.
Lgb
