Re: ReiserFS reading performance patch

["Vladimir 'φ-coder/phcoder' Serbinenko" <phcoder@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2182 bytes --]

On 28.04.2012 12:06, Bean wrote:
> 2012/4/28 Vladimir 'φ-coder/phcoder' Serbinenko <phcoder@gmail.com>:
>> On 27.04.2012 20:36, Vladimir 'φ-coder/phcoder' Serbinenko wrote:
>>> On 27.04.2012 16:05, Albert Zeyer wrote:
>>>> * Write a simplified compile.py script.
>>> We don't have any compile.py.
>>>> * Remove nested functions. They don't work on Mac because of stack
>>>> execution protection.
>>> It can be disabled. It's Apple fault for not allowing to disable it
>>> selectively. Replacing nested functions is possible if done properly but
>>> the "let's make it all static" is wrong and will create problems with
>>> recursion, i.a. with nested partitions.
>> And actually nested functions don't matter. What does are trampolines
>> and they are only emitted it nested functions access local variables of
>> parent and not inlined.
> Hi,
>
> I think mac is not alone, netbsd uses non-executable stack and heap as
> well. Anyway, lower system security just to run an application does
> seem a little undesirable.
NetBSD allows to enable and disable heap execution at runtime using
__enable_executable_stack which is called by GCC automatically when
emitting trampolines.
> I once wrote a small framework to get rid of the nasty trampolines, for example:
>
> #define DEFINE_CLOSURE struct __closure_struct {
> #define DEFINE_CLOSURE_END } __closure_data;
> #define CLOSURE_TYPE void *__closure
void has a disadvantage of not checking the alignment. It's better to
use opaque structs.
> #define CLOSURE __closure
> #define CLOSURE_DATA &__closure_data
> #define USE_CLOSURE struct __closure_struct *__closure_p = __closure;
> #define UNUSED_CLOSURE (void) __closure;
We systematically use __attribute__ ((unused))
This is a good way of doing things when there is no other option.
However, in many cases it's better to use macros like FOR_*. They also
result in smaller code. The major point where it can't be done is with
->dir but this interface should be changed anyway because it forces some
code to scan through directory just to e.g. find out mtime of a single
file.



-- 
Regards
Vladimir 'φ-coder/phcoder' Serbinenko



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 294 bytes --]