Does anyone know why exactly machine code is disabled when OpenBSD is detected? On 7.7 at least, make verify gives the same results on OpenBSD with machine code enabled or disabled. All the CVS commit says is "Seems to violate some checks on the compiler".
Adam
Le 7 févr. 06 à 23:53, Adam Montague a écrit :
Does anyone know why exactly machine code is disabled when OpenBSD is detected? On 7.7 at least, make verify gives the same results on OpenBSD with machine code enabled or disabled. All the CVS commit says is "Seems to violate some checks on the compiler".
Machine code on sparc64 was not working... maybe on x86 it should work...
By the way in general I allways disable that because of security problems that it can give us.
But you are free to enable it if you like :)
/Xavier
On Wed, Feb 08, 2006 at 11:14:28AM +0100, Xavier Beaudouin wrote:
By the way in general I allways disable that because of security problems that it can give us.
what security problems does machinecode give?
this is the first time i hear that claim.
to my understanding the ability to produce machine code is an essential part of what makes pike as fast as it is. ideally it should be supported on all platforms.
greetings, martin.
On Thu, 9 Feb 2006 19:33:37 +0100 Martin B_hr mbaehr@email.archlab.tuwien.ac.at wrote:
On Wed, Feb 08, 2006 at 11:14:28AM +0100, Xavier Beaudouin wrote:
By the way in general I allways disable that because of security problems that it can give us.
what security problems does machinecode give?
this is the first time i hear that claim.
to my understanding the ability to produce machine code is an essential part of what makes pike as fast as it is. ideally it should be supported on all platforms.
greetings, martin.
Speaking of which, I don't see any performance difference with machine code enabled or disabled. I tried testing some of my pike code, as well as pike -x benchmark. What exactly is enabling machine code supposed to affect, and how can I verify if its actually enabled or not?
Adam
Normally Pike compiles Pike source code into a binary format which is then interpreted by the Pike VM. If possible Pike instead generates machine code which is intepreted directly by the CPU.
On Mon, 13 Feb 2006 15:30:16 +0000 (UTC) "Martin Nilsson (Opera Mini - AFK!) @ Pike (-) developers forum" 10353@lyskom.lysator.liu.se wrote:
Normally Pike compiles Pike source code into a binary format which is then interpreted by the Pike VM. If possible Pike instead generates machine code which is intepreted directly by the CPU.
After taking some more time to run pike -x benchmark a few times and average the results, it does look like using machine code is faster (except the compiling test, which maybe makes sense).
Given that it is faster, and that it works fine on the most common arch at least (i386), shouldn't it be enabled by default, and only disabled on arch+OS combos that are known to have actual problems?
Adam
pike-devel@lists.lysator.liu.se
-
Adam Montague -
Martin B�hr -
Martin Nilsson (Opera Mini - AFK!) ＠ Pike (-) developers forum -
Xavier Beaudouin