Re: lang/clisp build fails on i386 -current

[Sam Steingold]

> * Sam Steingold <[EMAIL PROTECTED]> [2006-06-30 11:21:43 -0400]:
>
>> * Hannah Schroeter <[EMAIL PROTECTED]> [2005-08-13 14:13:44 +0200]:
>>
>> I've received the suggestion to build clisp with these Makefile's
>> CFLAGS:
>> -DNO_SINGLEMAP -DNO_TRIVIALMAP -DNO_MULTIMAP_FILE -DNO_MULTIMAP_SHM
>>
>> However I don't know how to pass them in using the port, and I didn't
>> have the time/energy to try it out.
>
> ./configure ... --disable-mmap

however, this is not necessary on OpenBSD, see


see also clisp/unix/PLATFORMS in the recently released clisp 2.39:

On PC 386 running OpenBSD 3.8:

If you get mmap errors on the first invocation of lisp.run, this is
likely due to the new randomizing malloc.
Pass "--with-gmalloc" to ./configure to work around this problem.


>> However I've been advised that that'll decrease the performance of
>> clisp, among others it'll make the GC non-generational.
> I don't think this is true.

It certainly is true, I don't know what I was thinking.

-- 
Sam Steingold (http://www.podval.org/~sds) on Fedora Core release 5 (Bordeaux)
http://memri.org http://mideasttruth.com http://honestreporting.com
http://openvotingconsortium.org http://truepeace.org http://camera.org
There are many reasons not to use Linux - but no good ones.

Re: lang/clisp build fails on i386 -current

[Sam Steingold]

> * Hannah Schroeter <[EMAIL PROTECTED]> [2005-08-13 14:13:44 +0200]:
>
> I've received the suggestion to build clisp with these Makefile's
> CFLAGS:
> -DNO_SINGLEMAP -DNO_TRIVIALMAP -DNO_MULTIMAP_FILE -DNO_MULTIMAP_SHM
>
> However I don't know how to pass them in using the port, and I didn't
> have the time/energy to try it out.

./configure ... --disable-mmap

> However I've been advised that that'll decrease the performance of
> clisp, among others it'll make the GC non-generational.

I don't think this is true.


-- 
Sam Steingold (http://www.podval.org/~sds) on Fedora Core release 5 (Bordeaux)
http://mideasttruth.com http://iris.org.il http://palestinefacts.org
http://openvotingconsortium.org http://pmw.org.il http://honestreporting.com
.sigs are like your face - rarely seen by you and uglier than you think

Re: lang/clisp build fails on i386 -current

[Marc Espie]

On Sat, Aug 13, 2005 at 02:13:44PM +0200, Hannah Schroeter wrote:
> And... While clisp seems to be the only port/package affected by all
> that randomization stuff, there *are* other things affected, such as
> cmucl/sbcl, just that they're not in ports anyway. But there were
> discussions on how to fix/workaround problems for similar randomization
> (in that case shared library load address randomization) on some
> "flavours" of Linux.

That's not true, gprolog is affected as well.
And precompiled headers in recent gcc do not work because of randomization
as well.

Re: lang/clisp build fails on i386 -current

[Hannah Schroeter]

Hello!

On Sat, Aug 13, 2005 at 08:25:36AM +0200, Andreas Vögele wrote:
>>Known problem, everybody does.

>Do you know if the OpenBSD kernel will be fixed before 3.8 is released?

>Thierry Deval wrote on 4th May:

>"While a proper solution has to be devised, you can use the following
>diff that diminishes the impact of randomization on malloc... And makes
>clisp work again. [...]"

>I still use the patch that Thierry provided.

I've received the suggestion to build clisp with these Makefile's
CFLAGS:
-DNO_SINGLEMAP -DNO_TRIVIALMAP -DNO_MULTIMAP_FILE -DNO_MULTIMAP_SHM

However I don't know how to pass them in using the port, and I didn't
have the time/energy to try it out.

However I've been advised that that'll decrease the performance of
clisp, among others it'll make the GC non-generational.

Btw, doesn't that high degree of mmap randomization cause the virtual
memory space, especially on 32-bit machines, to be quite fragmented, so
big allocations might fail?

And... While clisp seems to be the only port/package affected by all
that randomization stuff, there *are* other things affected, such as
cmucl/sbcl, just that they're not in ports anyway. But there were
discussions on how to fix/workaround problems for similar randomization
(in that case shared library load address randomization) on some
"flavours" of Linux.

Kind regards,

Hannah.

Re: lang/clisp build fails on i386 -current

[Andreas Vögele]

Christian Weisgerber writes:


J. Lievisse Adriaanse <[EMAIL PROTECTED]> wrote:


When trying to build lang/clisp on both of my laptops (running
-current), the build fails in "make build" because it can't map the
memory correctly, it says something  like: [...]



Does anybody else has this problems too?


Known problem, everybody does.


Do you know if the OpenBSD kernel will be fixed before 3.8 is released?

Thierry Deval wrote on 4th May:

"While a proper solution has to be devised, you can use the following
diff that diminishes the impact of randomization on malloc... And makes
clisp work again. [...]"

I still use the patch that Thierry provided.

Re: lang/clisp build fails on i386 -current

[J. Lievisse Adriaanse]

On Wed, 10 Aug 2005 15:11:56 + (UTC)
[EMAIL PROTECTED] (Christian Weisgerber) wrote:

> J. Lievisse Adriaanse <[EMAIL PROTECTED]> wrote:
> 
> > When trying to build lang/clisp on both of my laptops (running
> > -current), the build fails in "make build" because it can't map the
> > memory correctly, it says something  like: [...]
> 
> > Does anybody else has this problems too?
> 
> Known problem, everybody does.
> 
> -- 
> Christian "naddy" Weisgerber  [EMAIL PROTECTED]
> 
Yes, I found out after reading through the archives of the mailinglist.

Jasper


-- 
"Security is decided by quality" -- Theo de Raadt

Re: lang/clisp build fails on i386 -current

[Christian Weisgerber]

J. Lievisse Adriaanse <[EMAIL PROTECTED]> wrote:

> When trying to build lang/clisp on both of my laptops (running
> -current), the build fails in "make build" because it can't map the
> memory correctly, it says something  like: [...]

> Does anybody else has this problems too?

Known problem, everybody does.

-- 
Christian "naddy" Weisgerber  [EMAIL PROTECTED]

Re: lang/clisp build fails on i386 -current

[Alec Berryman]

J. Lievisse Adriaanse on 2005-08-10 14:34:55 +0200:

> When trying to build lang/clisp on both of my laptops (running -current), the
> build fails in "make build" because it can't map the memory correctly, it says
> something like:
> 
> ./lisp.run -B . -N -Efile -UTF-8 -Eterminal UTF-8 -norc -m 750KW -x 
> "(and(load \ "init.lisp"\) (sys::%saveinitmem) (ext::exit))"
> Cannot map memory to address 0xe7bb1000 .
>  errno = EINVAL: Invalid argument.
> Trying to make room through a GC...
> Cannot map memory to address 0xe7bb1000 .
>  errno = EINVAL: Invalid argument.
> Trying to make room through a GC...
> Segmentation fault (core dumped)
> *** Error code 139
> 
> Stop in /usr/ports/lang/clisp/w-clisp-2.33.p0/build-i386 (line 2319 of 
> Makefile)
> *** Error code 1
> 
> 
> Does anybody else has this problems too?

I can reproduce this problem on the August 8th snapshot.

lang/clisp build fails on i386 -current

[J. Lievisse Adriaanse]

Hello,

When trying to build lang/clisp on both of my laptops (running -current), the 
build fails in "make build" because it can't map the memory correctly, it says 
something  like:

./lisp.run -B . -N -Efile -UTF-8 -Eterminal UTF-8 -norc -m 750KW -x "(and(load 
\ "init.lisp"\) (sys::%saveinitmem) (ext::exit))"
Cannot map memory to address 0xe7bb1000 .
 errno = EINVAL: Invalid argument.
Trying to make room through a GC...
Cannot map memory to address 0xe7bb1000 .
 errno = EINVAL: Invalid argument.
Trying to make room through a GC...
Segmentation fault (core dumped)
*** Error code 139

Stop in /usr/ports/lang/clisp/w-clisp-2.33.p0/build-i386 (line 2319 of Makefile)
*** Error code 1


Does anybody else has this problems too?
The RELEASE package installs fine, though.

Cheers
Jasper
-- 
"Security is decided by quality" -- Theo de Raadt