gmane.lisp.scheme.gauche

Re: Gauche fails to build on FreeBSD/i386

Shiro Kawai — 2012-05-25T23:29:21

Bravo.  Thanks, Vitaly.

Changing the use of NSIG in gauche/vm.h is a bit more involved than
just replacing the constant, since doing so would break ABI.
Let me try to put a reasonable fix in git master.   I'm not sure
I can make a new release soon, for I won't have time to check
on various platforms for at least 2-3 weeks, but I can make a branch
from where you can pull.

--shiro

From: Vitaly Magerya <vmagerya< at >gmail.com>
Subject: Re: [Gauche-devel] Gauche fails to build on FreeBSD/i386
Date: Fri, 25 May 2012 19:48:02 +0300




------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Gauche fails to build on FreeBSD/i386

Vitaly Magerya — 2012-05-25T16:48:02

And the winner is sigHandlers.

It appears that on FreeBSD NSIG is 32, but SIGRTMIN is 65, so
GAUCHE_PTHREAD_SIGNAL is 70 and when Scm_SetMasterSigmask tries
to do this:

    sigHandlers.handlers[GAUCHE_PTHREAD_SIGNAL] = DEFAULT_SIGHANDLER;

... next_parameter_index is overwritten.

It apears that you need to use _SIG_MAXSIG+1 instead of NSIG on
FreeBSD.

In summary, if I update the bundled gc to the latest git sources,
fix NSIG usage in src/signal.c and src/gauche/vm.h (are both
needed?), and comment out killgnutls.sh, Gauche builds and passes
all tests on both amd64 and i386.

So, how do we proceed? I want to update FreeBSD port of Gauche
(which is currently at 0.9.2 and does not work at all), but I
need a publically accessible tarball for that; will you apply
those fixes and make a new release, or should I distribute a
patched tarball separately for the time being?

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will co

Re: Gauche fails to build on FreeBSD/i386

Vitaly Magerya — 2012-05-25T11:36:28

Here's what I have:

    00257310 b keywords
    00257320 b sigHandlers
    00257434 B parameter_mutex
    00257438 b next_parameter_index
    00257440 b nullModule
    00257480 b schemeModule
    002574c0 b gaucheModule

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Gauche fails to build on FreeBSD/i386

Shiro Kawai — 2012-05-24T23:42:34

From: Vitaly Magerya <vmagerya< at >gmail.com>
Subject: Re: [Gauche-devel] Gauche fails to build on FreeBSD/i386
Date: Fri, 25 May 2012 02:33:40 +0300


Hmm... that strongly suggests that something is overrunning
the memory region, which is worrysome.

Can you "nm libgauche-0.9.so | sort" and see what's allocated
around next_parameter_index in the bss area?

(Side note: I noticed I forgot to say 'static' in parameter_mutex;
it isn't supposed to be exported.)

On linux/x86_64, it's like this:

00000000004e3e60 b keywords
00000000004e3ea0 b sigHandlers
00000000004e4360 B parameter_mutex
00000000004e4388 b next_parameter_index
00000000004e43a0 b gaucheModule
00000000004e4400 b nullModule
00000000004e4460 b schemeModule

--shiro

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile

Re: Gauche fails to build on FreeBSD/i386

Vitaly Magerya — 2012-05-24T23:33:40

I did some debugging, and the problem comes from Scm_InitParameterLoc,
where ensure_parameter_slot is called with that value as "index",
which is caused by "next_parameter_index" jumping from 1, 2, 3,
4, ... to something silly. Unfortunately hardware watchpoints
don't seem to work under VirtualBox (at least on my machine),
so I can't really determine exactly what overwrites it.

As a workaround I found that adding a dummy int variable just
before next_parameter_index is defined solves the problem... (If
you can call that a solution).

Anyway, I'll try to find real i386 hardware in the next few days
to try debugging there (unless you beat me to it).

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/

Re: Gauche fails to build on FreeBSD/i386

Vitaly Magerya — 2012-05-24T11:07:46

Ah, yes. I overlooked that.


I'm testing in VirtualBox myself; I don't use i386 daily.


I just tried that with both gc-7.2 and latest git sources -- and
I hope I didn't miss any of the changes you made; it's all in
configure scripts, right? Anyway, there appears to be no difference
for i386.


Yes. It actually kills itself by executing "kill -9 0".


It's the "atoms" section; here's the last two lines of the log:

    test atom, expects #t ==> ok
    test atomic counting, expects 1000 ==>

I tried removing other tests to see if it'll hang only on atoms,
and the more tests I remove, the less frequently it hangs. Just
the single atoms section isn't enough to hang it (even if I
increase the loop count and/or put the whole test into a loop).

BTW, I didn't notice before, but gc's own test "gctest" fails
too (infrequently; it segfaults).

Anyway; all the tests appear to work fine on amd64 when I replace
gc with it's latest git sources (gc-7.2 release hangs too though).

----------------------------------------

Re: Gauche fails to build on FreeBSD/i386

Shiro Kawai — 2012-05-23T20:33:11

I remember I did work to make Gauche work on FreeBSD.
I think ultimately updating bdwgc to 7.2alpha6 did the
trick.  I didn't have FreeBSD-i386 platform and couldn't
check on it.  


This is really weird message.  The 'Heap size' is too small,
and the number of bytes allocating is ridiculously large.
Definitely, something is really messed up.

I need the platform to dig into it.  However, a new release
of bdwgc seems close based on the conversation in gc ML, so
we can hope that it is taken care of.  (You can check it by
using git HEAD of bdwgc, but there are a few modifications
you need to apply---check the parts marked as "[SK]" in the
gc tree in Gauche.)


Actually, gnutls isn't used at all in our test, so killgnutls.sh is
no use.  I just leave it to make changes minimum.  Does the script
made the test fail?  The I'll just remove the call to the script
altogether.  (What matters is killopenssh.sh, and for that I intercept
the original script to avoid using "ps -ef".)

For the thread tests, do you know whic

Gauche fails to build on FreeBSD/i386

Vitaly Magerya — 2012-05-23T13:24:14

Hi, folks. I'm trying to build Gauche 0.9.3.2 on FreeBSD, and while
it works fine on amd64 (e.g. passes all tests [1]), it fails on i386
either like this (full log at [2]):

../../src/gosh -ftest ../../src/precomp -e -P -o gauche--collection ../../libsrc/gauche/collection.scm
Deallocating excessively large block.  Too large an allocation?
*** Signal 6

Or prints many error messages like this:

../../src/gosh -ftest ../../src/recomp [..stuff..]
GC Warning: Out of Memory! Heap size: 4 MiB. Returning NULL!
out of memory (2693136576).  aborting...

... and then:

../src/gosh -q -I../src -I../lib -lgauche-init ./extract -en -o gauche-refe.texi ./gauche-ref.texi
gosh: "error": Compile Error: Compile Error: Compile Error: Compile Error: cannot find "util/match" in ("../lib" "../src" "/usr/local/share/gauche/site/lib" "/usr/local/share/gauche/0.9/lib")
"../lib/gauche/condutil.scm":36:(define-module gauche.condutil (use  ...
"../lib/gauche/regexp.scm":106:(define-condition-type <regexp-inval ...
"../lib/gauche/parseo

Re: test sometimes fails,maybe increase number of retries?

2012-05-23T13:48:30


debian/testing
$uname -a:
Linux amalthea 3.2.0-2-amd64 #1 SMP Sat May 12 23:08:28 UTC 2012 x86_64 GNU/Linux

karme

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: test sometimes fails, maybe increase number of retries?

Shiro Kawai — 2012-05-23T11:51:48

Hm, I never saw it failed, but the change does no harm so I'll take it.
For the record, what platform you see the test fails on?

--shiro

From: karme< at >karme.de
Subject: [Gauche-devel] test sometimes fails, maybe increase number of retries?
Date: Wed, 23 May 2012 13:07:26 +0200

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

test sometimes fails,maybe increase number of retries?

2012-05-23T11:07:26

Index: gauche-0.9.3.2/test/control.scm
===================================================================
--- gauche-0.9.3.2.orig/test/control.scm2012-03-20 10:07:45.000000000 +0100
+++ gauche-0.9.3.2/test/control.scm2012-05-23 12:57:09.299612417 +0200
< at >< at > -160,7 +160,7 < at >< at >
            (job-status (dequeue/wait! (thread-pool-results pool))))
     (test* "shutdown check" 'finished
            (let retry ([n 0])
-             (cond [(= n 10) #f]
+             (cond [(= n 1000) #f]
                    [(symbol? gate) gate]
                    [else (sys-nanosleep #e1e8) (retry (+ n 1))])))
     )

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: $ and .$

shuji yamamoto — 2012-05-13T00:57:45

Thanks for interesting & exciting advice .
I used .$ many time , because of n-in,m-out scene .
And , It save cost of typing :-)

Now I understood , $ macro suits n-in,1-out scene.
I am Japanese but not good schemer ,  recommended deep discussion page 
is difficult for me (T_T)


(2012/05/13 08:52), Shiro Kawai wrote:


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: $ and .$

Shiro Kawai — 2012-05-12T23:52:08

From: shuji yamamoto <shuuji.yamamoto< at >gmail.com>
Subject: [Gauche-devel] $ and .$
Date: Sun, 13 May 2012 08:00:40 +0900

This is true oniy if g returns a single value.

This is an invalid program---it expands to (+ (quotient&remainder 22 4)),
and you are returning two values where only one value is expected
(expressions in the argument position expect exactly one value).
Gauche doesn't bother to check every violation of number of values,
though, so it "happens" to work in this case.  It is not guaranteed
to work in other context, or in future versions.

OTOH, "compose" is defined explicilty to work with multiple values.

(.$ f g) == (lambda args (call-with-values (lambda () (apply g args)) f))

However, it is an interesting question to ask whether '$' should
"splice" multiple values into an argument list.

Functionally it would be a backward-compatible extension.  However,
allowing implicit splicing of multiple values may open a can of
worms.  You can't tell the number of arguments passed to f or g
by look

$ and .$

shuji yamamoto — 2012-05-12T23:00:40

Hi.
I testing $ macro.

gosh> ($ + $ quotient&remainder 22 4)
5
gosh> ((.$ + quotient&remainder) 22 4)
7

I thought $ is "compose" , but result is different.
I mean:
($ f a b $ g c d $) == (.$ (pa$ f a b ) (pa$ g c d) )

Shuji Yamamoto

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Another follow-up release

Shiro Kawai — 2012-05-11T21:45:21

0.9.3 build fails if you specify sjis or eucjp to --enable-multibyte
configuration option.  The fixed version (0.9.3.2) is available now.

http://practical-scheme.net/gauche/index.html

If you compile utf-8 version, you probably don't need to update.

--shiro







------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Release 0.9.3

Shiro Kawai — 2012-05-11T06:44:35

Good to know it works.
No need to blame yourself.  I don't think I've documented it
('maintainer-clean' and 'distclean' targets are GNU's convention).

--shiro

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Release 0.9.3

shuji yamamoto — 2012-05-11T05:52:13

Thanks! it works.

$ gosh -V
Gauche scheme shell, version 0.9.3.1 [utf-8,pthreads], i686-pc-linux-gnu

2012/05/11 12:54, Shiro Kawai wrote:
No....
I have known "maintainer-clean" now.
I'm ashamed my ignorance ^^;



------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Release 0.9.3

Shiro Kawai — 2012-05-11T03:54:43

Weird.  Did you start from completely clean state (I mean,
right after git clone, or after make maintainer-clean)?

--shiro


From: shuji yamamoto <shuuji.yamamoto< at >gmail.com>
Subject: Re: [Gauche-devel] Release 0.9.3
Date: Fri, 11 May 2012 11:17:02 +0900


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Release 0.9.3

shuji yamamoto — 2012-05-11T02:17:02

I install 0.9.3

$ gosh -V
Gauche scheme shell, version 0.9.3 [utf-8,pthreads], i686-pc-linux-gnu

then build git HEAD:
Looks like same error.

../src/gosh -q -I../src -I../lib -lgauche-init ./extract -en -o 
gauche-refe.texi ./gauche-ref.texi
gosh: "error": Compile Error: wrong number of arguments for #<subr 
%vm-parameter-ref> (required 2, got 3)
"././extract":20:(use file.util)


(2012/05/11 10:47), Shiro Kawai wrote:


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Release 0.9.3

Shiro Kawai — 2012-05-11T01:47:06

Indeed I found a bug, but which only appears when you try
to use modules precompiled with 0.9.2 or before for
HEAD build.  Does it still happen if you build from 0.9.3
tarball, install it, then build HEAD?

--shiro


From: shuji yamamoto <shuuji.yamamoto< at >gmail.com>
Subject: Re: [Gauche-devel] Release 0.9.3
Date: Fri, 11 May 2012 09:35:33 +0900


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Re: Release 0.9.3

shuji yamamoto — 2012-05-11T00:35:33

Thanks for release 0.9.3. ^^

I tried to build git HEAD on ubuntu , and got some error:

../src/gosh -q -I../src -I../lib -lgauche-init ./extract -en -o 
gauche-refe.texi ./gauche-ref.texi
gosh: "error": Compile Error: wrong number of arguments for #<subr 
%vm-parameter-ref> (required 2, got 3)
"././extract":20:(use file.util)

It seem , some trouble on document.

Shuji Yamamoto< at >Ako City , Hyogo pref,  Japan.


(2012/05/11 08:09)Shiro Kawai wrote:


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/