error on startup with latest guile-vm

error on startup with latest guile-vm

[Julian Graham]

Hey Andy,

I just pulled the latest patches from `vm' and did a make; make
install, and got the following the exception when I tried to launch
the REPL.  I should point out that this is the first time I've been
actually able to build the vm branch all the way to installation, so
it could very well be something weird in my environment -- I just
don't know where to start looking.  Here's the stack trace:

Backtrace:
In unknown file:
   ?:  0* [map #<program b7c8a100> ((#) (#) (#) (#) ...)]
   ?:  1* [#<program b7c8a100> ((ice-9 session))]
   ?:  2* [dynamic-wind #<program b7c9a8f0> #<program b7c9a890>
#<program b7c9a888>]
   ?:  3* [#<program b7c9a890>]
   ?:  4* [dynamic-wind #<program b7c9a358> #<program b7c9a370>
#<program b7c9a350>]
   ?:  5* [#<program b7c9a370>]
   ?:  6* [load-compiled/vm "/usr/local/share/guile/1.9/ice-9/session.go"]
   ?:  7* [dynamic-wind #<program b7c9ce88> #<program b7c9ce80>
#<program b7c9ce60>]
   ?:  8* [#<program b7c9ce80>]
   ?:  9* [dynamic-wind #<program b7c9cb38> #<program b7c9cb68>
#<program b7c9cb30>]
   ?: 10* [#<program b7c9cb68>]
   ?: 11* [load-compiled/vm "/usr/local/share/guile/1.9/ice-9/documentation.go"]
   ?: 12* [dynamic-wind #<program 80ee340> #<program 80ee348>
#<program 80ee350>]
   ?: 13* [#<program 80ee348>]
   ?: 14* [dynamic-wind #<program 80ee4f0> #<program 80ee4d8>
#<program 80ee4f8>]
   ?: 15* [#<program 80ee4d8>]
   ?: 16* [load-compiled/vm "/usr/local/share/guile/1.9/system/vm/program.go"]
   ?: 17* [dynamic-call "scm_init_programs" #<dynamic-object "libguile">]

ERROR: In procedure dynamic-func:
ERROR: /usr/lib/libguile.so.17: undefined symbol: scm_init_programs


Regards,
Julian

Re: error on startup with latest guile-vm

[Ludovic Courtès]

Hi Julian,

"Julian Graham" <joolean@gmail.com> writes:

> #<program 80ee4f8>]
>    ?: 15* [#<program 80ee4d8>]
>    ?: 16* [load-compiled/vm "/usr/local/share/guile/1.9/system/vm/program.go"]
>    ?: 17* [dynamic-call "scm_init_programs" #<dynamic-object "libguile">]
>
> ERROR: In procedure dynamic-func:
> ERROR: /usr/lib/libguile.so.17: undefined symbol: scm_init_programs

It appears that the wrong `libguile.so' is being picked up.  Assuming
you installed the thing with `--prefix=/usr/local' (which is the
default), it should be using `/usr/local/lib/libguile.so'.  You may need
to adjust your LD_LIBRARY_PATH or something.

Hope this helps,
Ludo'.

Re: error on startup with latest guile-vm

[Julian Graham]

> It appears that the wrong `libguile.so' is being picked up.  Assuming
> you installed the thing with `--prefix=/usr/local' (which is the
> default), it should be using `/usr/local/lib/libguile.so'.  You may need
> to adjust your LD_LIBRARY_PATH or something.

Yep, that was it -- durr, thanks.  Sorry about that.

Next up on the list of stupid questions: I was able to compile one of
my modules, but the resulting .go file got created with (what looks
like) a weirdly restrictive umask -- mode 600, and not, say, 664 or
644.  Did I do something wrong?

Re: error on startup with latest guile-vm

[Andy Wingo]

Hi,

On Wed 15 Oct 2008 07:17, "Julian Graham" <joolean@gmail.com> writes:

> Next up on the list of stupid questions: I was able to compile one of
> my modules, but the resulting .go file got created with (what looks
> like) a weirdly restrictive umask -- mode 600, and not, say, 664 or
> 644.  Did I do something wrong?

Nope, it's just abuse of POSIX -- we create a temporary file then relink
it if the compilation succeeded. Temp files have those perms by default.
Patches welcome, to fix this to use the user's umask, to
call-with-output-file/atomic in (system base compile).

As an aside, it used to be that the file was truncated then rewritten.
This was bad. That part of the kernel is notoriously buggy. More
relevantly, truncating an mmapped file will cause SIGBUS, which we must
avoid. It seems that we are fine with our mmap strategy -- like with ELF
files -- but we have to take some care.

While I'm rambling, I was looking at the ELF format today, inspired by
razor (an *incredibly* fast package manager --
http://github.com/krh/razor/wikis/razor-design), and we really should do
more of that style of coding. Specifically regarding ELF, we should be
able to dump tagged SCM values into an ELF file, which would help with
undumping. And we should be able to write native code to ELF files. But
this is all future work. There is lots of exciting stuff to do in this
regard.

Andy
-- 
http://wingolog.org/

Re: error on startup with latest guile-vm

[Ludovic Courtès]

Hi,

Andy Wingo <wingo@pobox.com> writes:

> While I'm rambling, I was looking at the ELF format today, inspired by
> razor (an *incredibly* fast package manager --
> http://github.com/krh/razor/wikis/razor-design), and we really should do
> more of that style of coding. Specifically regarding ELF, we should be
> able to dump tagged SCM values into an ELF file, which would help with
> undumping. And we should be able to write native code to ELF files. But
> this is all future work. There is lots of exciting stuff to do in this
> regard.

The key to serializability here is to store offsets within a region
instead of raw pointers (which is often referred to as "pointer
swizzling").  The problem is that there's a run-time overhead associated
with it: you can't just dereference pointers the usual way.

Thanks,
Ludo'.