Embeddable Common-Lisp

Scribit Pierre THIERRY dies 10/02/2007 hora 10:17:
> There is still a flaw in my last patch, in that it won't check
> filename size against a value fetched by sysconf. It seems to me the
> value should only be fetched once (POSIX says the value won't change
> during the lifetime of a process), and I wanted to ask where to put
> such initialization and data.

Is it OK if I add a global variable path_max in ecl.h, and set it in
cl_boot()?

Curiously,
Pierre
--=20
nowhere.man@...
OpenPGP 0xD9D50D8A

2007/2/10, Pierre THIERRY <nowhere.man@...>:
> Is it OK if I add a global variable path_max in ecl.h, and set it in
> cl_boot()?

ECL uses _no_ global variables. All common data is stored either in
cl_env or in cl_core. Both are C structures with different fields. The
former is per-thread data, the second is made of constants or data
common to all threads. path_max should go into cl_core. Perhaps it
makes sense to define it _always_, because I do not like configuration
flags to populate the ECL headers.

I will gladly commit any patches for supporting Hurd.

Regards,

Juanjo

-- 
Dpto. de Fisica Teorica I, Fac. de CC Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://teorica.fis.ucm.es/~jjgarcia/

Scribit Juan Jose Garcia-Ripoll dies 11/02/2007 hora 00:20:
> ECL uses _no_ global variables.

That is rare... Was it for embeddability?

> Perhaps it makes sense to define it _always_, because I do not like
> configuration flags to populate the ECL headers.

OK. I'll define it statically where PATH_MAX or MAXPATHLEN are defined,
in cl_boot() otherwise.

> I will gladly commit any patches for supporting Hurd.

Thanks.

Quickly,
Pierre
--=20
nowhere.man@...
OpenPGP 0xD9D50D8A

2007/2/11, Pierre THIERRY <nowhere.man@...>:
> Scribit Juan Jose Garcia-Ripoll dies 11/02/2007 hora 00:20:
> > ECL uses _no_ global variables.
> That is rare... Was it for embeddability?

Keeping all "variables" in a single C structure is good for two
reasons: namespace and garbage collection. It ensures that all
routines know where the global variables are, and the programmer
ensures also that not too many variables are defined :-)

> > Perhaps it makes sense to define it _always_, because I do not like
> > configuration flags to populate the ECL headers.
>
> OK. I'll define it statically where PATH_MAX or MAXPATHLEN are defined,
> in cl_boot() otherwise.

No. I meant adding a new field to cl_core, and setting up a value in
cl_boot(). In case PATH_MAX is defined, just set cl_core.path_max to
that value, otherwise use a system call to guess it.

Juanjo

-- 
Dpto. de Fisica Teorica I, Fac. de CC Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://teorica.fis.ucm.es/~jjgarcia/

2007/2/11, Pierre THIERRY <nowhere.man@...>:
> There is one thing I'm not sure, not knowing ECL's internals: in
> backup_fopen, should the malloc'ed buffer be freed before any call to
> FElibc_error() or only before the return statement (as I do now)?

malloc() can not be used in ECL. ECL is a garbage collected program
and as such you should use the garbage collector to allocate memory.
The required routines are cl_alloc(), cl_alloc_atomic(), etc The call
to cl_dealloc() should then be placed before the main return. I will
audit your patches later.

Juanjo

-- 
Dpto. de Fisica Teorica I, Fac. de CC Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://teorica.fis.ucm.es/~jjgarcia/

Scribit Juan Jose Garcia-Ripoll dies 11/02/2007 hora 10:32:
> > OK. I'll define it statically where PATH_MAX or MAXPATHLEN are defined,
> > in cl_boot() otherwise.
> No. I meant adding a new field to cl_core, and setting up a value in
> cl_boot(). In case PATH_MAX is defined, just set cl_core.path_max to
> that value, otherwise use a system call to guess it.

That's what I did, in fact.

Homogeneously,
Pierre
=20
PS: no need to CC me, I had to subscribe to post anyway
--=20
nowhere.man@...
OpenPGP 0xD9D50D8A

Scribit Juan Jose Garcia-Ripoll dies 11/02/2007 hora 00:20:
> I will gladly commit any patches for supporting Hurd.

Did you find time to review the latest version of my patches?

Curiously,
Pierre
--=20
nowhere.man@...
OpenPGP 0xD9D50D8A

2007/3/20, Pierre THIERRY <nowhere.man@...>:
> Did you find time to review the latest version of my patches?

Sorry, it slipped out of my notebook. I have a fixed set of patches,
checked against current CVS. Could you test them?

Juanjo

-- 
Dpto. de Fisica Teorica I, Fac. de CC Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://juanjose.garciaripoll.googlepages.com

Scribit Juan Jose Garcia-Ripoll dies 20/03/2007 hora 21:40:
> Sorry, it slipped out of my notebook. I have a fixed set of patches,
> checked against current CVS. Could you test them?

At first glance, this one looks right. I'll check out the CVS HEAD and
see if the others patches apply cleanly and ECL builds fine on Hurd. Did
you use my original patches for the other ones, about
pthread_mutex_errorcheck_np, sa_siginfo and autoconf?

Quickly,
Pierre
--=20
nowhere.man@...
OpenPGP 0xD9D50D8A

2007/3/20, Pierre THIERRY <nowhere.man@...>:
> Scribit Juan Jose Garcia-Ripoll dies 20/03/2007 hora 21:40:
> > Sorry, it slipped out of my notebook. I have a fixed set of patches,
> > checked against current CVS. Could you test them?
>
> At first glance, this one looks right. I'll check out the CVS HEAD and
> see if the others patches apply cleanly and ECL builds fine on Hurd. Did
> you use my original patches for the other ones, about
> pthread_mutex_errorcheck_np, sa_siginfo and autoconf?

No, I just found one last email saying it was the definitive patches.
I missed then something?

Juanjo

-- 
Dpto. de Fisica Teorica I, Fac. de CC Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://juanjose.garciaripoll.googlepages.com

The port to GNU/Hurd has been incorporated to the CVS sources. Thanks, Pierre.

Juanjo

-- 
Facultad de Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://juanjose.garciaripoll.googlepages.com

Scribit Juan Jose Garcia-Ripoll dies 24/03/2007 hora 11:57:
> The port to GNU/Hurd has been incorporated to the CVS sources. Thanks,
> Pierre.

Thanks for the feedback and review. I see in the Changelog in CVS that
you're planning a 1.0 release. Do you already know when?

I also see you now support GNU/kfreeBSD, which makes ECL usable on all
arches and ports of Debian, which is great!

Curiously,
Pierre
--=20
nowhere.man@...
OpenPGP 0xD9D50D8A

2007/3/25, Pierre THIERRY <nowhere.man@...>:
> Thanks for the feedback and review. I see in the Changelog in CVS that
> you're planning a 1.0 release. Do you already know when?

I am making a long stay abroad on May. I hope to have some free time
then and work on the compiler. 1.0 will only come once the compiler
has become correct -- now, particularly with optimizations, it may
produce wrong code. I am happy about the rest of the environment,
though: the interpreter and the library are very stable and need only
some work to make them faster (which should be done after the 1.0)

> I also see you now support GNU/kfreeBSD, which makes ECL usable on all
> arches and ports of Debian, which is great!

BTW, is it easy to install Hurd? I might give it a try on a
virtualizer some day.

Juanjo

--
Facultad de Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)
http://juanjose.garciaripoll.googlepages.com