Bug in gencmdlist.sh (or is it?)

Bug in gencmdlist.sh (or is it?)

[Duboucher Thomas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

	Hi,

	I was working on some Grub2 module a few days ago and I spent a lot of
time on this. Basically, I was unable to correctly generate the file
command.lst; my new commands simply didn't appeared in it.
	I finally found that gencmdlist.sh is only processing line per line
using sed, so that if one is writing with a different indentation, the
script silently fail. For instance,
> grub_extcmd
> (
>     "foo",
does not produce any output. I am working now with a small script
written in Lua, but it is neither efficient, nor a good idea to add Lua
as a dependency. Perhaps a sed guru can have a look at this?

	Thomas.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkpL5HwACgkQBV7eXqefhqg3hwCfX4Fn6icORyZKygqZJo1eqwWH
LSAAnjUCLniZX9pu0OCp4CV7EEjNXAp7
=AKuH
-----END PGP SIGNATURE-----

Re: Bug in gencmdlist.sh (or is it?)

[Pavel Roskin]

On Thu, 2009-07-02 at 00:34 +0200, Duboucher Thomas wrote:

> 	I was working on some Grub2 module a few days ago and I spent a lot of
> time on this. Basically, I was unable to correctly generate the file
> command.lst; my new commands simply didn't appeared in it.
> 	I finally found that gencmdlist.sh is only processing line per line
> using sed, so that if one is writing with a different indentation, the
> script silently fail. For instance,
> > grub_extcmd
> > (
> >     "foo",
> does not produce any output.

I confirm that it's indeed a limitation of gencmdlist.sh.  It may be
possible to handle it by using C preprocessor before sed, but I don't
think it's an urgent issue.  It's shouldn't be a problem for properly
indented source.

>  I am working now with a small script
> written in Lua, but it is neither efficient, nor a good idea to add Lua
> as a dependency.

I don't understand how this is related.

>  Perhaps a sed guru can have a look at this?

I actually don't feel good about using anything other that a C compiler
or preprocessor to parse C sources.

It's working for now, but if we want to make it more reliable, I'd
rather not ask a "sed guru".

-- 
Regards,
Pavel Roskin

Re: Bug in gencmdlist.sh (or is it?)

[Duboucher Thomas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Pavel Roskin a écrit :
> On Thu, 2009-07-02 at 00:34 +0200, Duboucher Thomas wrote:
> 
>> 	I was working on some Grub2 module a few days ago and I spent a lot of
>> time on this. Basically, I was unable to correctly generate the file
>> command.lst; my new commands simply didn't appeared in it.
>> 	I finally found that gencmdlist.sh is only processing line per line
>> using sed, so that if one is writing with a different indentation, the
>> script silently fail. For instance,
>>> grub_extcmd
>>> (
>>>     "foo",
>> does not produce any output.
> 
> I confirm that it's indeed a limitation of gencmdlist.sh.  It may be
> possible to handle it by using C preprocessor before sed, but I don't
> think it's an urgent issue.  It's shouldn't be a problem for properly
> indented source.

No, it is not really urgent. As I said, I have already found a
workaround. But based on Grub2 being a module-based project, I don't
think an indentation-dependant parser is a good idea, nor is it robust.

> 
>>  I am working now with a small script
>> written in Lua, but it is neither efficient, nor a good idea to add Lua
>> as a dependency.
> 
> I don't understand how this is related.
> 

I have replaced gencmdlist.sh by a script in Lua that does the same, and
can handle any kind of indentation - or at least the resulting file is
identical even if I mess the indentation; but based on how and when I
have writen it, well, I'm not very confident ;) -.

>>  Perhaps a sed guru can have a look at this?
> 
> I actually don't feel good about using anything other that a C compiler
> or preprocessor to parse C sources.
> 
> It's working for now, but if we want to make it more reliable, I'd
> rather not ask a "sed guru".
> 

There are tools to parse source code, but that is not my domain. That's
the reason I was reporting this. :)

	Regards,
	Thomas.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkpMSM8ACgkQBV7eXqefhqgzGwCcD5cXmwsCREdwNUBjgvACc9xK
xxoAoLIBToOcZ5oig5T/ihd9av/wioed
=6XCt
-----END PGP SIGNATURE-----

Re: Bug in gencmdlist.sh (or is it?)

[Pavel Roskin]

On Thu, 2009-07-02 at 07:42 +0200, Duboucher Thomas wrote:

> No, it is not really urgent. As I said, I have already found a
> workaround. But based on Grub2 being a module-based project, I don't
> think an indentation-dependant parser is a good idea, nor is it robust.

I agree.

> >>  I am working now with a small script
> >> written in Lua, but it is neither efficient, nor a good idea to add Lua
> >> as a dependency.
> > 
> > I don't understand how this is related.
> > 
> 
> I have replaced gencmdlist.sh by a script in Lua that does the same, and
> can handle any kind of indentation - or at least the resulting file is
> identical even if I mess the indentation; but based on how and when I
> have writen it, well, I'm not very confident ;) -.

I think an improved approach would be one of those:

1) Make gencmdlist.sh call cpp with a special macro, that would expand
into something that can be reliably extracted by sed or grep and placed
into command.lst.

2) Use a macro to write the symbol into a special linker section.
gencmdlist.sh would read it to generate command.lst at the build time.

3) Use a macro to write the symbol into a special linker section.
grub-install would run a new utility to generate command.lst from the
modules in /boot/grub.

The later is very appealing, but it would be a big change in the way we
are doing things.

-- 
Regards,
Pavel Roskin