`char * s = "abc";' causes compile to crash with a segfault

`char * s = "abc";' causes compile to crash with a segfault

[Alexander Sorockin]

I realize that compile is only an example and you are probably familiar  
with the problem, but just in case... I was playing around with compile  
and here's what got.

------ In Linux (sparse-0.4.1, gcc 4.3.3):

$ ./compile -
char * s = "abc";
-:1:6: warning: symbol 's' was not declared. Should it be static?
	.file	"-"
  =
Segmentation fault

------ In CygWin (sparse-0.4.1, gcc 3.4.4):

$ ./compile -
char * s = "abc";
-:1:6: warning: symbol 's' was not declared. Should it be static?
         .file   "-"
  =
       7 [main] compile 1676 _cygtls::handle_exceptions: Error while  
dumping state (probably corrupted stack)
Segmentation fault (core dumped)

------

Just out of curiosity, is sparse stable enough in general to be used as a  
full-fledged front-end for a C compiler?

Best,
Sandy

Re: `char * s = "abc";' causes compile to crash with a segfault

[Jeff Garzik]

Alexander Sorockin wrote:
> I realize that compile is only an example and you are probably familiar 
> with the problem, but just in case... I was playing around with compile 
> and here's what got.

Yes, that is expected behavior for the program merged as "silly sparse 
x86 backend" :)  There are many, many, many code generator shortcomings.


> Just out of curiosity, is sparse stable enough in general to be used as 
> a full-fledged front-end for a C compiler?

As long as you don't do super-weird, non-standard C:  sure.

	Jeff

Re: `char * s = "abc";' causes compile to crash with a segfault

[David Given]

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

Jeff Garzik wrote:
[...]
>> Just out of curiosity, is sparse stable enough in general to be used
>> as a full-fledged front-end for a C compiler?
> 
> As long as you don't do super-weird, non-standard C:  sure.

Well --- I did find a whole bunch of bugs when working on Clue. OTOH
that was a while ago, so some of them might have been fixed by now. Look
back in the archives for 'Odd sparse behaviour' posted on 2008-07-06,
and 'Odd behaviour with OP_SCAST' posted on 2008-09-08.

In addition, I found a number of places where minor tweaks to what data
Sparse keeps around can make a compiler writer's life much easier; for
example, keeping more type data makes finding the type of a pseudo much
easier. There's a combined patch shipped with Clue that fixes and
enhances a bunch of this stuff:

http://cluecc.svn.sourceforge.net/viewvc/cluecc/clue/sparse.patch?revision=34&view=markup

(In fact, if you're generating llvm assembly directly, you may find it
easier just to add a backend to Clue rather than write your own, er,
middle-end --- a lot of the simplifications Clue does for languages like
Javascript also apply to llvm bitcode.)

<plug> Did I mention that Clue is now converting C into Java source that
runs at 40% of compiled C? </plug>

- --
┌─── ｄｇ＠ｃｏｗｌａｒｋ．ｃｏｍ ───── http://www.cowlark.com ─────
│
│ "People who think they know everything really annoy those of us who
│ know we don't." --- Bjarne Stroustrup
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFJ+Nf6f9E0noFvlzgRAvpIAKCcJG0/SZ0mbx5Iel0iR2GbfJpclACgiQuw
/pkelbkI9I0FkdXqX/bTbQ0=
=581X
-----END PGP SIGNATURE-----
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: `char * s = "abc";' causes compile to crash with a segfault

[Christopher Li]

On Wed, Apr 29, 2009 at 3:43 PM, David Given <dg@cowlark.com> wrote:
>
> In addition, I found a number of places where minor tweaks to what data
> Sparse keeps around can make a compiler writer's life much easier; for
> example, keeping more type data makes finding the type of a pseudo much
> easier. There's a combined patch shipped with Clue that fixes and
> enhances a bunch of this stuff:
>
> http://cluecc.svn.sourceforge.net/viewvc/cluecc/clue/sparse.patch?revision=34&view=markup

I think some of your patch already went into sparse. If there are some
missing. Feel free to submit it again.

> (In fact, if you're generating llvm assembly directly, you may find it
> easier just to add a backend to Clue rather than write your own, er,
> middle-end --- a lot of the simplifications Clue does for languages like
> Javascript also apply to llvm bitcode.)

If every back end needs to do those "simplifications", that means sparse
should be fixed to do it right.

> <plug> Did I mention that Clue is now converting C into Java source that
> runs at 40% of compiled C? </plug>

Oh really, Java?

Chris

Re: `char * s = "abc";' causes compile to crash with a segfault

[Alexander Sorockin]

On Thu, 30 Apr 2009 01:08:59 +0400, Jeff Garzik <jeff@garzik.org> wrote:

>
> As long as you don't do super-weird, non-standard C:  sure.
>

Hmm. It's interesting that sparse itself doesn't seem to compile with  
anything other that gcc. Does that mean that most other compilers are  
"non-standard"?

Sandy

Re: `char * s = "abc";' causes compile to crash with a segfault

[Christopher Li]

On Thu, Apr 30, 2009 at 1:55 AM, Alexander Sorockin <sorockin@yandex.ru> wrote:
> Hmm. It's interesting that sparse itself doesn't seem to compile with
> anything other that gcc. Does that mean that most other compilers are
> "non-standard"?

Sparse use some C99 feature. Those "non-standard" compiler without C99
support can't compiling it. I don't think sparse has much gcc specific
stuff though.

Chris

Re: `char * s = "abc";' causes compile to crash with a segfault

[David Given]

Christopher Li wrote:
[...]
> I think some of your patch already went into sparse. If there are some
> missing. Feel free to submit it again.

The bulk of it's the byte-size patch, which has already made it in, and 
the rest is Clue-specific tweaks (like keeping type information in 
pseudos, and so on).

[...]
> If every back end needs to do those "simplifications", that means sparse
> should be fixed to do it right.

Actually that stuff's in Clue rather than in the patch --- things like 
ultra-simple register allocation, flattening Sparse's parse tree to 
Clue's backend model, etc. None of this really makes sense for anything 
involving a real compiler (for example, Clue assumes you have an 
infinite number of registers available).

In fact, now I think of it, Clue also has a rather intrinsic assumption 
that sizeof(int) == sizeof(char) == 1, which means it wouldn't work with 
llvm out of the box. It may be productive to rip off some of the code, 
though.

-- 
David Given
dg@cowlark.com