visibility attribute not supported?!

visibility attribute not supported?!

[Jeremy Kolb]

I'm running linux 2.6.3, gcc 3.3.3 on an amd t-bird.  When I try to
compile glibc I get a warning about the visibility attribute not being
supported. I've tested this out with a test program that uses the
visibility attribute and that fails as well.  So glibc won't build for
me.  How do I enable the visibility attribute? I've recompiled binutils
and gcc but I still get the same error.

Jeremy

Re: visibility attribute not supported?!

[llewelly]

Jeremy Kolb <jkolb@brandeis.edu> writes:

> I'm running linux 2.6.3, gcc 3.3.3 on an amd t-bird.  When I try to
> compile glibc I get a warning about the visibility attribute not being
> supported. I've tested this out with a test program that uses the
> visibility attribute and that fails as well.  So glibc won't build for
> me.  How do I enable the visibility attribute? I've recompiled binutils
> and gcc but I still get the same error.

First, make sure you are using fsf gcc 3.3.3 and not the derivative
shipped by your distro maker (it seems all distros hand-roll their own
gcc derivative ... don't ask me why.)
If you can still reproduce with fsf gcc, I think you should report
this as a bug. There's a similar bug 
reported against 3.4-branch: 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13134
but it doesn't mention 3.3.3 .

Re: visibility attribute not supported?!

[Jeremy Kolb]

Yeah It's stock fsf gcc. The distro I use is Sourcemage and it downloads
from gcc directly from the gnu site (and it's md5 verified).  The thing
is this has only happened to a few people and not the rest, we don't
know why.

Jeremy

On Thu, 2004-03-18 at 23:18, llewelly@xmission.com wrote:
> Jeremy Kolb <jkolb@brandeis.edu> writes:
> 
> > I'm running linux 2.6.3, gcc 3.3.3 on an amd t-bird.  When I try to
> > compile glibc I get a warning about the visibility attribute not being
> > supported. I've tested this out with a test program that uses the
> > visibility attribute and that fails as well.  So glibc won't build for
> > me.  How do I enable the visibility attribute? I've recompiled binutils
> > and gcc but I still get the same error.
> 
> First, make sure you are using fsf gcc 3.3.3 and not the derivative
> shipped by your distro maker (it seems all distros hand-roll their own
> gcc derivative ... don't ask me why.)
> If you can still reproduce with fsf gcc, I think you should report
> this as a bug. There's a similar bug 
> reported against 3.4-branch: 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13134
> but it doesn't mention 3.3.3 .
> 

Re: visibility attribute not supported?!

[llewelly]

Jeremy Kolb <jkolb@brandeis.edu> writes:

> Yeah It's stock fsf gcc. The distro I use is Sourcemage and it downloads
> from gcc directly from the gnu site (and it's md5 verified).  The thing
> is this has only happened to a few people and not the rest, we don't
> know why.

As near as I can tell, gcc should build with visibility support iff
    you are using gnu ld.

Since you are running linux, I can't imagine you might be using
    anything else. So I suspect a bug somewheres. When you file a bug
    report, you should include the output of ld --version, gcc
    --version, and uname -a, in addition to the info requested at
    gcc.gnu.org/bugs.html .

I suppose you could reconfigure and rebuild gcc, adding --with-gnu-ld
    to your configure arguments, but it seems a bug that that should
    be necessary on a platform where SFAIK only gnu ld is available.

new computer language with GCC

[Eljay Love-Jensen]

Hello everyone,

I would like to try my hand at implementing a computer language in GCC.

The language is "lambda calculus brainf---".  Yes, "brainf---" is a dirty 
word.  And best of all, it's a VERY VERY SIMPLE language.  (Look it up in 
Google.  Amazingly terse.)

Does anyone know of a good "Idiot's Guide to Get Your Pet Programming 
Language Working With GCC As The Backend"?  An O'Reilly book or a HowTo URL?

I've been through the Compilers (Aho, Sethi, Ulman) "dragon" book.  I'd 
like to try my hand at the real thing.

I don't have any delusions that I'll be the next Niklaus Wirth.  I don't 
think anyone will be as prolific as he has been in formulating programming 
languages.

I could write a LCB --> C translator.  But that's a CFront kind of 
solution; I want to get some experience working with GCC.

Thanks,
--Eljay

Re: new computer language with GCC

[llewelly]

Eljay Love-Jensen <eljay@adobe.com> writes:

> Hello everyone,
> 
> I would like to try my hand at implementing a computer language in GCC.
> 
> The language is "lambda calculus brainf---".  Yes, "brainf---" is a
> dirty word.  And best of all, it's a VERY VERY SIMPLE language.  (Look
> it up in Google.  Amazingly terse.)
> 
> Does anyone know of a good "Idiot's Guide to Get Your Pet Programming
> Language Working With GCC As The Backend"?  An O'Reilly book or a
> HowTo URL?

http://gcc.gnu.org/readings.html has 3 links which are helpful for
    frontends. The front-end howto, the COBOL front-end manual, and
    the toy example language. However I understand they are of varying
    degrees of obsolesence.

Most of the experts on this topic frequent the gcc mailing list.

Re: new computer language with GCC

[Eljay Love-Jensen]

Looks like <http://www.tldp.org/HOWTO/GCC-Frontend-HOWTO> is exactly what I 
needed.

Thanks,
--Eljay

Crafting Better Source Code in C++

[Eljay Love-Jensen]

Hi everyone,

#define SOAPBOX

I've been asked this question several times about C++ (in particular, GCC's 
C++), and I thought I'd share my thoughts.

The GCC compiler has a suite of warnings which will point out dangerous or 
suspicious coding practices.  By and large, I feel that these GCC warnings 
are indicative of either bad code ("bugs"!  ~10% of the time), or poor 
coding practices (~90%).

In either case, I think it is a worthwhile endeavor to use GCC's warning 
facilities to help improve the source code, and improve the portability of 
the source code.

At a minimum, I suggest everyone ALWAYS use these flags:
-pedantic -Wall -W

Other flags are documented here:
http://gcc.gnu.org/onlinedocs/gcc-3.3.3/gcc/Warning-Options.html

I use almost all the additional flags.  Call me ultra-pedantic.  I wish 
there was an -WALL switch, to enable all warnings (even the annoying ones 
which I disregard), just so I wouldn't have the creepy feeling I've missed one.

I have heard of people unhappy that there isn't a good GNU Lint++ for C++ 
code in general.  In my opinion, a good chunk of Lint++ facility is built 
right into the GCC compiler.  If only people would enable it.

Granted... Gimpel's PC-lint / FlexeLint software does an extensive analysis 
beyond what GCC has built in.  Gimpel's software has a different objective 
than GCC.

#undef SOAPBOX

Thanks,
--Eljay

Re: Crafting Better Source Code in C++

[llewelly]

Eljay Love-Jensen <eljay@adobe.com> writes:

> Hi everyone,
> 
> #define SOAPBOX
> 
> I've been asked this question several times about C++ (in particular,
> GCC's C++), and I thought I'd share my thoughts.
> 
> The GCC compiler has a suite of warnings which will point out
> dangerous or suspicious coding practices.  By and large, I feel that
> these GCC warnings are indicative of either bad code ("bugs"!  ~10% of
> the time), or poor coding practices (~90%).
> 
> In either case, I think it is a worthwhile endeavor to use GCC's
> warning facilities to help improve the source code, and improve the
> portability of the source code.
> 
> At a minimum, I suggest everyone ALWAYS use these flags:
> -pedantic -Wall -W

As more and more C libraries incorporate C99 features, fewer and fewer
    people will be able to use -pedantic with C++ . Many real projects
    will need to seperate that code which can be written with strict
    conformance from that which requires non standard features, and
    use appropriately different compile flags.

While we are at it, however, I suggest another warning flag: -O
    :-) . Certain warnings (like 'foo unused', and 'foo may be used
    before it is initialized' ) require data-flow analysis, and so are
    only enabled with -On, e.g. -Wunitialized.

> 
> Other flags are documented here:
> http://gcc.gnu.org/onlinedocs/gcc-3.3.3/gcc/Warning-Options.html
> 
> I use almost all the additional flags.  Call me ultra-pedantic.  I
> wish there was an -WALL switch, to enable all warnings (even the
> annoying ones which I disregard), just so I wouldn't have the creepy
> feeling I've missed one.
[snip]

Compile this program:

    #include<iostream>
    #include<ostream>

    int main()
      {
        std::cout << "Hello World!" << std::endl;
      }

with: g++ -Weffc++ -Wsystem-headers

Alternatively, try building any boost-using C++ with -Weffc++, without
    -Wsystem-headers.

Personally, I think the flags not enabled by -W -Wall are only useful
    in very rare circumstances.