[distcc] gcc bootstraps with distcc

[Alexandre Oliva]

On Jul 11, 2003, Neil Booth <neil at daikokuya.co.uk> wrote:

> Alexandre Oliva wrote:-
>> >> +      _cpp_do_file_change (pfile, LC_RENAME, dir_with_slashes, 1, 0);
>> >> +      _cpp_do_file_change (pfile, LC_RENAME, name, 1, 0);
>> >> +    }
>> 
>> > Is the second rename necessary?  I don't see why it should be;
>> > if not let's not do it.
>> 
>> AFAICT, yes.  Without it, -E -fpreprocessed is not equivalent to
>> `cat'.

> <builtin-in> doesn't do this; you used to cite <built-in> as precedent
> for what you do.

Err...  This has nothing to do with the built-in.  It has to do with
the fact that, without the second rename, we'll emit a

# 2 "main_filename"

after the directory line marker when doing -E -fpreprocessed.

> cpplib is a library.  It couldn't care less if the client has a problem.

In my book, libraries should propagate errors in callbacks out to its
callers.

> Debug output is a client-specific issue.

This is not about debug output.  This is about preserving directory
information in preprocessing.

> We're talking about a cpplib feature, and so a specific client error
> is obviously out of place in cpplib.

This means the error, at its location, should be rephrased, not
removed.  But I'll instead move it into the callback.  This should be
fine, even though it won't trigger an error in case of -E
-fpreprocessed, which is why I placed the logic for printing the error
where I did.

> Do you tell cpplib to report if the compiler meets an invalid tree node?

No, but cpplib does report an error if it encounters an invalid
pragma.

> This whole thing is a cpplib feature for which debug output is
> orthogonal - it just happens to be the use that this client puts the
> feature to.  -gpwd and -gno-pwd therefore, whilst better than -M stuff,
> are not ideal.

Agreed.

> -fpwd has builtin-in support for the negative, is forwarded to CPP
> automically, and is therefore the clear choice.

I don't see anything clear about it.  -f to me is an option to the
compiler.  This one isn't.  It's an option to the preprocessor.  I'm
currently thinking -Pwd would be the best choice, since -P has to do
with printing line markers, but If you care so much about -fpwd, I'll
go with that.  I don't care.  I want the feature, I don't want
complexity, I don't want this debate to go on forever, I just want you
and Zack to agree on what both of you find acceptable such that I can
go ahead and get the patch in.  I'm tired of getting agreement from
one of you, waste the time implementing the change, just to have the
other disagree with what the one explicitly agreed with.  This has
been very frustrating.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist                Professional serial bug killer