Re: bootstrap fails on Debian ppc64 with fatal error: bits/libc-header-start.h: No such file or directory

[Dennis Clarke <dclarke@blastwave.org>]

>>> Honestly, I find that hard to believe.
>>
>> Me too!
>>
>>>
>>> One or the other is needed. If you need both, I think you're the only
>>> person in the world who needs that.
>>>
>>
>> Not bloody likely right?  I am looking into that and I think the real
>> issue is that the versions wanted in the source tree must be the same as
>> the stuff dragging in by the script kicker download_prerequisites :
>>
>>       https://gcc.gnu.org/install/prerequisites.html
>>
>>       The in-tree build is only supported with the GMP version that
>>       download_prerequisites installs.
> 
> Right, you should be using that script.
> 
> https://gcc.gnu.org/wiki/InstallingGCC
>

Well what difference does it make anyways?
The correct versions are the correct versions.

> Or (as that page says) just install the debian packages and stop
> installing them by hand.

But I want to have the libs around as fully non-optimized and no
assembly single steppable debug enabled full of goodness. Which I
can certianly do.


>> However what do I know? There is another page that says do a bootstrap4
>> for extra spice and flavour in your result. That doesn't fly.
> 
> Well stop trying to do weird stuff and just Keep It Simple.
>

That just drains the joy out of everything. However to be really clear
all I wanted was to boot strap gcc on a few machines and as usual I run
into problems. Most people these days never see the sources and only get
packages pre-built by someone else somewhere magic.  Sort of defeats the
whole reason why the source is open right? May as well just install MS
Windows Vista and be happy.

Also there is great learning stuff to be had in the process. Not the 
least of which is from good folks like Segher Boessenkool :

     https://gcc.gnu.org/pipermail/gcc-help/2022-May/141568.html

There is a valid point there. Am I building a compiler which is already
crippled? Hard to know for sure.

>>
>> Whoa ... hold on a sec. I generally never use the little magic script
>> download_prerequisites. I had better check that :
> 
> You should use it. It makes everything much simpler than you're making it.
>

Well it does the same thing and all that is left is the patch for mpfr.

>> Okay so it needs 'wget'. Fine. Easy to fix that.
> 
> It uses curl instead if wget is not found. Most people have one or the
> other already installed.

Yeah, I prefer curl. Regardless ...

> 
>> Same stuff I use however the mpfr *needs* a patch.
> 
> Eh? Why?

Those folks working on gmp and mpfr really are the experts. The page
says :

     https://www.mpfr.org/mpfr-current/#download

     The bugs listed below have been fixed (latest update: 2021-05-17).

     See https://www.mpfr.org/mpfr-current/#fixed

Thirteen bug fixes and from what I have heard directly from some good
folks the code in GMP has been subjected to an actual proof. The MPFR
code is in process. I gather those bug fixes are valuable and trivial
to apply. In fact the download_prerequisites script could be updated
to fetch the patches and just apply them.

> 
>> I must have made a mistake somewhere.
>>
.
.
.
> 
> I would also delete the source tree and untar it again, in case you've
> fouled it up somehow. And then run contrib/download_prerequisites
> again.

Yep.

Starting over from scratch seems like a good idea on Friday the 13th.

-- 
Dennis Clarke
RISC-V/SPARC/PPC/ARM/CISC
UNIX and Linux spoken
GreyBeard and suspenders optional