Building binutils "bootstrap"-wise

Building binutils "bootstrap"-wise

[Christian Jönsson]

I would like to discuss the idea of building the binutils suite using
a "bootstrap" approach similar to gcc's.

The idea is to reduce the impact that the system-installed binutils
has on the to-build binutils executables and friends and thus its test
results.

However, this means longer compilation time to build the binutils
suite.

Comments anyone?

Cheers,

/ChJ

Re: Building binutils "bootstrap"-wise

[Nick Clifton]

Hi Christian,

> I would like to discuss the idea of building the binutils suite
> using a "bootstrap" approach similar to gcc's.
> 
> The idea is to reduce the impact that the system-installed binutils 
> has on the to-build binutils executables and friends and thus its
> test results.  However, this means longer compilation time to build
> the binutils suite.

Hmm - Well there is certainly no reason why bootstrapping targets
could not be included in the makefiles.  I guess it is a question of
whether people would actually use them when testing their patches.

Building binutils at the moment is a relatively quick process.
(Compared to building gcc say).  So I doubt if the extra time required 
to run bootstrap target would be a great deterrent.

Note that the linker testsuite already has a small set of bootstrap
tests (ld/testsuite/ld-bootstrap).

Cheers
        Nick

Re: Building binutils "bootstrap"-wise

[Christian Jönsson]

On Fri, Mar 01, 2002 at 09:13:30AM +0000, Nick Clifton wrote:

Hej Nick

> Hi Christian,
> 
> > I would like to discuss the idea of building the binutils suite
> > using a "bootstrap" approach similar to gcc's.
> > 
> > The idea is to reduce the impact that the system-installed binutils 
> > has on the to-build binutils executables and friends and thus its
> > test results.  However, this means longer compilation time to build
> > the binutils suite.
> 
> Hmm - Well there is certainly no reason why bootstrapping targets
> could not be included in the makefiles.  I guess it is a question of
> whether people would actually use them when testing their patches.

Sure, or even if we'd encourage people to do so...

> 
> Building binutils at the moment is a relatively quick process.
> (Compared to building gcc say).  So I doubt if the extra time required 
> to run bootstrap target would be a great deterrent.

I wouldn't hope so...

> 
> Note that the linker testsuite already has a small set of bootstrap
> tests (ld/testsuite/ld-bootstrap).

I don't follow you on this comment and perhaps I'm unclear about what
I mean. I mean that I would like to have a "three" stage build of the
binutils sources. A first build with the already installed "binutils",
a preliminary build with the first stage built binutils, a "final"
build to check wether there are binary differences in the built files
(*.o mainly...) Such an approach would level some problems at the cost
of some build time.

And, as always, I'm a great guy (humble, right?) in having ideas, I am
not that great in implementing them and suggesting patches... So, I'm
not pushing this, just advocating a little...

Cheers,

Re: Building binutils "bootstrap"-wise

[Ian Lance Taylor]

Nick Clifton <nickc@cambridge.redhat.com> writes:

> Hmm - Well there is certainly no reason why bootstrapping targets
> could not be included in the makefiles.  I guess it is a question of
> whether people would actually use them when testing their patches.

There is already a bootstrap target in the gas Makefile and the ld
Makefile.  In either case, it will do a full three-stage bootstrap
test.  (Well, they used to; I haven't tested them recently.)

There is no full three stage bootstrap which uses both the linker and
the assembler.  That is, there is no three stage test which ensures
that the linker can correctly link the files which the assembler
generates.

It would be quite easy to modify the gas bootstrap test to use the
newly built linker, if there is one.  That would be a full three-stage
bootstrap.

Ian