Parallel make also broken now...

Parallel make also broken now...

[Steven Bosscher]

Who approved this patch:

        * Makefile.in (rtl.o): Remove -DGENERATOR_FILE.
        (bitmap.o): Likewise.

It breaks parallel make for me.

Gr.
Steven

Re: Parallel make also broken now...

[Zack Weinberg]

Steven Bosscher <stevenb@suse.de> writes:

> Who approved this patch:
>
>         * Makefile.in (rtl.o): Remove -DGENERATOR_FILE.
>         (bitmap.o): Likewise.

Me.

> It breaks parallel make for me.

Changes to commands can't break parallel make; the bug is elsewhere.
Please say how it failed.

zw

Re: Parallel make also broken now...

[Steven Bosscher]

On Sep 06, 2004 06:53 PM, Zack Weinberg <zack@codesourcery.com> wrote:
> > It breaks parallel make for me.
> 
> Changes to commands can't break parallel make; the bug is elsewhere.
> Please say how it failed.

It breaks when building bitmap.o, but insn-conditions.h and
another insn-something-file were not yet generated.

Gr.
Steven

Re: Parallel make also broken now...

[Kelley Cook]

> On Sep 06, 2004 06:53 PM, Zack Weinberg <zack@codesourcery.com> wrote:
>>> It breaks parallel make for me.
>> 
>> Changes to commands can't break parallel make; the bug is elsewhere.
>> Please say how it failed.
> 
> It breaks when building bitmap.o, but insn-conditions.h and
> another insn-something-file were not yet generated.
> 
> Gr.
> Steven

Is this occuring during stage2?

If so it is, as far as I can guess, the same apparent gmake bug that prevented me 
from committing an essentially approved patch which IMO would have had a side 
effect of solving Mohan's gcc-none.o problem more elegantly than his patch does.

http://gcc.gnu.org/ml/gcc-patches/2004-08/msg02155.html

The problem with its generic rules, make sometimes fails to serialize something that 
seems to be clearly is written out in the make file.

Essentially one of the insn-*.h files is getting rebuilt at the same time as an .o 
file that needs it, but that file does in fact list that .h file on its dependencies.

I circumvented the problem with a followup to the patch which didn't 
unnecessarily erase the generated *.[ch] files.  Nathanael was hesitant 
on that idea though.

Worse my original concept of only building those helper programs only once 
goes against the philosophy of the toplevel bootstrap.  It seems that in 
order to implement it, those files intended only to be built by the build 
compiler would probably have to be move to the toplevel.  

File a PR, cc: it to me and I'll look at another way to kludge around it tonight.

Kelley Cook

Re: Parallel make also broken now...

[Zack Weinberg]

Kelley Cook <kcook@gcc.gnu.org> writes:

>> On Sep 06, 2004 06:53 PM, Zack Weinberg <zack@codesourcery.com>
>> wrote:
>>>> It breaks parallel make for me.  Changes to commands can't break
>>> parallel make; the bug is elsewhere.  Please say how it failed.
>> It breaks when building bitmap.o, but insn-conditions.h and another
>> insn-something-file were not yet generated.  Gr.  Steven
>
> Is this occuring during stage2?

No, during stage 1, and the dependencies of rtl.o and bitmap.o really
were wrong.  This is something that automatic dependency generation
would have avoided, but not much else.  Removing -DGENERATOR_FILE from
the command line causes tm.h to include more files, see.

This has, AFAIK, been corrected as best we can.  I don't really follow
the rest of your comments, so I'm not sure whether I should still file
a bug report.

zw

Re: Parallel make also broken now...

[Mohan Embar]

Hi People,

>>>> It breaks parallel make for me.
>>> 
>>> Changes to commands can't break parallel make; the bug is elsewhere.
>>> Please say how it failed.
>> 
>> It breaks when building bitmap.o, but insn-conditions.h and
>> another insn-something-file were not yet generated.
>> 
>> Gr.
>> Steven
>
>Is this occuring during stage2?

>File a PR, cc: it to me and I'll look at another way to kludge around it tonight.

FWIW: Zack checked in a fix for this:

http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00577.html

-- Mohan
http://www.thisiscool.com/
http://www.animalsong.org/

Re: Parallel make also broken now...

[Mohan Embar]

Hi Kelley,

>If so it is, as far as I can guess, the same apparent gmake bug that prevented me 
>from committing an essentially approved patch which IMO would have had a side 
>effect of solving Mohan's gcc-none.o problem more elegantly than his patch does.

>http://gcc.gnu.org/ml/gcc-patches/2004-08/msg02155.html

When you say "his patch", which patch are you referring to? I rescinded this one:

http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00550.html

...and the second one:

http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00586.html

...merely reintroduces #include "config.h" to ggc-none.c, which we'd need to
do anyway given that the host-specific build  of ggc-none.o was removed here
(cf. item 5):

http://gcc.gnu.org/ml/gcc-patches/2004-08/msg00265.html

-- Mohan
http://www.thisiscool.com/
http://www.animalsong.org/