[gentoo-user] Reducing repeated checkings

[gentoo-user] Reducing repeated checkings

[Al]

Some thoughts of the day

With multicore machines compilation itself is relativly fast. The
bigger part of time meanwhile is spend for the system check before
compilation. Each  program das it again and 90% of the checks check
the same stuff.

I wounder if efforts have been made to reduce this repeated checks.

Al

Re: [gentoo-user] Reducing repeated checkings

[Alan McKinnon]

Apparently, though unproven, at 21:27 on Monday 23 August 2010, Al did opine 
thusly:

> Some thoughts of the day
> 
> With multicore machines compilation itself is relativly fast. The
> bigger part of time meanwhile is spend for the system check before
> compilation. Each  program das it again and 90% of the checks check
> the same stuff.
> 
> I wounder if efforts have been made to reduce this repeated checks.
> 
> Al

cmake.

It really does decrease the amount of time in the configure step - witness 
KDE4.

Then promptly causes a huge rash of other problems elsewhere.

It's software, what were you expecting? That it should actually work?


-- 
alan dot mckinnon at gmail dot com

Re: [gentoo-user] Reducing repeated checkings

[Al]

>>
>> I wounder if efforts have been made to reduce this repeated checks.
>>
>> Al
>
> cmake.

Didn't know that, Time to study the build tools, even if I am not a C-coder.

>
> It really does decrease the amount of time in the configure step - witness
> KDE4.
>
> Then promptly causes a huge rash of other problems elsewhere.

Nothing else expected. New features, new troubles. Once I really
believed programming would save me time.

Re: [gentoo-user] Reducing repeated checkings

[Paul Hartman]

On Mon, Aug 23, 2010 at 2:27 PM, Al <oss.elmar@googlemail.com> wrote:
> Some thoughts of the day
>
> With multicore machines compilation itself is relativly fast. The
> bigger part of time meanwhile is spend for the system check before
> compilation. Each  program das it again and 90% of the checks check
> the same stuff.
>
> I wounder if efforts have been made to reduce this repeated checks.

confcache

It can (or used to be able to) be used with portage, though I think it
is discouraged because it can cause some other problems. I used it for
a while and never noticed any ill effects, but maybe I wasn't looking
in the right places. I read enough about it being bad that I stopped
using it out of fear.

The better time-saver is probably parallel emerges (emerge -j) which
can configure multiple packages at once. It also has quirks but works
most of the time. :)

Re: [gentoo-user] Reducing repeated checkings

[Al]

2010/8/23 Paul Hartman <paul.hartman+gentoo@gmail.com>:
> On Mon, Aug 23, 2010 at 2:27 PM, Al <oss.elmar@googlemail.com> wrote:
>> Some thoughts of the day
>>
>> With multicore machines compilation itself is relativly fast. The
>> bigger part of time meanwhile is spend for the system check before
>> compilation. Each  program das it again and 90% of the checks check
>> the same stuff.
>>
>> I wounder if efforts have been made to reduce this repeated checks.
>
> confcache
>

configure knows the option `--cache-file=file':
http://sources.redhat.com/autobook/autobook/autobook_14.html#SEC14

Does emerge make use of it? Is there a way to tell emerge to use it,
without writing an overlay to each package?

Al

Re: [gentoo-user] Reducing repeated checkings

[Alex Schuster]

Al writes:

> 2010/8/23 Paul Hartman <paul.hartman+gentoo@gmail.com>:

> > confcache
> 
> configure knows the option `--cache-file=file':
> http://sources.redhat.com/autobook/autobook/autobook_14.html#SEC14
> 
> Does emerge make use of it? Is there a way to tell emerge to use it,
> without writing an overlay to each package?

Emerge did this, when confcache was installed and FEATURES=confcache was 
set. But I guess this only works well with once cache file per package, 
not with a global cache that is being used during all emerges.
The configure script tests for lots of features, each test has a name. But 
there is no global naming scheme, so tests for different features can have 
the same name in different packages. I think this is why it did not work 
so well.

	Wonko

Re: [gentoo-user] Reducing repeated checkings

[Al]

>
> Emerge did this, when confcache was installed and FEATURES=confcache was
> set. But I guess this only works well with once cache file per package,
> not with a global cache that is being used during all emerges.
> The configure script tests for lots of features, each test has a name. But
> there is no global naming scheme, so tests for different features can have
> the same name in different packages. I think this is why it did not work
> so well.
>

Yes, that is probably a bad idea to use the same cache file for all
packages, if they use different naming schemes. The Autobook suggest
that option inside one package with many subpackages.

<quote>The presence of a well primed cache file makes a big
improvement when configuring a complex tree which has `configure'
scripts in each subtree. </quote>

I think I will give it a try. Make my an overlay and test it with the
gcc, which has such a complex tree.

Al