Bug 431614 – [science overlay] intel-sdp.eclass: please extract all RPMs in one call to tar

• Product: Gentoo Linux
• Status: CONFIRMED ---

[Description Michał Górny 2012-08-16 09:46:30 UTC]

Right now, intel-sdp_src_unpack() does a 'for' loop over all requested RPMs and extracts them one by one from the huge tarball.

I believe it would take much less time if all the possible paths were just passed to 'tar -xf' and thus were extracted in a single run, and then the necessary logic was applied.

[Comment 1 Sébastien Fabbro (RETIRED) 2012-11-22 15:28:42 UTC]

may be.

[Comment 2 Justin Lecher (RETIRED) 2012-11-22 15:50:53 UTC]

I will look into it.

[Comment 3 Justin Lecher (RETIRED) 2012-11-22 19:01:10 UTC]

is it worth saving 25% of time and using 650% of disk space?

[Comment 4 Michał Górny 2012-11-22 19:14:10 UTC]

(In reply to comment #3)
> is it worth saving 25% of time and using 650% of disk space?

Are those actual numbers? Could you elaborate?

[Comment 5 Justin Lecher (RETIRED) 2012-11-22 19:24:29 UTC]

Wrong testing. Doing it again. Will be much more.

[Comment 6 Justin Lecher (RETIRED) 2012-11-22 19:49:31 UTC]

okay I found a solution. 

Full extract    30 s    2100 MB
Old solution    62 s      81 MB
New solution    37 s      81 MB

This should be fine.

[Comment 7 Michał Górny 2012-11-22 19:54:55 UTC]

(In reply to comment #6)
> okay I found a solution. 
> 
> Full extract    30 s    2100 MB
> Old solution    62 s      81 MB
> New solution    37 s      81 MB
> 
> This should be fine.

Yes, it seems that you found the right way. Care to attach a diff or the new version?

[Comment 8 Justin Lecher (RETIRED) 2012-11-22 21:19:29 UTC]

I will send it off to the ml soon. The idea is simple, get a list with "t" and use that list to identify the location of the rpm inside the tarball. Then decompress alls at once.

[Comment 9 Michał Górny 2012-11-22 21:41:37 UTC]

(In reply to comment #8)
> I will send it off to the ml soon. The idea is simple, get a list with "t"
> and use that list to identify the location of the rpm inside the tarball.
> Then decompress alls at once.

Hmm, not sure about that but couldn't we just create a 'max' list of possible rpm locations and pass them all to tar? This would cause an error exit but we could probably just ignore it and verify the RPMs instead.

[Comment 10 Justin Lecher (RETIRED) 2012-11-22 21:59:58 UTC]

Maybe an evan faster solution.