Developers: I've got a (great) Portage suggestion

GaMMa · l33t Joined: 23 Aug 2002 Posts: 684 Location: USA

Why don't you add a feature where it tells the estimated compile time at the beginning of an emerge? I'm sure you're thinking it'd be different for every processor, but hear me out. You have the user compile a 100kb test code, and base the compile time results off that. After that to find the results to a 1MB transfer you just do 5X100kb = whatever time. Also if you're doing multiple emerges at a time, you can have the estimated compile time before you compile each source in the Emerging foo (1of20) line. Any feedback? I think it'd be a great idea, especially for the impatient, who wonder how long it'd take to compile all this sutff.

Lovechild · Posted: Fri Jan 03, 2003 8:41 pm Post subject:

DUPE ALERT...

Nice idea, but flawed, because it's the complexity of the code not the size that determines the compile time.

The only way to do this is to painfully meassure, say how long it takes to compile bash and then use that as a base... thus we need to time every ebuild and divide with the bash standard compiletime base (BSCB)... so that we have a multiplier to use... then all we need is to compile bash once to get the BSCB and then the estimated compiletime will be the ebuild multiplier times BSCB.

Problem, hell yes, what if the computer is under load, thus the realworld BSCB is no longer equal to the idea BSCB and the estimate will be worthless...

It can't be done gracefully, sorry.

axxackall · Posted: Fri Jan 03, 2003 8:57 pm Post subject:

There is another way to implement it. Imagine the following:

When you install and configure gentoo-stats, allow it to gather compilation statistics from portage. Some appropriate patch must be developed and applied to both gentoo-stats and portage.

Portage must register (log) compilation time, gentoo-stats must use that log to send it to stats.gentoo.org.

After the compilation time statistics is gathered on stats.gentoo.org then periodically this information is used to update some text comments inside ebuilds.

No need to use any AI where a simple statistics may help.

By the way. the other statistics (i.e. amount of installations generally or on specific CPU) can be useful would it be automatically updated in text comments in ebuild files.

jigma · n00b Joined: 10 May 2002 Posts: 51 Location: NZ

I think the gentoo-stats is a great idea...

It is not like you are after pinpoint accuracy...you just want a ball park figure like 10-15 minutes or 30-40 minutes.
_________________
"linux is only free if your time has no value" - Jamie Zawinski

Dalrain · Posted: Fri Jan 03, 2003 11:31 pm Post subject:

I believe that though this is a good idea, the bandwidth and resources it would require rather outweigh the small return. (A little convenience.) However, what would be rather easy to do (comparatively) would be to label known HUGE compiles as such. (The kinds of things my dual Athlon chokes on for a few hours, for instance.) Perhaps this could be another "attribute" as shown in the search table, making an easy filter for things that take forever across the board. Even ballpark figures for machines like that would be horrible to try and acquire statistics for, in my thinking at least. Resources better spent elsewhere.

echeslack · Posted: Sat Jan 04, 2003 1:27 am Post subject:

I always thought that this type of thing should be implemented at make's level. make already has an option to output all the commands its going to execute. While it wouldn't give you an actual amount of time, it could give you a semi-accurate progress bar by tracking how many commands need to be executed and how many have been. If I ever had enough good ideas to add to make maybe I would start some sort of supermake project, but that's the only thing i have at the moment, and I don't think its worth it, so its not going to happen

. But if somebody did want to do it, I bet it would be pretty straightforward to get a basic system going and just move make and replace it with your supermake (make may be a link to the actual program, which would be even better b/c you could just redirect the link to your version, but I can't check that theory as I'm on windows right now).

-ewen

pilla · Posted: Sat Jan 04, 2003 3:05 am Post subject:

Very nice idea -- but not new. Sorry guy, moving to Dups.
_________________
"I'm just very selective about the reality I choose to accept." -- Calvin