1 |
On Thu, Feb 11, 2010 at 04:54:38PM +0100, Torsten Veller wrote: |
2 |
> Can we please move the mips profiles from "dev" to "exp" in |
3 |
> profiles/profiles.desc? |
4 |
> |
5 |
> |
6 |
> The ~150 mips development profiles increase the time for a |
7 |
> `repoman -d full` run in dev-perl/ from three to five minutes. That is |
8 |
> an increase of roughly 66 percent. |
9 |
> repoman further prints more than 2000 lines of output for two keywording |
10 |
> problems. |
11 |
|
12 |
Quick pcheck visibility scan of the full tree, stats follow: |
13 |
|
14 |
mips profiles still enabled: |
15 |
* 116191 seperate dependency issues, 1 line per profile/dependency |
16 |
issue |
17 |
* roughly 2m39s run time |
18 |
|
19 |
mips profiles disabled (leaving mips-irix however) |
20 |
* 9550 seperate dependency issues, 1 line per profile/dependency issue |
21 |
* roughly 1m54s run time. |
22 |
|
23 |
So... mips accounts for about 30% of the pcheck runtime, and *92%* of |
24 |
known visibility issues. As for the runtime difference between |
25 |
pcheck/repoman, pcheck has some tricks internally to reduce the # of |
26 |
profiles it has to scan down to just the unique USE/mask set- I'd |
27 |
expect the mips impact to be far larger w/out that trick in place. |
28 |
|
29 |
At the very least if it's going to be kept around, experimental or |
30 |
not, the number of profiles in use there *really* needs reduction- |
31 |
mips has roughly 117 profiles listed in profiles.desc out of 217- |
32 |
literally ~54% of all dev/stable/experimental profiles. |
33 |
|
34 |
Either way, stats to chew on. |
35 |
~harring |