LWN: Comments on "Some 5.15 development statistics"

Some 5.15 development statistics

Wol — Thu, 11 Nov 2021 23:49:22 +0000

I believe Huawei was blamed for a load of "exploit code" discovered in some of their routers. It was then discovered that this code was old, and had been directly copied from Cisco or similar ...

So even there you need to make sure you blame the right people :-)

Cheers,
Wol

Some 5.15 development statistics

Baylink — Thu, 11 Nov 2021 19:56:48 +0000

Hmmm. I don't mean to imply -- oh hell, yes I do -- that there might be anything skeevy about the contributions from Huawei *just because* the US federal government banned their gear from deployment in communications networks, and is buying out that gear from network operators (along with that of ZTE)...

But has anyone done any kernel analysis with a *specific* eye to the blame-map outcome of Huawei-employed developers?

As other articles in this week's LWN note, some of that type of attack code can be exceedingly subtle, and hard to spot even if you *are* looking for subterfuge, let along if you are not.

And no, the fact that I'm rereading _The Bear And The Dragon_ this week has nothing whatever to do with this comment...

Some 5.15 development statistics

linusw — Sun, 07 Nov 2021 16:17:19 +0000

I think it is interesting to note significant contributions from the organization "SoMainline".

This is just another name for "organized hobbyists", some of them also working on things such as PostmarketOS:

https://www.somainline.org/

Some 5.15 development statistics

fpeters — Thu, 04 Nov 2021 07:47:00 +0000

fyi for this kind of analysis there's also https://github.com/erikbern/git-of-theseus that can plot code survival (by year, not tag). Reading its README file now it also points to https://github.com/src-d/hercules that does the same and apparently much more.

5.15 afaics was slow as the merge window opened at the end of summer vacation season

knurd — Wed, 03 Nov 2021 08:13:58 +0000

> Instead, perhaps, the development community is just taking a well-deserved mid-pandemic break.

I had a suspicion that it was simply summer and the summer vacation season that were a big factor in this. As I needed a distraction I wrote a small script to compile some numbers, which seem to confirm my theory: people simply author and commit less patches in August, e.g. right before the merge window for 5.15 opened. For the numbers and two charts that show this see https://twitter.com/kernellogger/status/1455799355919699969 or https://docs.google.com/spreadsheets/d/1Q60lmo8JklcDUN5SS...

The script can be found here: https://gitlab.com/knurd42/kernel-scripts/-/blob/master/s... It likely has a bug or two, but a quick rough check seems to confirm the findings:

[thl@truhe mainline]$ git log --pretty='%as' --date=local v2.6.14.. | grep -c -e '-08-1[0-6]'
17944
[thl@truhe mainline]$ git log --pretty='%as' --date=local v2.6.14.. | grep -c -e '-12-1[0-6]'
21567

Code

corbet — Tue, 02 Nov 2021 19:39:46 +0000

The data comes from the linetags utility, which is in the gitdm repo.

The plotting of the chart was done by a brute-force one-time script using pyplot, which is a wonderful module for these things. At this point, I'm much more likely to bash out a script to make a plot with pyplot than to try to coax a spreadsheet into doing something reasonable.

The whole script can be found over here ...again, not really meant for public consumption and I don't have the time to change that...

Code

smurf — Tue, 02 Nov 2021 19:33:29 +0000

Messiness is not a problem.

I couldn't find the part that creates the code longevity chart (or the table that it is based on), though. Which, incidentally, is the part I asked about :-P

Some 5.15 development statistics

Paf — Tue, 02 Nov 2021 14:58:08 +0000

Two questions.

1. Is 4.4 another GPU register file drop?
2. What did you use to create the graphs from the extracted data? (Apologies if that’s in the code dump)

Code

corbet — Tue, 02 Nov 2021 13:45:15 +0000

It's all in git://git.lwn.net/gitdm.git

I'll warn you that the code is messy, but it does get the job done...

Some 5.15 development statistics

smurf — Tue, 02 Nov 2021 11:11:56 +0000

Is the code (esp. the "Code Longevity" part) available somewhere?