[mkroot] To the people who keeping pinging me off list...

[Rob Landley]

On 03/12/2018 02:43 PM, David Seikel wrote:
> On Mon, 12 Mar 2018 13:11:52 -0500 Rob Landley <rob at landley.net> wrote:
> 
>> FYI, I disabled all the subscriptions to the old aboriginal linux
>> mailing list when I moved 'em over (and made new subscriptions
>> require approval), so if you send a message there it goes into
>> moderation.
> 
> My current "mkroot" email folder is still called "FWL".  I've been
> around here for a while.  lol

Ok, yes, that was a while ago. :)

(And like that project, I may still rename this one. I should decide that this
week...)

> I started building LFS 8.1 on Sunday, and later today (Tuesday my time)
> I'll continue.  I'm starting with the "build it manually yourself"
> method suggested in the automated LFS page, and scripting it myself as
> I go along.  I'm up to the beginning of chapter 6, I'll likely start on
> bits of BLFS after finishing that.  Chapter 6 should go quickly, half
> of it is building the stuff already built with some minor variations.
> 
> Today I downloaded the 8.2 stuff, I'll switch over to that next, I was
> trying 8.1 first coz I had downloaded it's source some time ago.
> Bandwidth in Australia is expensive, I have to carefully ration out
> things like downloading 7 GB of LFS + BLFS source code, and 8.2 was
> only recently released.

Fun trick, git ls-remote and then fetch to a tag:

git ls-remote git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux
git fetch git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux v4.16-rc5

This not only lets you grab in smaller chunks (possibly spreading the bandwidth
over multiple billing periods), but it also checkpoints the "download stopped
halfway and it threw out the partial results so you need to restart" problem.
(It'll still do it, but you only regress to the last checkpoint.) It'd be nice
if there was a way to log remotely with better granularity, but the git
developers have been asked for that repeatedly for a decade now and still can't
undrstand why anyone would want to.

(In general git has some of the same problems as android does: it's made by
privileged people for whom certain resources are basically free. Can't log the
repo without fetching gigabytes of data? Why would you ever _not_ spend
gigabytes of data and disk space to produce 200k of text? Doesn't _everyone_
have a multi-terrabyte ssd they can afford to replace annually hooked up to
unmetered fiber?)

> Then I'll move to using Aboriginal Linux for the job, which I have
> experience with, or skip straight to mkroot which I haven't tried yet.

As with aboriginal, if you have the downloads directory already populated with
the right tarballs mkroot won't re-download them. (Or if you have a git repo
checked out in there.)

This time around I didn't implement the fallback mirrors, patches directory, or
package cache. But it's got the basics. :)

> Currently building it on my x86_64 test box natively, I will at some
> point tell it to build for 486.  Then I can test things on my real 486
> hardware, which my old Aboriginal Linux build still works fine on.
> Earlier this year the 486 device was being audited by the government
> test labs, as it's the sort of device that requires that sort of thing.
> 
> I did have the idea while working on this, that the various test suites
> that LFS builds for all the software that toybox replaces, might be a
> good idea to see if we can get them to test toybox.  Dunno if you have
> tried that yet.

I have not but it sounds like fun.

Rob