#pest | 2022-05-09

shinohai: hm, and what dependency hell is tcl going to drag in?

00:11

*

signpost does want blatta. might be worth it for me to help replace blatta's storage.

00:12

looks like maybe tcl just wants bdb, which will already be present for bitcoind.

00:12

signpost: Best I can tell mostly C code, a small bit of Ada in there too: https://github.com/tcltk/tcl

00:14

I don't want to be so religious we don't do anything, just mindful of how much stuff gets dragged in.

00:15

this exercise at least tells us how many things we actually depend upon.

00:15

That's what I was thinking, at least gives one thorough education in *precisely* what exists on system.

00:16

what was the patch you needed for python?

00:16

I've got sbcl working over here, and the sbcl test suite pointed out to me that I hadn't created /etc/group yet.

00:16

once I did, all sbcl tests passed without patching. curious if python was also mad about that.

00:18

(I've meanwhile added a minimal /etc/group to /src/build/root)

00:19

probably about time I cut a genesis for pbuild and the build scripts. I'll get to that next week, and then you'll be unblocked to patch.

00:21

hang tite will paste python patch

00:21

http://paste.deedbot.org/?id=iim6

00:23

there's the problem with python. METH

00:23

Now there may be more things that surface once I try building w/ sqlite support - I'm prolly not gonna compile w/ ssl and all that stuff unless someone knows reason I should.

00:23

lol @ METH

00:23

interesting, saving https://bugs.python.org/issue20594 in my notes, and here for posterity.

00:24

shinohai: I did a build of python last week sometime and also skipped ssl, though we should explicitly kill it off in the configure step so it doesn't see bitcoind's old openssl and decide to enable

00:25

*

signpost would much rather have zero ssl than w/e old-ass openssl that happened to work in trb.

00:26

I imagine openssl eventually gonna rear ugly head though in something :/

00:27

iirc the tls implementation in busybox doesn't verify shit.

00:27

and is limited in other ways.

00:28

but I'd recommend letting it bump into the need first.

00:33

shinohai: hm, ugh, now I wonder how much openssh depends on openssl.

00:34

That was one of the next things I wanted to try after I got past blatta.

00:34

well anyway, I don't want to apply such conservatism that nothing at all can be done. pressure can be applied to at least minimize optional interdependency.

00:35

cool

00:36

I'm addicted to it though already, everything just makes sense, is readable.

00:36

And protable!

00:36

ugh portable

00:42

glad to hear it. put a lot of sweat into it, and intend to use it for a few embedded projects soon. (home defense stuff with perimeter sensors and cameras, without obscene cloud shit)

00:43

will very much appreciate your and others help making pentacle more useful, and keeping the thing tidy/small.

00:44

have you tried building it on a non-x86 machine yet?

00:49

nah, had the thing fart out a ppc-targeting gcc the other day, seemed to work, but haven't built on anything else yet.

00:49

would be a good experiment on anything (arm, etc)

00:50

my intent is to make targeting another arch brain-dead simple on any pentacle instance; you'll see me passing BUILD/HOST/TARGET through pbuild to that end.

00:52

oughta be able to do say HOST=ppc-linux-musl /src/pbuild install busybox -d /ppc and get oneself a busybox binary they can run on ppc.

00:52

see also: crossdev on gentoo

00:53

That would be neato (similar to doing crosdev currently)

00:53

oh lol

00:53

yup yup

00:55

Hmmmm tcl build fails with tclEpollNotfy.c:28:23: fatal error: sys/queue.h: No such file or directory

00:58

awt

I am shocked tcl is required for sqlite

00:59

tclsh is what it bitches about not having, could find nothing yet on method to build without it.

01:01

awt

It seems as though sqlite was initially developed as an extension for tcl.

01:01

Could have sworn I built by hand years ago w/ only awk scripts and no tcl, but guess I dreamed it.

01:06

http://logs.bitdash.io/pest/2022-05-09#1005299 << is used there for the ecdsaism. asciilifeform has notions of replacing w/ ffaistic routines.

01:06

Logged on 2022-05-09 00:25:16 signpost[asciilifeform]: would much rather have zero ssl than w/e old-ass openssl that happened to work in trb.

01:07

http://logs.bitdash.io/pest/2022-05-09#1005311 << asciilifeform plans cameras around new pad, quite interested in what signpost digs up re: usable irons to plant the thing on

01:07

Logged on 2022-05-09 00:42:56 signpost[asciilifeform]: glad to hear it. put a lot of sweat into it, and intend to use it for a few embedded projects soon. (home defense stuff with perimeter sensors and cameras, without obscene cloud shit)

01:08

*

asciilifeform currently has ones which simply config'd to upload to local ftp on motion

01:08

they annoyingly crash tho, coupla times / yr

01:09

"USE CODE YOUTUBE FOR 20% SIMPLY SAFE!!!!"

01:09

God I hate those fuckers

01:09

lol

01:10

So appears musl doesn't have queue.h - alpine solves by introducing a bsd-compat-headers package.

01:11

hrm

~ 24 minutes ~

01:35

asciilifeform: will let ya know if I find/build anything worthwhile in re: cameras.

01:37

shinohai: sounds about right. elfutils (which contains libelf) wanted a bunch of other non-posix glibc funcs, which brought in argp-standalone, musl-fts, musl-obstack

01:38

the spittoon effect.

01:39

in the specific case of elfutils, only libelf is actually needed to build a recent kernel. perhaps some sawing would knock loose a few of those other deps too.

01:39

*

signpost bbl

~ 39 minutes ~

02:19

http://logs.bitdash.io/pest/2022-05-09#1005323 << yep, per sqlite website, "The build process makes extensive use of the Tcl scripting language"

02:19

Logged on 2022-05-09 00:58:31 awt[asciilifeform|billymg]: I am shocked tcl is required for sqlite

02:20

imho the correct thing is to capture the dependency as-is, taking care to not introduce *unnecessary* cross-dependency by disabling optional features, and once captured as vpatches, lop off complexity as feasible.

02:23

perhaps one path would be standardizing an on-disk storage format with jonsykkel, such that both implementations eat/shit same format.

~ 10 hours 9 minutes ~

12:33

Morning #pest

12:35

http://logs.bitdash.io/pest/2022-05-09#1005345 << plan to tackle tcl today after cafecito, since the bsd-headers aren't that girthy just gonna add 'em and link for now.

12:35

Logged on 2022-05-09 02:20:21 signpost[asciilifeform]: imho the correct thing is to capture the dependency as-is, taking care to not introduce *unnecessary* cross-dependency by disabling optional features, and once captured as vpatches, lop off complexity as feasible.

~ 1 hours 16 minutes ~

13:51

Adding the bsd headers got me further, now tcl fails with `find: unrecognized: -empty` and ` [Makefile:784: libtcl9.0a4.zip] Error 1`

13:51

(so gotta dig into that now)

~ 1 hours 33 minutes ~

15:24

Finally build success using tcl8.6.11 & sqlite 33.8

15:25

Avoid the Shithub repos at all costs.

~ 16 minutes ~

15:41

shinohai: https://wiki.musl-libc.org/faq.html << mentions that queue.h is self-contained.

15:41

if so would be best to isolate that as it's own pbuild rather than whole wad of bsd's headers.

15:42

but nice!

15:43

lol yeah I cheated and used queue.h by itself, ugly hack but it worx

15:43

ah ok, just copied it in?

15:43

yup

15:44

yeah, I typically do the same thing, just get something to build with own hands first, then carve off into tidy vpatches

15:44

*

signpost bbl after pouring a coffee into his head as well

15:59

annnnnd blatta now runs after recompiling python2.7 w/ sqlite

~ 36 minutes ~

16:35

awesome.

16:36

*

signpost gets to work on cutting the pentacle_genesis.vpatch, 'bout time

16:43

*

shinohai expecting weechat to be a whole 'nother nightmare if using it for client side of blatta ....

16:49

so long as the minimal set of software remains small I don't see a problem with pulling in what we actually use.

16:50

rather than pretending to ideological purity while you know, installing gnome on ubuntu on the dl.

16:50

this way we actually get a handle on what makes our computers useful, and using a tool with which we can crank it smaller.

16:50

same problem as trb, just more shitwad.

16:51

I expect we'll see a PHP in this thing, mysql, etc.

16:51

problem becomes when some dork decides he'll e.g. rewrite pbuild in PHP, for reasons, and everything under the sun ends up hauled in for everybody.

16:52

*arises when

16:52

"tangled hierarchy"

16:53

which is not to say hierarchy isn't already tangled. I don't know what business an ADA compiler has being written in ADA, or c compiler in c.

16:53

*Ada

16:55

awt

shinohai: nice!

16:57

I like pbuild exactly because simple and won't stop working when python "upgrades" in 6 months.

16:59

10k man-years of "just want to" in all directions, you know? "wouldn't I be clever if gnat were written in... itself!"

17:00

fact is it's not fucking written in itself; it's written in the previous version, which is written in the previous version, which is...

17:00

the dependency chain of such an artifact is gigantic.

~ 1 hours 25 minutes ~

18:25

billymg

signpost: i'm building pentacle now. when i first tried i got the error "couldn't work out current architecture". i added '--arch=amd64' right after line 60 in bootstrap-debian.sh and now it seems to be working

18:26

interesting. what is uname -m on your host machine?

18:26

and what's the host distro?

18:28

billymg

`uname -m`: x86_64. glibc gentoo host

18:30