<winny>
samth: before i post to the discourse thread about the snapshot thing, would there be a more appropriate place to leave feedback why there may not be much inclination from packagers to invest the time and effort
<winny>
i think it's important to point out how special unicorn racket is in terms of testing non-release versions
<samth>
winny: do you mean what you were saying about not wanting to package snapshots regularly?
<samth>
or something else?
<winny>
yes
<samth>
and in what sense is racket a special unicorn?
<winny>
using a tarball as a snapshot in time instead of a git rev
<samth>
because no one is suggesting that people package snapshots all the time
<winny>
i think i misunderstand, if packagers aren't always testing there'll be issues like the autoconf bugs that kept cropping up last year
<winny>
or the year before idr
<samth>
the point is that this is a big change in the build process and likely to impact packaging, so we want people to try it out before the 8.6 release so we can get those bugs fixed
<samth>
obviously there's always the possibility of bugs that impact packaging but that's usually much less likely than right now
<winny>
i'll not post but i find this entire prospect kind of ridiculous
<samth>
what do you mean?
<samth>
i feel like somehow you and i are not communicating effectively but I'm not sure what the disconnect is
ec_ has quit [Ping timeout: 240 seconds]
<samth>
also the idea that there's a non-trivial step between the source repository and the tarball that you'd download seems very normal -- many open source projects work that way
<samth>
but like i said clearly i'm not understand what you're saying so I would like to
<samth>
maybe a longer post would clarify things
szkl has joined #racket
badkins has joined #racket
<winny>
I think I should write something long form and gather some data to help convey what I'm trying to share
<winny>
all i know is the packaging story for racket is pretty unfortunate
<samth>
winny: that would be helpful, thanks!
<winny>
np, thx for being patient with this discussion
ec_ has joined #racket
littlebobeep has quit [Ping timeout: 240 seconds]
msiism has joined #racket
ec_ has quit [Ping timeout: 240 seconds]
<bremner>
samth: fwiw, I think fewer projects have a big disconnect between version-control and distribution these days. I don't really have data for that, but racket is the only one I maintain in Debian (not counting git submodules, which is a different kind of pain)
<samth>
yes, i agree and that this is a somewhat generational difference -- when Racket was started this was the normal way projects worked
<samth>
it's not clear to me if the best way forward is just distributing tarballs of the git repository, or some other change
<samth>
right now, i think the main disconnect is that the top-level directory doesn't work in the "configure && make && make install" way, but I might be wrong about that
kenran has joined #racket
ec_ has joined #racket
ttree has joined #racket
<samph>
One thing that occurs to me that makes a tarball attractive is that there are several component packages on top of "racket-minimal", and the tarball defines a supported configuration of those parts.
<samth>
the other big reason is that `make` accesses the network whereas the tarball for non-minimal racket includes everything
jao has quit [Ping timeout: 240 seconds]
<winny>
compare that behavior to gradle
<samth>
i can't even tell how to build gradle from source from their github
<samth>
so i don't know what to compare
<winny>
gradle gets a little bit of a bad rap because it pulls dependencies from network, it's not designed to pull dependencies from another package or bundle them for distribution, afaict. Debian appears to kludge it with using a local maven repository and patching the gradle build files to look there instead. Other distros don't bother such as gentoo, their official stance is it's not going to be
<winny>
This feels a bit reminiscent of building racket from git with all the standard collects installed
<samth>
winny: that seems a little different since it's about using gradle to build things rather than packaging gradle directly
<samth>
but I guess I still don't understand what the comparison is
<winny>
In both cases a build system is designed primarily to pull deps from the internet
<samth>
this is a big, general challenge for linux package managers (cf cargo, npm, etc). But I don't think anyone is worrying about packaging individual racket packages, right?
<samth>
we're just concerned about packaging either minimal or main-distribution racket
<winny>
xgt (xgqt?) is doing that in his racket overlay on gentoo, it's pretty impressive. Same goes for the nix folks with their racket2nix
<winny>
though i feel like that's not a concern agreed
badkins has quit [Remote host closed the connection]
<samth>
i agree that that's impressive, and i would want to help with that if they need our help, but yeah I think that's not the thing I'm generally focused on when thinking about packaging
<xgqt>
samth, winny: when Racket pkgs can not be built w/o network I sure will bring it up to Racket team :) Right now it works. I found a way to relieably build & install individual non-main-distribution pkgs by the system pkg manager (on Gentoo, that is Portage).
<winny>
Oh hey! Yep I think I crossed my wires a little - was highlighting the behavior of gradle to build something versus building racket from git with all the normal bits-and-bobs pulled in (from the network unfortunately)
<xgqt>
its good that Racket team makes periodical snapshot and releases dont pull main-dist pkgs from net, very helpful
<xgqt>
if it didnt we would have to list all main-dist in Racket sources
<winny>
xgqt: since you're here, do you invest time using the snapshots regularly to test pre releases?
badkins has joined #racket
<xgqt>
I was doing that, after 8.5 I stopped. That was a bad idea.
<xgqt>
I brought the zuo issue to mflatt when I tedted post-8.5 snapshot
<xgqt>
I havent tested the pkg, but I build Racket from git manually and it worked with installing to custom image, so I believe 8.6 should be fine
<xgqt>
But i WILL test post-8.5 snapshots for pkgs some time :)
<samth>
xgqt: why was it a bad idea?
<samth>
(or was stopping the bad idea?)
<winny>
Do you think it would save some time & effort to just have a live (git) ebuild that somehow handles a full racket installation (assuming it ships collects in a fast-to-build way)? I know it's not quite possible in the current iteration, but let's imagine for a moment
<winny>
or same effort?
<xgqt>
samth: I meant that it was a bad idea to stop testing snapshots :^)