[GHC DevOps Group] GHC 8.4.3 release, and release policies?

[Gershom B]

Update: as per https://github.com/commercialhaskell/stack/issues/4019 it
turns out that stripping is in fact enabled by default in stack, so this
bug only hits when people _turn off_ stripping in their config files (which
in stack just happens to also be coupled with turning on -g).

So a fresh installed stack works on ubuntu, and only breaks when people
change stripping settings. Anyway, this seems to lend more weight to the
“no release necessary” side of things.

—Gershom


On May 13, 2018 at 4:11:06 AM, Gershom B (gershomb at gmail.com) wrote:

On May 13, 2018 at 3:59:04 AM, Simon Marlow (marlowsd at gmail.com) wrote:

Will Stack also be broken for other GHC versions on Ubuntu 18.04, or does
it only enable -g for 8.4+? i.e. even if we roll a new 8.4.3, will people
still using 8.2.x have problems?

Indeed, as best I undesrtand it, it will be broken on the 8.2 series as
well. So anyone pinned to any older ghc (via a particular lts or daily)
will continue to have this problem. From what I can tell, the -g flag is
added basically universally. C.f.
https://github.com/commercialhaskell/stack/blob/master/src/Stack/Build/Source.hs#L147


The "purist" side of me agrees with you that this is a downstream issue.
However, Stack doesn't have its own builds of GHC unlike Ubuntu, so it
can't easily fix this by patching GHC. They could fix Stack to not use -g
on affected distros / GHC versions though.

As far as I know, stack does produce its own builds these days, although
they should be conferred with, to confirm. In particular, see
https://raw.githubusercontent.com/fpco/stackage-content/master/stack/stack-setup-2.yaml
which
is where it draws the compilers it installs from. Currently all the urls
seem to point to files hosted on
*https://github.com/commercialhaskell/ghc/releases/
<https://github.com/commercialhaskell/ghc/releases/>* and from what I can
tell there are far more ghc builds provided there than directly produced by
ghchq.

—gershom



Cheers
Simon


On 12 May 2018 at 18:23, Gershom B <gershomb at gmail.com> wrote:

> I believe this list/group is intended to help set GHC release policies
> and plan when releases occur.
>
> I'd like to draw people's attention to
> https://ghc.haskell.org/trac/ghc/ticket/15068
>
> I think that the "nay" side has the right of it.
>
> 1) To summarize my understanding: there is an issue with upstream gnu
> binutils that are included in Ubuntu 18.04 and is triggered by GHC
> _only_ when the -g flag for debug symbol output is passed. This bug is
> triggered by GHCs going back through the 8.2 series. Note that 18.04
> was released _after_ ghc 8.4.2.
>
> There is a patch to GHC to work around this issue, that also
> incidentally improves another possible issue with debug symbol
> emitting.
>
> Normally the -g flag is relatively rarely used, and by those who would
> be able to work around such things. However, stack runs with -g by
> default, unless users explicitly compile with symbol stripping turned
> on.
>
> The proposed solution is an 8.4.3 release of GHC with _just_ this patch.
>
> 2) To summarize my objection to the plan -- I do not think that
> spinning up a full new release to just patch one thing to work around
> an upstream bug is a good precedent for GHC release management. I can
> imagine exceptional circumstances, but in general the right answer
> should happen downstream of a full GHC release. We already want a more
> timely cadence. That cadence should suffice for these sorts of things.
>
> In the meantime, the binary provided by hvr's ubuntu ppa already
> incorporates this patch.
>
> One might say -- what about stack and platform users? My suggestion
> would be that since stack provides its own binaries, patched binaries
> can be provided by stack in this circumstance, just as the ubuntu ppa
> includes the patch. Similarly we could respin a platform release with
> the patch, just for linux, rather than across all three major
> platforms. (Or not -- I don't think that users that know about -g
> would mind too much using the ppa for the patched release, and I
> suspect those users are more likely ppa users already anyway, or wise
> enough not to immediately upgrade to a new ubuntu system for devwork
> without time for bugs to shake out).
>
> Anyway, I can't recall a precedent for a point-release over an issue
> like this in the past, and I think if we start to do so too much it
> will overstrain resources away from the important goals of regularity
> and stability we do want in releases.
>
> I would like to know what others think.
>
> Cheers,
> Gershom
> _______________________________________________
> Ghc-devops-group mailing list
> Ghc-devops-group at haskell.org
> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devops-group
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devops-group/attachments/20180513/f4c50033/attachment.html>