[GHC] #15068: Small number of test case failures on Ubuntu Bionic (GCC 7.3)

[GHC]

#15068: Small number of test case failures on Ubuntu Bionic (GCC 7.3)
-------------------------------------+-------------------------------------
        Reporter:  jrp               |                Owner:  (none)
            Type:  bug               |               Status:  merge
        Priority:  highest           |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.4.1
  (CodeGen)                          |
      Resolution:  fixed             |             Keywords:
Operating System:  Linux             |         Architecture:  x86_64
                                     |  (amd64)
 Type of failure:  Incorrect result  |            Test Case:  T13658
  at runtime                         |  T14779a T14779b T14868 debug
                                     |  parsing001
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):  Phab:D4654
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by hvr):

 Sven, I still believe you're not seeing the point I'm making and trying to
 get back into a line of argument that's besides the point (don't get me
 started on the validity of those "statistics" you quote) and would only
 detract from the technical arguments that are pertinent to this issue.
 Please stop bringing up the irrelevant claims about Stack's (questionable)
 popularity and thus Stack would be too-big-to-fail that we have to adjust
 the surrounding ecosystem to Stack's needs, or even your (questionable)
 subjective perception about what a "happy Haskell life" constitutes,
 unless you really want me to get into an extended off-topic rant about of
 Stack's negative impact on Haskell's ecosystem and community and now even
 reaching into teaching (e.g. the popular data61 fp-course publicly getting
 attacked and its course materials even forked over Stack ideology)...

 But let's try to get back on-topic. Maybe I'm not explaining myself well
 enough to get across the technical point, that rushing a GHC 8.4.3
 quickfix just to workaround bad decisions made in Stack won't be enough
 and that Stack will need to address this on their end or Stack users will
 be still be "broken" with GHC 8.2.1, 8.2.2, 8.4.1 and 8.4.2 based install-
 plans on their end; and that we should rather focus our limited resources
 on letting GHC 8.4.2 mature a bit more (which is essentially a
 quickfixed/proper GHC 8.4.1 which is what some communities would declare a
 "NUKED" release; iow, for all practical purposes GHC 8.4.1 should be
 considered as if it never happened). To summarise, this bug has been
 present in GHC since GHC 8.2.1. If we follow your logic, Stack's mere
 existence would require us to make a GHC 8.2.3 point release as well
 rather than to fix this in Stack, in order not to leave those unfortunate
 GHC 8.2/Stack/Ubuntu18 users out in the rain.

 Moreover, the most principled and thus recommended way to install and
 manage GHC on Debian or Ubuntu if you need a version that's not packaged
 by Debian/Ubuntu is via the instructions mentioned on
 http://downloads.haskell.org/debian/ which provides non-generic bindists
 specifically configured and built for the respective distribution release
 (this may not be so relevant on OSX or Windows, but for Linux's ABI compat
 story it is); this btw also means that a redundant(!) GHC 8.4.3 release
 would cause me a lot of busy work to "fix" something that my packaging
 doesn't even suffer from as I'd need to package up 8.4.3 and build it for
 various non-EOL'ed distros; not to mention I'd have to scrap the rather
 costly/time-consuming builds of a GHC 8.4.2 bindist for AIX I was working
 on, and start over w/ GHC 8.4.3, as well as having to carefully prepare
 Hackage releases of a couple GHC bootlibs and their associated Haddocks.

 Finally, this doesn't affect non-Stack users, unless they are on a just
 released Ubuntu 18 release (NB: there's currently 3 LTS releases of Ubuntu
 that aren't EOL'ed; nobody is forced to use the still very infantly young
 new Ubuntu LTS yet) *and* don't use the packages provided either by Ubuntu
 or by my Ubuntu PPA (or aren't capable of building GHC from source) *and*
 use a non-default flag which is only relevant if you actually intend and
 know how to debug a Haskell program at a low-level via a debugger such as
 GDB. I seriously doubt this is a large enough group of people to demand
 this urgency. I for one I can't use (nor do I recommend) using GHC 8.4.x
 for anything mission critical until it has matured at least 2 more months
 and seen more field testing, and we've had time to accumulate fixes and
 release a non-redundant GHC 8.4.3 with those, to get it to the same level
 of confidence the GHC 8.2.2 release has.

 I've already written more about this that I wanted to; I haven't heard any
 compelling reason yet that would justify rushing a redundant GHC 8.4.3
 release only to keep Stack from fixing its bad choice of defaults. This is
 like asking to move a mountain because you don't feel like going around
 it. I'd rather want to see a GHC 8.4.3 released shortly before its
 support-window closes (which is in a couple months from now when GHC 8.6
 becomes a thing), and releasing a GHC 8.4.3 now would make it more likely
 we won't see a GHC 8.4.4, thereby leaving us with a potentially less-
 reliable GHC 8.4.3 than it would be if it had had more time to mature and
 collect more fixes.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15068#comment:33>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler