One less item on the V3INTEGRATION list.

3 patches for repository http://darcs.net/screened:

patch 0a26c618d2c049a8f5039a858486581f1d891218
Author: Ben Franksen <ben.franksen@online.de>
Date:   Wed Aug 21 10:39:17 CEST 2019
  * Darcs.Util.Graph: make helper functions local to ltmis

patch 8168201f5f78ff2d72c23c90450c5030afcbdf38
Author: Ben Franksen <ben.franksen@online.de>
Date:   Wed Aug 21 10:40:48 CEST 2019
  * Darcs.Util.Graph: add hadocks

patch cd1c64911ce98f347cc2d6381025664236471042
Author: Ben Franksen <ben.franksen@online.de>
Date:   Wed Aug 21 12:41:32 CEST 2019
  * Darcs.Util.Graph: add properties and test them

Attachments

• darcs_util_graph_-make-helper-functions-local-to-ltmis.dpatch
• patch-preview.txt
• unnamed

Looks good

>  * Darcs.Util.Graph: add properties and test them

Just a note for future reference. After applying this patch, if I do an
incremental build I get a segfault from some darcs commands, like 
obliterate, only on darcs-3 repositories (!).

I recently had problems with my msys installation and with some other 
software on my machine causing standard binaries like openssh to segfault,
but this feels a bit different.

Anyway, if I blow away my build tree and rebuild all is ok, so I'm just
going to move on. But in case we see anything similar in future I wanted
to leave a note.

> Just a note for future reference. After applying this patch, if I do an
> incremental build I get a segfault from some darcs commands, like 
> obliterate, only on darcs-3 repositories (!).

Can you try again by obliterating, building, applying, and rebuilding? I
guess you won't get the error then.

I remember that I also had strange segfaults with experimental darcs
versions that had worked fine in the past. I got the segfaults when
working on darcs itself, which is in V2 format, so this is not limited
to V3 repos. Same as for you, rebuilding the darcs in question fixed it.

I recently switched compiler versions a few times, sometimes building
with ghc-8.2.2, sometimes with ghc-8.6.3 and now with ghc-8.6.5, and I
think this is what caused the problems for me. Perhaps related to shared
libraries not being found in the correct places?

>> Just a note for future reference. After applying this patch, if I do an
>> incremental build I get a segfault from some darcs commands, like 
>> obliterate, only on darcs-3 repositories (!).
> 
> Can you try again by obliterating, building, applying, and rebuilding? I
> guess you won't get the error then.

I tried that repeatedly before I established what was going on. The
sequence that failed was clean build without the patch; apply patch;
incremental build. The failure persisted even after I obliterated until
I did a clean build. (I only actually observed the failure as a segfault
once, when I dug deeper into the test failures, but the symptoms I did
observe each time were consistent.)

> I remember that I also had strange segfaults with experimental darcs
> versions that had worked fine in the past. I got the segfaults when
> working on darcs itself, which is in V2 format, so this is not limited
> to V3 repos. Same as for you, rebuilding the darcs in question fixed it.

In this case it was definitely V3-specific, several tests were passing
on V1 and V2 and failing on V3.

> I recently switched compiler versions a few times, sometimes building
> with ghc-8.2.2, sometimes with ghc-8.6.3 and now with ghc-8.6.5, and I
> think this is what caused the problems for me. Perhaps related to shared
> libraries not being found in the correct places?

I was definitely using a consistent compiler version (same shell and
same cabal.project.local).

Okay. This looks like a compiler bug to me. Which ghc version are you
using and which repo state(s) exactly? I'll try to reproduce. If I can,
we should file an issue with ghc.

I'm getting it again with patch1891 :-(

I discovered that --disable-optimization is part of the circumstances, which 
may explain why we haven't seen it before (I only recently started using 
it). GHC 8.2.2, haven't tried with a later GHC yet. My suspicion is that 
it's cabal file dependency changes that trigger it, but I haven't confirmed 
it. I'll start a thread on darcs-devel as it's not specific to one patch.