This should now be fixed. Apologies for the wait -- you got trapped by the Mailgun suppression filter, because apparently some emails bounced off your user account @acm.org -- I imagine the SMTP gateway might have been down at some point and you probably don't need many CC's in that time window to get blacklisted.
Feb 22 2017
Dec 17 2016
Jul 11 2016
Nice, thank you! This looks fine.
Wow, nice catch. I think we should definitely put the milestone on Trac #12227 for 8.0.2 for this one.
Jul 8 2016
Jul 4 2016
Jun 21 2016
Jun 4 2016
Have you tested this on a newer Arch system, like mentioned in Trac #12147?
Jun 3 2016
Jun 1 2016
Okay, @rwbarton with the save as usual. But numeric/should_run/T10011.hs didn't catch this, which is clearly my fault somewhere, so we should look at that, too.
Maybe there should just be a consolidated when_ppc64 conditional. But this otherwise looks fine to me since it just marks the Linker.c tests specifically; I'd wait on @hvr for any AIX followups.
Big win - but please put a changelog note about the constraint removal for Ratio - it can cause people warnings with the new redundant constraint machinery when people upgrade, so probably worth pointing out.
May 30 2016
LGTM, and I think @thomie is right as well.
May 29 2016
why do we have another md5.h? I recently saw one in ./libraries/base/include/md5.h
I also use binary for -fexternal-interpreter, and I really want a faster replacement. I was going to look at Store, but if cbor is not far away maybe I should wait for that. Is it likely to be as fast as Store?
May 27 2016
Do we think that incorporating binary-serialise-cbor into GHC is a viable option? Won't this significantly increase our dependency set? AFAICT it would add text, unordered-containers, hashable, if taken as-is. Are these dependencies just for defining instances? If so, what is going to become of them?
love to wordsmith
But can we switch to CBOR yet?
It's not released yet, and @duncan and I are still tuning it. I mean, it's definitely seen a lot of production use, we're just not committed to the initial API (yet).
I expect this patch will probably still give some additional benefit, even when combined with CBOR (a lot of the redundancy will be stripped, e.g. because Ints will be encoded much more densely), so I imagine the overall effect will be somewhere in the 50-70% range over what we're currently doing with the two of them combined. Just speculation.
I have a private branch with a tiny amount of work starting a transition of GHC to use binary-serialise-cbor, but it doesn't work.
Should there be a test for this? Looks OK either way.
May 26 2016
May 24 2016
Great job, thanks. We should honestly probably merge nofib into ghc.git at some point tbh...
Okay, so I don't want this to hang out forever (Hadrian will have to accommodate it but it'll eventually have to do that), and this change overall looks OK to me. @ezyang feel free to merge this (and point Cabal to an even newer commit if you think it's safe too).
Yeah, I'm always vaguely scared of such changes because encoding is so complex, but I'll buy it since it's mostly fixing the test suite.
May 23 2016
LGTM. I almost feel like you should make it even more robust: change the #if parts from:
if it builds, ship it
May 21 2016
May 20 2016
LGTM, I like code reduction.
May 19 2016
- Remove some more dead code
- Make encoding simpler
- docs/relnotes: add blurb
Minor wibble, but not a big deal. LGTM.
May 17 2016
May 14 2016
May 13 2016
Should this have a release note? I'll let you decide.
May 12 2016
Nice, look at that reduced allocation footprint!
May 10 2016
Great, thanks for finishing this off! Can you just add a simple test? (Or maybe even modify/consolidate with the test from D2012?)
Planning changes based on Ben's feedback.
May 9 2016
I suppose the trouble here is that this would change the current meaning of some doc comments (since "the name of the foo" would currently be applied to the Foo tycon.
Indeed. I'm loathe to make a breaking change to the way Haddock works here, since that would affect the way comments are rendered for a good amount of data constructors in the wild. And in any case, this seems like an orthogonal concern - perhaps we should open a separate Trac issue for this?
May 8 2016
May 4 2016
Apr 30 2016
Apr 29 2016
@bgamari Should I merge this into master before it needs rebasing?
I too would prefer this - and nice catch on the config simplication!
Apr 28 2016
The other mystery is why the hell mingw doesn't come with a CC alias, but that's for another day.
Apr 27 2016
Apr 25 2016
Apr 18 2016
Apr 13 2016
[flashbacks of writing NTFS code intensifies]
Looks fine, but why not just change the local decl to an unsigned char? Maybe this isn't better. I don't know.
Apr 6 2016
Looks good. Generally I'd ask for a comment here, but for hackers the W/W transform here should be fairly obvious, I guess (independent of the upstream issue @simonpj mentioned, Trac #1168, which I think you're right, we're best not to confuse too much).
I think this is fine (and consistent with normal signatures), personally. But it needs a release note entry. (Yet this isn't going to go into 8.0, so you'll have to wait on D2066 or something.)
Apr 5 2016
I was the first person to implement support for -N on OS X (c.f. 97c4b27494addf) in some of my earliest patches to GHC. Good memories!
Mar 30 2016
This looks fine, but yeah, one more test would be nice.