Ben Lippmeier:
Manuel M T Chakravarty wrote:
Platform-specific breakage is probably going to stay a nuisance -
at least, I have no good idea how to avoid it. If we could somehow
get an infrastructure where we could at least validate patches
easily on a range of platforms, so that the perso
On 21/08/2009, at 10:45 PM, Simon Marlow wrote:
2) Everyone pulls from the blessed ghc-head, but people only push to
their own ghc-head-username repos.
Yes, thought the per-developer overhead is quite high. Probably ok
while there are only a few developers.
Yeah, the people I talked to
On 19/08/2009, at 8:55 PM, Simon Marlow wrote:
Ok, I suggest doing this:
* change #undef DEBUG_DUMP to #define DEBUG_DUMP in
GHC/IO/Encoding/IConv.hs
GHC/IO/Handle/Internals.hs
* cd libraries/base; make
then try hello world and see what output we get.
You can also try setting an encodin
On 21/08/2009 07:14, Ben Lippmeier wrote:
Manuel M T Chakravarty wrote:
Platform-specific breakage is probably going to stay a nuisance - at
least, I have no good idea how to avoid it. If we could somehow get an
infrastructure where we could at least validate patches easily on a
range of platfor
Manuel M T Chakravarty wrote:
Platform-specific breakage is probably going to stay a nuisance - at
least, I have no good idea how to avoid it. If we could somehow get
an infrastructure where we could at least validate patches easily on a
range of platforms, so that the person who creates the b
On 20/08/2009 04:01, Manuel M T Chakravarty wrote:
I fully understand your frustration. Many times when I sat down in the
morning to work on something specific, I found that the MacOS build
broke and spent my time picking up somebody else's pieces instead. This
has gotten *much* better with the
Ben,
I fully understand your frustration. Many times when I sat down in
the morning to work on something specific, I found that the MacOS
build broke and spent my time picking up somebody else's pieces
instead. This has gotten *much* better with the introduction of the
validate script,
Simon Marlow wrote:
Right - but we've explicitly said which platforms are everyone's
problem, and which are not. You have to draw the line somewhere, and
it's better for the line to be clear than fuzzy.
Unfortunately both Sparc and Solaris are on the wrong side of the
line. But without a r
On 19/08/2009 08:44, Ben Lippmeier wrote:
Simon Marlow wrote:
ghc-stage1: panic! (the 'impossible' happened)
(GHC version 6.11.20090817 for sparc-sun-solaris2):
Error in array index
I have no idea what's going on here, but it looks suspiciously
interface-file related (though I'm not sure why th
On 19/08/2009 10:06, Ben Lippmeier wrote:
Simon Marlow wrote:
We (i.e. Simon PJ, Ian and myself) commit to keeping our tier-1
platforms working, namely Windows, x86 and x86-64/Linux and x86/MacOS
X. We can't realistically add any more platforms to this list without
having an expert with access
On Wed, Aug 19, 2009 at 07:06:02PM +1000, Ben Lippmeier wrote:
> Perhaps what we need is some better build bots. It would be good to have
> a mechanism to run a build with "the current head + these patches" on
> all architectures easily.
It would also be helpful for those who are running build s
Simon Marlow wrote:
We (i.e. Simon PJ, Ian and myself) commit to keeping our tier-1
platforms working, namely Windows, x86 and x86-64/Linux and x86/MacOS
X. We can't realistically add any more platforms to this list without
having an expert with access to a box to help us fix things.
Yeah, I
On 19/08/2009 00:40, Ben Lippmeier wrote:
Simon Marlow wrote:
I don't think it's as bad as you make out - all the Tier 1 platforms
are validating at the moment, as far as I know. The nightly builds
test a lot more than validate (e.g. binary distributions), but for
development all you need is val
Simon Marlow wrote:
ghc-stage1: panic! (the 'impossible' happened)
(GHC version 6.11.20090817 for sparc-sun-solaris2):
Error in array index
I have no idea what's going on here, but it looks suspiciously
interface-file related (though I'm not sure why that would be
Sparc-specific at a
Simon Marlow wrote:
I don't think it's as bad as you make out - all the Tier 1 platforms
are validating at the moment, as far as I know. The nightly builds
test a lot more than validate (e.g. binary distributions), but for
development all you need is validate.
x86_64/Linux is fine - I use it
On 18/08/2009 14:36, Simon Marlow wrote:
The unregisterised build is broken, probably for tedious header-file
related reasons. I need to look into that, but it doesn't affect the
register allocator.
And I just realised after making a ticket for the above, that I have a
patch to fix it sitting
On 18/08/2009 13:54, Ben Lippmeier wrote:
#2790: Use -fregs-graph by default
...
Comment (by simonmar):
Ben - do you plan to look into this in time for 6.12.1? (RC 11 Sept)
I'd like to, but I can't do much with the NCG when the head is not
building on some architectures. From the bui
#2790: Use -fregs-graph by default
...
Comment (by simonmar):
Ben - do you plan to look into this in time for 6.12.1? (RC 11 Sept)
I'd like to, but I can't do much with the NCG when the head is not
building on some architectures. From the build bot logs, the x86_64
build has been
18 matches
Mail list logo
