- Glasgow-haskell-users

GHCi hacking
by Ulf Norell 21 Aug '06

21 Aug '06

Hi. I want to use GHCi as the interface to my Application. The simple solution is to have the Application store its state in global IORefs. A user can then start up ghci with -package Application and use the provided functions to communicate with the Application. This works nicely. Now the tricky part. I want the user to be able to implement her own layers on top of the Application API. A typical user interaction could look something like this: $ ghci -package Application *Application:API> startApplication *Application:API> :load UserLayer *main:UserLayer> myCleverFunction 42 The problem, of course, is that as soon as the user says ':load' ghci forgets all about the state of the Application. My solution to the problem is to compile my own version of ghci (copy InteractiveUI.hs and use -package ghc) and remove the call to rts_revertCAFs when loading new modules. This seems to work, but since I don't really have a clue what I'm doing I wanted to ask a few questions: 1) What is a CAF? 2) What breaks down if you don't revert them? 3) Does not reverting the CAFs really solve my problem? / Ulf

5 6

Mac/PPC threaded RTS problems -- potential clue
by David Kirkman 21 Aug '06

21 Aug '06

I managed to build ghc-6.5.20060804 on a powerpc mac, and I spent a little time Saturday night trying to work out what might be going on with the threaded RTS. Running the testsuite with make WAY=threaded1 shows many (73) failures. Many of them (the conc??? set) have to do with TVars, in particular writing TVars. On my dual-proc G5 macs, with ghc-6.5.20060804, the following program will hang when running with the threaded RTS, but will work fine when compiled without -threaded. >module Main where > >import GHC.Conc >import Control.Concurrent > >main = do > t1 <- atomically (do t1 <- newTVar 80 > return t1) > atomically ( do writeTVar t1 1 ) > putStr ("done\n") This problem seems to be powerpc specific -- it works fine with the threaded RTS on a multi-processor intel mac (built by me, from the same ghc-6.5.20060804 source tree I used to build the ppc compiler). Looking around in STM.c (via decidedly low-tech printfs) I quickly zoomed in on cond_lock_tvar, which lead me to cas (atomic compare and swap) in SMP.h, where I found (I think) a fairly clear error in the powerpc code -- I've appended a patch to the end of this message. The problem is that the inline assembler code was placing the result of an operation in a register that is used as input later in the code. At the bottom of this message I've extracted a short short code fragment that you can run through gcc (on a powerpc machine) to see the generated assembly output. The changes to fix the problem are fairly simple. The first adds an ampersand to the output list of the assembly fragment ("=r" (result) --> "=&r" (result)) The ampersand just tells gcc that result can not be placed in a register used for any of the input parameters (o, n, or p). Otherwise, it feels free to place output parameters in the same registers used by the inputs -- but because of the flow of control here we need everything in a distinct register. This change fixes the TVar program above. The second change adds a clobber list (the :"cc", "memory"). This tells gcc that the condition code (due to the compare) and memory (due to the store) might be changed during the asm execution. The lack of a clobber list did not seem to be causing any trouble, but without it gcc is free to assume that no state is changed during the execution. Applying the following patch to SMP.h, and rebuilding everything, I not only fixed the simple writeTVar program, but it also fixed 8 programs in the testsuite (conc043 -> conc049, conc052 and conc053). The only conc test program that still fails is conc039. But there are still many mac problems, I still have 132 unexpected failures with make fast. At least the patch does not cause any new failures (in either make fast or make WAY=threaded1). Anyway, seeing as the change to SMP.h fixes a fair number of test cases in the testsuite, I figure there is some chance that it might fix some of the problems that people are having with the threaded RTS. On the other hand, I'm not real happy the the large number of testsuite failures my build gets, so I can't really call this 'tested'. But I'm posting because it might be a useful clue for somebody with a little more mac/ghc experience. Cheers, -david k. // // Short code to run through gcc -S. On my powermac, without the change // the generated assembly produces // 1: lwarx r0, 0, r0 // load to here ---^ ^------- from address here // But we need this value for the stwcx. // // with the fix, the first line of the generated assembly becomes // 1: lwarx r11, 0, r0 // and r0 remains unmodified if we need to use it later in the stwcx. /* * CMPXCHG - the single-word atomic compare-and-exchange instruction. Used * in the STM implementation. */ long cas(long* p, long o, long n) { long result; // // Change // :"=r" (result) --> :"=&r" (result) // to get result and p in different registers __asm__ __volatile__ ( "1: lwarx %0, 0, %3\n" " cmpw %0, %1\n" " bne 2f\n" " stwcx. %2, 0, %3\n" " bne- 1b\n" "2:" :"=r" (result) :"r" (o), "r" (n), "r" (p) ); return result; } Here's a "diff -cp" for SMP.h *** SMP.h Sun Aug 13 01:08:53 2006 --- SMP-new.h Sun Aug 13 01:08:47 2006 *************** cas(StgVolatilePtr p, StgWord o, StgWord *** 76,83 **** " stwcx. %2, 0, %3\n" " bne- 1b\n" "2:" ! :"=r" (result) :"r" (o), "r" (n), "r" (p) ); return result; #else --- 76,84 ---- " stwcx. %2, 0, %3\n" " bne- 1b\n" "2:" ! :"=&r" (result) :"r" (o), "r" (n), "r" (p) + :"cc", "memory" ); return result; #else

3 2

Latest ghc build error
by Joel Reymont 21 Aug '06

21 Aug '06

/usr/local/bin/ghc -H16m -O -istage1/utils -istage1/basicTypes - istage1/types -istage1/hsSyn -istage1/prelude -istage1/rename - istage1/typecheck -istage1/deSugar -istage1/coreSyn -istage1/ specialise -istage1/simplCore -istage1/stranal -istage1/stgSyn - istage1/simplStg -istage1/codeGen -istage1/main -istage1/ profiling -istage1/parser -istage1/cprAnalysis -istage1/ ndpFlatten -istage1/iface -istage1/cmm -istage1/nativeGen - istage1/ghci -Istage1 -DGHCI -DBREAKPOINT -package template-haskell - threaded -package readline -DUSE_READLINE -cpp -fglasgow-exts -fno- generics -Rghc-timing -I. -IcodeGen -InativeGen -Iparser -package unix -ignore-package lang -recomp -Rghc-timing -H16M '-#include "cutils.h"' -i../compat -ignore-package Cabal -fno-cse -c ghci/ InteractiveUI.hs -o stage1/ghci/InteractiveUI.o -ohi stage1/ghci/ InteractiveUI.hi /tmp/ghc17177_0.hc: In function 'smEl_ret': /tmp/ghc17177_0.hc:42472:0: error: void value not ignored as it ought to be /tmp/ghc17177_0.hc: In function 'smEf_ret': /tmp/ghc17177_0.hc:42504:0: error: void value not ignored as it ought to be <<ghc: 812568504 bytes, 1358 GCs, 15125404/31986692 avg/max bytes residency (10 samples), 80M in use, 0.00 INIT (0.00 elapsed), 2.39 MUT (10.94 elapsed), 1.75 GC (1.86 elapsed) :ghc>> -- http://wagerlabs.com/

2 1

Instruction to build current GHC 6.5 on Mac OS X/Intel x86
by Esa Ilari Vuokko 21 Aug '06

21 Aug '06

Hi, Paolo Martini got ghc to built using Audrey Tang's build from 20060608 on Mac OS X/Intel x86. Instructions are available at ghc trac: http://hackage.haskell.org/trac/ghc/wiki/X86OSXGhc Please edit the page if you have corrections. Best regards, --Esa Ilari Vuokko

2 1

Re: [Haskell] Haskell Workshop 2006 Call for participation
by Bulat Ziganshin 21 Aug '06

21 Aug '06

Monday, August 14, 2006, 10:34:22 AM, Andres Loeh wrote: > 14:30 Simon Marlow (Microsoft Research) > An Extensible Dynamically-Typed Hierarchy of Exceptions is this planned to be included in ghc 6.6? 6.8? > 15:00 David Himmelstrup (Denmark) > Demo: Interactive Debugging with GHCi and this? -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com

2 1

ANNOUNCE: GHC's new support engineer: Ian Lynagh
by Simon Marlow 21 Aug '06

21 Aug '06

We are delighted to announce that GHC now has a full-time support engineer. He is Ian Lynagh (aka Igloo on IRC), and is well known to many of you on the Haskell and GHC mailing lists. He'll be helping with all aspects of GHC, especially release management, bug diagnosis and tracking, documentation, packaging, and supporting other GHC hackers. Welcome Ian! We were absolutely delighted by both the number and quality of the people who applied for the post, and we want to thank them all very much for doing so. We love you all! Andrew Herbert, the lab director of Microsoft Research Cambridge lab director, also deserves a big thank-you, for supporting this post (no strings attached!). It'll make a huge difference. Looking forward to seeing many of you soon, at the GHC Hackathon [1] (25 people registered), and/or at ICFP in Portland. Simon Peyton Jones and Simon Marlow [1] http://hackage.haskell.org/trac/ghc/wiki/Hackathon

1 0

Efficiency of using field labels vs pattern matching
by Brian Hulley 20 Aug '06

20 Aug '06

Hi, I'm wondering if it is more efficient to pattern match against the components of a record instead of using the field functions, though I'd prefer to just use the field functions. For example, if I write: data R = R {_xRef :: !(IORef Int)} foo :: Int -> R -> IO () foo i R{_xRef = xRef} = writeIORef xRef i is the above more efficient than: foo i r = writeIORef (_xRef r) i I know this is probably a bit silly to worry about the overhead of an extra function call etc but I've at the moment got a lot of code using the first version and I think it would look much neater just using field labels but I don't want to modify it if it will be less efficient since speed is quite critical here. Thanks, Brian. -- Logic empowers us and Love gives us purpose. Yet still phantoms restless for eras long past, congealed in the present in unthought forms, strive mightily unseen to destroy us. http://www.metamilk.com

3 4

Digraph.lhs broken?
by Joel Reymont 19 Aug '06

19 Aug '06

/usr/local/bin/ghc -H16m -O -istage1/utils -istage1/basicTypes - istage1/types -istage1/hsSyn -istage1/prelude -istage1/rename - istage1/typecheck -istage1/deSugar -istage1/coreSyn -istage1/ specialise -istage1/simplCore -istage1/stranal -istage1/stgSyn - istage1/simplStg -istage1/codeGen -istage1/main -istage1/ profiling -istage1/parser -istage1/cprAnalysis -istage1/ ndpFlatten -istage1/iface -istage1/cmm -istage1/nativeGen - istage1/ghci -Istage1 -DGHCI -DBREAKPOINT -package template-haskell - threaded -package readline -DUSE_READLINE -cpp -fglasgow-exts -fno- generics -Rghc-timing -I. -IcodeGen -InativeGen -Iparser -package unix -ignore-package lang -recomp -Rghc-timing -H16M '-#include "cutils.h"' -i../compat -ignore-package Cabal -fglasgow-exts -c utils/Digraph.lhs -o stage1/utils/Digraph.o -ohi stage1/utils/ Digraph.hi utils/Digraph.lhs:133:12: Ambiguous occurrence `indices' It could refer to either `GHC.Arr.indices', imported from Array at utils/Digraph.lhs:43:0-11 or `Data.Array.Base.indices', imported from Data.Array.ST at utils/Digraph.lhs:48:0-19 utils/Digraph.lhs:139:18: Ambiguous occurrence `bounds' It could refer to either `GHC.Arr.bounds', imported from Array at utils/Digraph.lhs:43:0-11 or `Data.Array.Base.bounds', imported from Data.Array.ST at utils/Digraph.lhs:48:0-19 -- http://wagerlabs.com/

3 7

can not find prelude module
by Liang Guang 18 Aug '06

18 Aug '06

I originally used Redhat9.0, and the compilation is fine. But now I switched to Fedora 5, the compiler said it can not find prelude module! I used Yum to install the ghc-6.4.2, and it is put to into /usr/lib/ghc-6.4.2. I tried to add the directory to the makefile as : GHC_OPTS = -O2 -prof -auto-all -i../src:/usr/lib/ghc-6.4.2/ I also tried GHC_OPTS = -O2 -prof -auto-all -i../src:/usr/lib/ghc-6.4.2/imports but they didn't work. I am quite confused, isn't it that the base library(like prelude module) should be included in the search path by default? How can I deal with the problem? Thanks! TOm --------------------------------- Do you Yahoo!? Everyone is raving about the all-new Yahoo! Mail Beta.

2 1

Re: GHCi hacking
by Ulf Norell 18 Aug '06

18 Aug '06

On Aug 18, 2006, at 1:06 PM, pepe wrote: > If you have gone that far and modified ghci, wouldn't it be easier to > disable the :load command and hardwire the loading of the UserLayer > module ? Ah, but the thing is that the UserLayer is written by the user and so we don't know what it is before-hand. Also, the user should be able to switch between different Layers without destroying the application state. > 1) CAFs (Constant Applicative Form) are essentially top level > bindings. Top level bindings are closures, as anything in ghc, and > they are in suspended state until they are evaluated. When this > happens the closure is updated to store the computed value, and > because of this evaluation effectively happens only once. > > 2) I don't know, but probably you will notice during your experiments > > 3) Probably no. What happens to top level bindings in the interactive > environment? I think that you lose that environment too after load and > some other commands. An option is to identify and disable all these > commands. I don't want to disable the commands, just the clearing of the environment. At the moment I never revert the CAFs so I'm hoping that the environment will be preserved whatever I the user does. / Ulf

1 0