- Glasgow-haskell-users

No NoFib
by Kevin Glynn 01 May '02

01 May '02

Hi, I can't find a tarred up release of the NoFib suite, I can only find it in the CVS repository. Is this intentional? The link from GHC's home page points to a very old page at Glasgow which doesn't give a link to the software. (Perhaps the link in the paper still works - I haven't tried). k

1 0

announce: GHC internals library (version 0.1)
by Bernard James POPE 01 May '02

01 May '02

Hi all, Announcing the GHC internals library, version 0.1. Essentially this provides a means for polymorphically deconstructing values that reside on the GHC heap, all from the comfort of your Haskell program. For those who are familiar with the HugsInternals library that comes with Hugs, it offers similar support to getCell and classifyCell, in a GHC-centric way. What might you use this for? I don't know, but I'm using it for debugging, maybe you can too, or alternatively, you might find another use. You can find it here: http://www.cs.mu.oz.au/~bjpop/code.html For your curiosity, here's _one_ thing you can do with it: > import GhcInternals > > main :: IO () > main = do let val = zip [1..12] (iterate not True) > print $ take 3 val > metaVal <- reify False val > putStrLn $ prettyPrintMeta metaVal [] Running this program you get: [(1,True),(2,False),(3,True)] (: ((,) 1 True) (: ((,) 2 False) (: ((,) 3 True) THUNK))) How does it work? Basically, using some odd looking FFI code it peaks at the heap representation of a value, and constructs a meta representation of it. The deconstruction is only done one constructor at a time, although provision is made for the whole structure to be inspected (more or less) at once. Cycles and sharing in general are not yet supported, but future versions will attempt to solve this, perhaps with some tinkering with Stable Pointers. You can choose to inspect the structure lazily (ie no further evaluation) or strictly (ie force the structure to evaluate one constructor at a time). Functions can be printed, but mostly you'll get some funny string that the compiler generated. This might be improved in later versions of the library. Currently it relies on compiling your program with -prof to make sure the names of constructors are retained in the heap. No modifications are required to your compiler, it should compile out of the box without any hackery to GHC itself. Works with version 5.03 of GHC, but you need a very current version to avoid exposing some bugs in the Stable Pointer implementation. Cheers, Bernie.

2 1

announce: StableTable Printer for GHC
by Bernard James POPE 01 May '02

01 May '02

Hi all, Ever had that deep desire to see what was in that pesky Stable Pointer/Name table? Probably not, but I did, and I wrote some code to print it out in all its glory from the safe confines of a Haskell program. I thought I'd share it with you just in case one day you too want to see what is lurking in there. You can find the code here: http://www.cs.mu.oz.au/~bjpop/code.html Here's an example of what it can do: ----------------------------------------------------- > module Main where > > import Stable > foreign import ccall "printStableTable" printStableTable :: IO () > > main = do let list = [1,2,3,4] ++ list > print $ take 5 list > trueSName <- makeStableName True > listSPtr1 <- newStablePtr list > listSPtr2 <- newStablePtr list > printStableTable If you run this program, you get as output: [1,2,3,4,1] --- Begin Stable Table --- stable_ptr_free = 3 0.val = NULL 0.ref = 0 0.sn_obj = NULL 1.val = (addr = 0x80963e0) True 1.ref = 0 1.sn_obj = <STABLE_NAME> 2.val = (addr = 0x500c11c4) (: 1 (: 2 (: 3 (: 4 cycle)))) 2.ref = 2 2.sn_obj = NULL 3.val = free at pos 4 3.ref = 0 3.sn_obj = NULL 4.val = free at pos 5 4.ref = 0 4.sn_obj = NULL ... blah blah blah, until the end of the table ... ----------------------------------------------------- Some features: It shows each entry in the table, and implicitly the free list, starting at stable_ptr_free. Data structures are printed in all their glory, including the current address of the value and the representation of the value. Cycles are detected! (See 2.val), but other types of sharing are not yet shown, however it is possible with tweaks. Why might you want this? 1. Debugging the stable pointer table implementation. 2. Debugging your own code. Stable pointers introduce the potential for nasty space leaks in Haskell programs. Much like malloc/free in C. If you forget to free your pointers then the space is retained. If you suspect that your program is leaking space in this way, then just whip out the table printer and see what is looks like. 3. Checking how much sharing you have in your data (with tweaks, as I said ealier you can show sharing). What do you need to use it? Just a current version of GHC. It should compile out of the box without any modification to GHC. Oh, and you have to compile your program with -prof to see all the data constructors. How does it print the data structures with cycles? The library also comes with yet another library for prinitng arbitrary Haskell Data structures, done via the FFI interface. NOTE: this is different from the GHCinternals library. We can detect cycles more easily if we are just printing things and not building new values on the heap at the same time. Probably this extra library should live somewhere by itself, but I have just bundled them together at the moment. Cheers, Bernie.

1 0

RE: building with make -j2 ?
by Simon Marlow 30 Apr '02

30 Apr '02

> > "Simon Marlow" <simonmar(a)microsoft.com> writes: > > > > > The snapshot source distribution has now been updated to > include the > > > file mk/package.mk which was missing the first time around. > > > > Thanks. > > > > Btw, does ghc build with "make -j2"? I tried building the > > snapshot just now with it but got a make error at some > > point. I forget to save the build log, but I can easily > > fire up another build if you want to see it. > > > > It *mostly* worked for me a month or so ago. I did a single-processor > make until I saw it had started on ghc/compiler, and then hit ^C and > did make -j2 in ghc/compiler. This worked fine, and then I continued > the make with one processor for the rest. Presumably the same trick > would work in hslibs/. I think there's some setup that depends on > implicit ordering, but once it gets going in a directory it's all > fine. I've fixed a couple of things in the Makefile setup, so 'make -j' should be better, at least under fptools/ghc. I'm not sure it's totally robust, but please feel free to give it a whirl. Cheers, Simon

1 0

RE: resetting in ghci
by Simon Marlow 30 Apr '02

30 Apr '02

> Simon Marlow writes > > > :load with no arguments will unload all the currently > loaded modules and > > bindings. > > (so, the values go to garbage, and memory will be recovered?) All the memory will be recovered except for object files (it turns out to be difficult to do this safely, so we just didn't tackle it). > Thank you. > `:?' command of ghci does not tell this about :load > I wonder whether it is explained in GHC Manual. Well, it's just a degenerate case of the :load command when the list of modules to load is empty. The manual doesn't mention this use explicitly, but I'll add a note to it. Cheers, Simon

1 0

RE: resetting in ghci
by Simon Marlow 30 Apr '02

30 Apr '02

> Dear GHC, > > Does ghci have a command of clearing > out of all the currently defined values > (like x set by `let x = 1') ? > What about un-loading modules? > Does the user need to exit and re-run ghci in order to reset the > created environment? > Such situation is not a tragedy, but still ... :load with no arguments will unload all the currently loaded modules and bindings. Cheers, Simon

2 1

RE: sparc/linux, again
by Simon Marlow 30 Apr '02

30 Apr '02

> Dear all, I can't get the hc-boot process to work for > sparc64-linux-debian. > I compiled from the ghc-4.08.2 hc-files, but when I run the > executable, I get > > Starting program: /home/joe/ghc-4.08.2/ghc/compiler/hsc > (no debugging symbols found)...(no debugging symbols found)... > (no debugging symbols found)... > Program received signal SIGBUS, Bus error. > 0x7ccdd8 in forceIO_info () > (gdb) where > #0 0x7ccdd8 in forceIO_info () > #1 0x7ccec0 in stg_init_ret () > #2 0x7c45b4 in schedule () > #3 0x7c4bd8 in waitThread () > #4 0x7c30a4 in rts_evalIO () > #5 0x7c2ab4 in main () > #6 0x700df450 in __libc_start_main () from /lib/libc.so.6 Which .hc files did you use? To port to a new architecture you really need to start from an unregisterised build. As I recall, 4.08.2 also has some 64-bit incompatibilities, which might cause problems porting to sparc64. I don't want to put you off, but this *really* isn't for the faint-hearted ;-) You might want to talk to Ken Shan <ken(a)digitas.harvard.edu> who did the Alpha port recently, and there are some useful messages in the mailing list archives (use Google's "search site" facility to search haskell.org for "GHC Alpha port"). Cheers, Simon

1 0

RE: powerpc-linux port
by Simon Marlow 30 Apr '02

30 Apr '02

> On Mon, Apr 29, 2002 at 05:50:49PM +0100, Simon Marlow wrote: > > > No, actually I'd like to have GHC, because many libraries > I want to > > > use depend on it. But good that you point this out, maybe > it would be > > > possible to have NHC compile GHC... Although the FAQ says it > > > would most probably not work. > > > > It most *definitely* would not work ;-) > > > > To bootstrap the unregisterised version, I suggest starting > from version > > 4.08.2. The reason is that bootstrapping more recent > versions from .hc > > files is significantly more difficult (but hopefully should > get easier > > again soon...). > > OK, be it 4.08.2 then. I see there are HC files for x86 and > Sparc, how do > I get (produce) them for PowerPC? You need to start from unregisterised .hc files, which you have to build on a supported platform (eg. Linux/x86). Here's a message that has some details on how to do the bootstrap: http://www.haskell.org/pipermail/glasgow-haskell-users/2000-October/0014 45.html (which coincidentally is a reply to another guy who asked about PPC/Linux...) Cheers, Simon

1 0

resetting in ghci
by Serge D. Mechveliani 30 Apr '02

30 Apr '02

Dear GHC, Does ghci have a command of clearing out of all the currently defined values (like x set by `let x = 1') ? What about un-loading modules? Does the user need to exit and re-run ghci in order to reset the created environment? Such situation is not a tragedy, but still ... ----------------- Serge Mechveliani mechvel(a)botik.ru

1 0

sparc/linux, again
by Johannes Waldmann 30 Apr '02

30 Apr '02

Dear all, I can't get the hc-boot process to work for sparc64-linux-debian. I compiled from the ghc-4.08.2 hc-files, but when I run the executable, I get Starting program: /home/joe/ghc-4.08.2/ghc/compiler/hsc (no debugging symbols found)...(no debugging symbols found)... (no debugging symbols found)... Program received signal SIGBUS, Bus error. 0x7ccdd8 in forceIO_info () (gdb) where #0 0x7ccdd8 in forceIO_info () #1 0x7ccec0 in stg_init_ret () #2 0x7c45b4 in schedule () #3 0x7c4bd8 in waitThread () #4 0x7c30a4 in rts_evalIO () #5 0x7c2ab4 in main () #6 0x700df450 in __libc_start_main () from /lib/libc.so.6 Any hints? -- -- Johannes Waldmann ---- http://www.informatik.uni-leipzig.de/~joe/ -- -- joe(a)informatik.uni-leipzig.de -- phone/fax (+49) 341 9732 204/252 --

1 0