- Glasgow-haskell-users

profiling crashes my code
by Hal Daume III 02 May '02

02 May '02

I'm in a profiling kind of mood (read: am sick of my program taking too long to run), but when I ghc with -prof -auto-all (or -auto or whatever), when I try to run my program it seg faults. I don't want to attach all the code, since it's really long, but I *have* tested profiling on smaller programs and it works fine. (Note that the program crashes regardless of whether I put +RTS -p or not.) Here are some "unusual" things I'm doing; hopefully someone will be able to say "oh, there's your problem": - using the ghc binary package (which using FastMutInt -- I don't really know what this does, but it looks scary) - i'm only compiling the main program with profiling -- it imports a whole lot of modules which are compiled with -O2, including binary which is compiled using -cpp also - i use the Word8 type extensively - i use unsafePerformIO (but only once per run of the code) - i use UArray Int Double (i think this also uses upIO) - uhm...i do a lot of file io :) That's all I can think of at the moment. It dies regarless of whether I use -O, -O2, or neither... -- Hal Daume III "Computer science is no more about computers | hdaume(a)isi.edu than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

3 2

problems with ld
by Hal Daume III 01 May '02

01 May '02

This isn't a ghc specific question as I've just found a similar problem with gcc, but I was hoping someone here could lend me a hand...I recently installed RedHat 7.2 on my x86 laptop and then immediately installed ghc 5.02.3 from rpm. Compilation works fine as does interpretation but linking doesn't and it seems to be a problem with ld. It complains "/usr/bin/ld: cannot find -lc". A similar thing happens in C, except in that ld cannot find -lc. I'm wondering if anyone has any idea what this means and how to fix it (I'm not a *nix hacker by any means). I've attached outputs from ghc and gcc when run with -v; the important error is at the very bottom. The test programs where, for haskell, "main = print 5" and for c "int main() { printf("Hello world"); }". Thanks for any help! - Hal -- Hal Daume III "Computer science is no more about computers | hdaume(a)isi.edu than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

2 1

profiling (again)
by Hal Daume III 01 May '02

01 May '02

i'm using ghc 5.02.1 for sparc solaris -- Hal Daume III "Computer science is no more about computers | hdaume(a)isi.edu than astronomy is about telescopes." -Dijkstra | www.isi.edu/~hdaume

1 1

RE: No NoFib
by Simon Marlow 01 May '02

01 May '02

> 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). We haven't done releases of the NoFib suite for a long time - there doesn't seem to be a need now you can get the sources from CVS. We generally tag NoFib at the same time as a GHC release, so to get the NoFib sources that correspond to GHC 5.02.3, for example, 'cvs co -rghc-5-02-3 nofib'. Cheers, Simon

2 1

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