- Haskell-Cafe - Haskell.org

RE: Concurrent-Haskell in GHC
by Simon Marlow 04 Feb '03

04 Feb '03

> | I've started looking into the docs on GHC's > | implementation of Concurrent-Haskell, and I got > | confused about the architecture. Different sources > | seem to indicate that it either: > | - Uses one OS thread, and blocking IO calls will > | therefore block all the Haskell-threads, or > | - Uses one OS thread, occasionaly (50 times a second?) > | calling select() to simulate blocking IO calls without > | blocking other Haskell-threads, or > | - Uses a pool of OS threads so that when a blocking IO > | call is made, another thread takes over execution. > > Basically the last of these. (At least in the most recent > GHC.) Simon > Marlow implemented a web server using Concurrent Haskell, and has a > paper about it too. http://www.haskell.org/~simonmar/bib.html But we should really point out that you get the second method by default; the "pool of OS threads" method is currently experimental and not enabled unless you rebuild GHC's RTS with the right flags. The default single-threaded IO multiplexing works pretty well, although it doesn't scale to multiple processors of course. Cheers, Simon

1 0

Re: GHC and C++ (was RE: Creating COM objects and passing out pointers to them via a COM interface)
by Sarah Thompson 04 Feb '03

04 Feb '03

I'm replying to two threads at the same time and cross posting my reply, because they are very relevant to each other. I'm sorry if anyone here ends up seeing this more than once as a consequence. [s] ------ > Sarah > > Did you get this problem sorted out? Not directly. I ended up building an 'outer' COM component in C++ that mapped an MSXML-like interface onto a simpler, internal COM component that wraps up the Haskell code. This approach seems to work well - I might have been able to build the necessary wrapper directly in Haskell using HDirect, but I was becoming aware that I was probably starting to take up an unreasonable amount of Sigbjorn's time with increasingly detailed questions and I didn't have time to figure out the necessary intricacies from scratch. HDirect seems very good, but I don't think it's currently in quite as stable a state as the main GHC release. The problems I had in building a version that works with the current GHC are an example of this. I'm not complaining - it worked well in the end, but I do think it might be a 'good thing' to consider rolling HDirect into the main GHC release at some point. The other thing that bit me was exception handling, or rather the default behaviour when an uncaught exception 'falls out' and reaches the top level of the runtime system. Currently, the functionality appears to be to abort the current process. Whilst this is probably desirable for a typical application executable (be that console or GUI based), it is a problem if it happens inside a COM component. My thinking is quite straightforward on this. Preferred behaviour should really be for a call to a COM component to fail returning an error code, rather than abort the calling process. 'Server-style' applications are usually expected to keep running when minor faults occur, rather than terminate. HDirect/GHC's current behaviour makes this difficult to achieve - I managed to successfully catch exceptions, but the price was needing to use some code based on DeepSeq to force deep strict evaluation for all state updates. This (seemingly) was the only way I could stop exceptions from happening in the context of the automatically generated COM wrapper - passing it a fully evaluated data structure was apparently the only solution available. However, this does break the advantages of laziness as applied to the current state of the object - it would be pretty neat, for example, to be able to define a view of a database as the result of an expression involving relational algebra combinators, with the query evaluated one record at a time as the data is pulled back by the client application. The need for DeepSeq forces strict evaluation, so this benefit is lost. This is actually more of an issue than it might immediately seem - many queries don't actually need to retrieve all of a record set before useful things can be done. A particular example is updating a GUI, where a list box reflects the results of a query. Lazy evaluation will allow the screen to start updating more or less immediately even on a query that returns thousands of records - strict evaluation might incur a delay of a few seconds (for a very complex query or a large database) before updates can start happening. In an ideal world, what I'd dearly love to see would be an HDirect option that allows a choice between existing functionality and having calls return a failure code, or better still allowing a user-supplied handler function to determine the correct course of action. So, to sum up my wish list: 1. A facility to replace the default uncaught exception behaviour for COM objects and similar shared library code 2. HDirect rolled into the main GHC tree 3. A facility within HDirect to create COM objects based on existing (Haskell) data structures, returning an interface pointer as a return parameter on a call to an existing COM object. (This was the question Simon asked if I'd had an answer to - I've attached a copy below for reference). Sarah PS: I've not yet looked closely at the .NET integration in GHC yet, but I can imagine myself having a very similar wish list in that case too. -------- Hi again, A slightly harder problem this time. I want to implement something that looks something like the Microsoft XML parser. This is a very COM heavy contraption that provides an interface that looks to client applications like a tree of COM objects that reflects the internal structure of an XML document. The idea is, you start with the document object, make a method call requesting the root node, and have a COM object that wraps that node returned to you. You can then request the children of that node, and their children and siblings, all which involves being passed back interface pointers to COM objects. Effectively, there is one COM object per syntactic item in the XML document. I already have my document object working nicely. This is where the actual database resides. I want to be able to spawn views of this database as COM objects. When I've done this in the past (using ATL mostly, euuuugh), I've used an IDL definition something like HRESULT CreateView([in,string] BSTR query, [out,retval] IDbView **); What's the correct way to implement the necessary Haskell code to create the coclass and return a pointer to it? I fiddled around for ages, but either couldn't get GHC to like the types I was using, or when I finally got it to compile, I found that my test harness was being passed an invalid COM object. This should be simple stuff, so I'm sure someone will have done it! Secondly, it is a nasty, though fairly common trick, if you know your objects are going to be created in the same process space, to allow them some 'behind the scenes' communication. In my case, I'd like to avoid needing to serialise and deserialise the data in a view during creation of these child objects (of which there may be a great many, and which may be created on the fly rather often). Ideally, I'd like to be able to do something like: newView :: String -> State -> IO State newView db (State st) = do db <- readIORef st r <- newIORef db -- Insert some kind of COM magic here return (State r) so the database content gets copied within the Haskell environment (lazily, I'd hope), without needing to be serialised and deserialised. Is this feasible? One (messy) approach might be to create a bunch of extra interfaces on the main database object, then create 'thin wrapper' child objects that simply make calls back into the parent object. I'm not so keen on that for various reasons, however. Sarah

1 0

RE: Concurrent-Haskell in GHC
by Simon Peyton-Jones 04 Feb '03

04 Feb '03

| I've started looking into the docs on GHC's | implementation of Concurrent-Haskell, and I got | confused about the architecture. Different sources | seem to indicate that it either: | - Uses one OS thread, and blocking IO calls will | therefore block all the Haskell-threads, or | - Uses one OS thread, occasionaly (50 times a second?) | calling select() to simulate blocking IO calls without | blocking other Haskell-threads, or | - Uses a pool of OS threads so that when a blocking IO | call is made, another thread takes over execution. Basically the last of these. (At least in the most recent GHC.) Simon Marlow implemented a web server using Concurrent Haskell, and has a paper about it too. http://www.haskell.org/~simonmar/bib.html Simon

1 0

Re: Global variables?
by oleg＠pobox.com 03 Feb '03

03 Feb '03

Richard Uhtenwoldt wrote: > (2) the global variable has a lexical scope that extends over > the bulk of the program. > It strikes me as a simple and obvious application of lexical > scope, and I am surprised that it received no mention in the > discussions on this list and in Hughes's paper. Something like that has been mentioned before: http://www.haskell.org/pipermail/haskell-cafe/2002-September/003423.html If the global variables are used to store parameters that are read from configuration files at the beginning and are not changed since, it's possible to make such "global variables" _ordinary_ top-level variables scoped through the entire program. No unsafe operations, no pragmas, no implicit variables, and no reliance on the IO monad for variable access are needed. The previous article said the approach was ugly. I'd like to take it back. The approach relies on an ability to dynamically compile and link in a program -- something that we all do and something that is specifically emphasized in the GHCi documentation. The following code is snipped from the previous article and adjusted for GHC. The latter is used as a meta-evaluator/dynamic linker of sorts. Suppose file '/tmp/a.hs' contains the following user program. Suppose the program needs a global, configurational variable named Config.config_item. >>> File "/tmp/a.hs" > import Config (config_item) > > foo = "foo shows: " ++ (show config_item) > > bar = "bar shows: " ++ (show config_item) > > main = do > print foo > print bar > print foo Realistically a user program would do something more with the global variables than just showing them. We specifically illustrate multiple access to the global variable, to indicate that such access poses no trouble (in contrast to the unsafePerformIO approach). We also note that user computations 'foo' and 'bar' do not have to be annotated with the type and the name of the global variable -- in contrast to the implicit variable approach. Furthermore, foo and bar are pure functions rather than IO actions. The value of the global variable Config.config_item is computed by reading it from the configuration file. The computation occurs before running of the main function of the user program. The following code accomplishes the trick *>>> File "/tmp/b.lhs" *>>> To run this code, do *>>> ghci -package posix b.lhs *>>> Be sure /tmp/config exists and contains an integer value > import System (system, ExitCode(ExitSuccess)) > import Posix(executeFile) > > myconfig_file = "/tmp/config" > > phaseII_var = "/tmp/Config.hs" > phaseII_const = "/tmp/a.hs" > phaseII_eval = "ghc --make " > phaseII_result = "/tmp/a.out" > > nl = "\n" > > writeConfig :: Int -> IO () > writeConfig num = > do > writeFile phaseII_var $ > concat > ["module Config (config_item) where", nl, > "config_item =", show num, nl] > > > runSuperIO () = system (phaseII_eval ++ > phaseII_const ++ " -o " ++ phaseII_result) > >>= \ExitSuccess -> > executeFile phaseII_result False [] Nothing > > main = readFile myconfig_file >>= writeConfig . read >>= runSuperIO We can indeed be sure that the configurational variable is set before the first attempt to access it.

1 0

Concurrent-Haskell in GHC
by Jonathan Holt 03 Feb '03

03 Feb '03

Hi guys! I'm considering doing a server project in Haskell, using GHC, and I'm having the old dilema of trying to figure out the best way of handling multiple concurrent connections: multiplexing or multithreading. I've started looking into the docs on GHC's implementation of Concurrent-Haskell, and I got confused about the architecture. Different sources seem to indicate that it either: - Uses one OS thread, and blocking IO calls will therefore block all the Haskell-threads, or - Uses one OS thread, occasionaly (50 times a second?) calling select() to simulate blocking IO calls without blocking other Haskell-threads, or - Uses a pool of OS threads so that when a blocking IO call is made, another thread takes over execution. So which is right for the current stable GHC? And how efficient/scalable is it? Thanks, -Jony __________________________________________________ Do you Yahoo!? Yahoo! Mail Plus - Powerful. Affordable. Sign up now. http://mailplus.yahoo.com

1 0

Global variables?
by pavel＠joker.botik.ru 03 Feb '03

03 Feb '03

Hello, Is it even possible to make a "global variable" in Haskell? If yes, how? Thanks. -- Zhbanov Pavel

8 7

Re: Global variables?
by Jon Cast 02 Feb '03

02 Feb '03

Fergus Henderson <fjh(a)cs.mu.OZ.AU> wrote: > On 02-Feb-2003, Jon Cast <jcast(a)ou.edu> wrote: > > Claus Reinke <claus.reinke(a)talk21.com> wrote: > > > It is the programmer's responsibility to verify that none of these > > > problems matter in the particular case of usage. Since many > > > advances in compiler technology tend to invalidate those > > > verifications, it is almost impossible to guarantee safety in such > > > cases > > Even with sufficiently liberal use of {-# NOINLINE #-}? > Haskell Report, Appendix E (Compiler Pragmas): "An implementation is > not required to respect any pragma". Right. So perhaps the Haskell Report should take more notice of global variables. But considering that it doesn't, and there is no completely portable mechanism for getting global mutables, it seems to me that bashing any alternative as non-portable is missing the point :) > > > There you went.. into one of the many available traps in this > > > mine-field: > > > You argue that unallocated IORefs don't matter as long as "the" > > > IORef is allocated before it is dereferenced. But that's just part > > > of the problem - what about inlining globalVar, creating multiple > > > IORefs? > > {-# NOINLINE globalVar #-} :) > What about creating two copies of globalVar, neither of them inlined? > A compiler could do that to improve performance on some architectures, > e.g. by allowing the function to be called via a short jump > instruction rather than a long jump instruction. See above. > -- > Fergus Henderson <fjh(a)cs.mu.oz.au> | "I have always known that the pursuit > The University of Melbourne | of excellence is a lethal habit" > WWW: <http://www.cs.mu.oz.au/~fjh> | -- the last words of T. S. Garp.

1 0

Re: Global variables?
by Jon Cast 02 Feb '03

02 Feb '03

Claus Reinke <claus.reinke(a)talk21.com> wrote: > > > > import IORef > > > > import IOExts > > > > > > > > globalVar :: IORef Int > > > > globalVar = unsafePerformIO $ newIORef 0 > > > John Hughes wrote a nice pearl on the subject, see > > > http://www.math.chalmers.se/~rjmh/Globals.ps > > This paper claims ``unsafePerformIO is unsafe''. That's not > > actually true in the sense meant; unsafePerformIO merely has safety > > pre-conditions that the compiler can't check. > Which is the main sense in which the 'unsafe' prefix is usually meant > to be interpreted, and that is bad enough (see below, then re-read > John's quote of Simon PJ's description of unsafePerformIO;-). In > particular, the 'unsafe'-prefix is not a hint for the implementation > to treat something with extra care, but a hint for the programmer that > the implementation may shake unsafe expressions around like any other > ones (inlining, cse, ..), even though that is bound to lead to > problems with the hidden side-effects. So you have to be careful. If that weren't what I said, I'd disagree with you :) > It is the programmer's responsibility to verify that none of these > problems matter in the particular case of usage. Since many advances > in compiler technology tend to invalidate those verifications, it is > almost impossible to guarantee safety in such cases Even with sufficiently liberal use of {-# NOINLINE #-}? > - about the best one can hope for is to identify and document > precisely which assumptions need to be made to "guarantee" > safety. Unfortunately, this leaves it to users to figure out whether > the assumptions made by 'unsafe' authors (e.g., no inlining) are still > valid at the point of use.. > Btw, when talking about unsafety in that paper, John also happens to > point out the other little problem with unsafePerformIO: it permits to > break type safety (many a good spirit has stumbled over that > "polymorphic references" problem in other functional languages). > > The precondition (proof obligation) of unsafePerformIO is that the > > order in which unsafePerformIOs are performed cannot affect the > > outcome of the program. However, in this case, ordering doesn't > > matter: the only side effect is allocation of a new IORef, and > > IORefs are sufficiently opaque we don't care (or really know) about > > un-allocated IORefs while the only case we care about the > > now-allocated IORef is when we de-reference it. But, that forces > > the IORef, which executes the unsafePerformIO. So, whenever we > > access the variable, it is allocated. Therefore, the outcome of the > > program (regardless of the order of evaluation) is the same as if > > all such global variable declarations are executed before main > > begins executing. So, the outcome is independent of the order of > > evaluation. > > There you go: the precondition of unsafePerformIO is satisfied, so > > the usage is safe. > There you went.. into one of the many available traps in this > mine-field: > You argue that unallocated IORefs don't matter as long as "the" IORef > is allocated before it is dereferenced. But that's just part of the > problem - what about inlining globalVar, creating multiple IORefs? {-# NOINLINE globalVar #-} :) > Remember that, by using unsafePerformIO, you've given the compiler the > license to treat globalVar as an expression without side-effects, and > for those inlining is a common first step to enable further > optimisations (last time I checked, the language report didn't even > guarantee the sharing on which the globalVar trick depends)! Now there > are multiple side-effects instead of a single one, and read-and > write-accesses are spread out over the multiple copies of your IORef, > none of which is likely to hold the value you'd like it to have.. > Cheers, > Claus Jon Cast

2 1

Re: Global variables?
by Jon Cast 02 Feb '03

02 Feb '03

Andrew J Bromage <ajb(a)spamcop.net> wrote: > G'day all. > On Fri, Jan 31, 2003 at 01:54:26PM -0600, Jon Cast wrote: > > Otherwise, though, see my other post on this subject: > > unsafePerformIO will perform its action when the variable is > > accessed, so you can't write a Haskell program which differentiates > > between what any compiler actually does and running the variable > > allocations before main. > As has been pointed out, there is no language requirement for a > Haskell implementation to be "fully lazy". In particular, it is > technically possible for an implementation to garbage collect > globalVar and re-evaluate it on the next call. So, the technique is implementation-dependent. Since Hughes' recommendation is equally non-portable, I don't see this as a serious problem in this case. > Haskell 2 should probably have a pragma controlling this. IMO, that would be a good idea :) > Cheers, > Andrew Bromage

1 0

Re: Global variables?
by Jon Cast 02 Feb '03

02 Feb '03

Patrik Jansson <patrikj(a)cs.chalmers.se> wrote: > On Fri, 31 Jan 2003, Jon Cast wrote: > > Otherwise, though, see my other post on this subject: > > unsafePerformIO will perform its action when the variable is > > accessed, so you can't write a Haskell program which differentiates > > between what any compiler actually does and running the variable > > allocations before main. > If you initialize this "global variable" with a value from a file > which is changed by main (or over time) then order does matter. 1. Why would you do this? In any case, the solution adopted in C (initialize the global statically with ``no value'' (0, NULL, Nothing, etc.) and overwrite it at a fixed point in main or elsewhere) should be good enough. 2. In any case, my claim is actually: If a :: IO alpha, for monomorphic alpha, and a's only interaction with the state consists of either creating new IORefs or accessing IORefs created by a, then unsafePerformIO's result is unaffected by the order of evaluation. > I claim unsafePerformIO is very often unsafe It's always potentially unsafe. However, it is always possible to prove that a particular application is safe (under sufficiently stringent conditions, of course). > - the question is more like if the "level of unsafety" is acceptable > for a particular application. There is no "level of unsafety". There is only the level of trust you put in your proof of satisfaction of unsafePerformIO's proof obligation and your view of the acceptability of the portability restrictions of your proof. > /Patrik Jon Cast

1 0