- Glasgow-haskell-users

Where STM is unstable at the moment, and how we can fix it
by Sterling Clover 01 Sep '08

01 Sep '08

This email is inspired by the discussion here: http:// hackage.haskell.org/trac/ghc/ticket/2401 As the ticket discusses, unsafeIOToSTM is, unlike unsafePerformIO or unsafeInterleaveIO, genuinely completely unsafe in that there is no way to use it such that a segfault or deadlock is not at least somewhat encouraged. The code attached to the ticket creates a deadlock solely through using it to write to stdout. But, for the same reason that unsafeIOToSTM is unstable, unsafeInterleaveIO now is very unstable as well -- conceivably, data generated from functions with lazy IO (including those in the prelude) could cause deadlocks within STM, and even segfaults. In summary, a "validation" step is performed on all threads inside atomically blocks during garbage collection. This validation step will, on encountering invalid threads (i.e. ones which should be rolled back) immediately kill them dead and retry. This is different than the implementation described in the STM paper, where rollbacks only occur on commit. However, it does add a measure of efficiency. The problem is that the validation code disregards exception handlers, since rollback is not an exception, and so anything embedded in STM that brackets an IO action, for example, can be rolled back without the final part of the exception even being called. As Simon M. notes, the obvious solution would be to turn rollbacks into regular exceptions, but this would open a number of cans of worms. A start, though not sufficient, would be for stm validation to respect blocked status -- not to block on it, obviously, but simply to refuse to rollback a transaction within it. Validation on GC is, after all, only an efficiency trick and implementation detail, and if it lets the occasional invalid transaction stand due to its blocked status, that transaction will simply be cleaned up later anyway. A more thorough solution would be, as I suggest at the end of the ticket, to add a new primitive with similar semantics to block -- blockRollback, of type STM () -> STM (). Anything that took place within blockRollback could not be stopped by validation. Finally, we could "split the difference" between block and blockRollback, by simply setting a rollbackBlocked flag on a *top level* invocation of block within STM, and thenceforth, not unsetting it until that block is exited, regardless of calls to unblock nested inside. This would effectively, without introducing a new primitive, ensure that rollback did not disrupt things terribly, and thus would be the solution that handled the lazyIO issue the best as well. There are lots of interesting applications of STM that require the ability to extend its semantics. To do this is going to require unsafeIOToSTM, just as unsafePerformIO is used on occasion as a low level tool to create safer and better things on top of (or as unsafeCoerce is, for that matter). However, the current state of STM means that writing these extensions of STM semantics safely is 100% impossible. I'm not sure which, if any, of the solutions that I'm presenting seem the most reasonable. However, without some sort of resolution for this issue, STM is far less powerful and useful than it can and should be. --Sterl.

2 1

Version control systems
by Simon Marlow 29 Aug '08

29 Aug '08

Following lots of useful discussion and evaluation of the available DVCSs out there, the GHC team have made a decision: we're going to switch to git. It came down to two things: the degree of support available, and flexibility of the tools (git is much happier to let you modify the history than Mercurial). Speed ruled out bzr, and Windows support is less of an issue: git appears to work reasonably well on Windows these days. So we need a plan for switching. We aim to make the switch shortly before branching the repository for 6.10, which would mean we need to make the switch early September, in around 5 weeks time. Before then, the goal is to get the infrastructure to the point where we can switch with minimum fuss. We already have an up-to-date git mirror thanks to Thomas Schilling: git clone http://darcs.haskell.org/ghc.git (notice how fast that is :-) darcs-all will be able to work with either the git repository or the darcs repository (Max Bolingbroke is working on this, I believe). We can switch the automatic builds over to git as soon as darcs-all is working, and as long as the git mirror is kept up to date (Thomas: is the mirror being automatically updated now?). I'd urge people to try out the git mirror and let us know how you get on. We'll also work on updating the build documentation on the wiki and creating a page of getting-started info on using git. Cheers, Simon

35 157

"dataflow rewriting engine"
by Chad Scherrer 27 Aug '08

27 Aug '08

Hello GHC, This page http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/IntegratedCode… mentions a to-be-developed "dataflow rewriting engine". Can someone please send a description of what this will do? Thanks! -- Chad Scherrer "Time flies like an arrow; fruit flies like a banana" -- Groucho Marx

9 11

GHC's build system
by Simon Peyton-Jones 26 Aug '08

26 Aug '08

Friends There's been quite a bit of discussion about GHC's build system recently, and in particular about the use of Cabal. Responding to that discussion we now have a new plan, described here: http://hackage.haskell.org/trac/ghc/wiki/Design/BuildSystem If you've taken an interest in the earlier thread, now would be a good time to look at our plan and help us improve it. In concrete terms, we propose to develop the changes (which should not be a radical upheaval) on a branch, and merge them into the head *after* the 6.10 code fork. So the new build system won't be on the 6.10 branch. It'll take a while to make the new system stabilise, and we want to get 6.10 out! Some comments and suggestions are best done by email, but please feel free to clarify or amplify wording on the Wiki itself, or add lists of questions and open issues. The two media are suitable for different things. But we do want to end up with a *comprehensible* description, with long-term value, of how the new system works, so where things are obscure please help us to clarify it. All this is somewhat separate to the question of what version control system to use for what. We're still working on that question! Simon

1 0

Re: SHA1.hs woes, was Version control systems
by C Rodrigues 22 Aug '08

22 Aug '08

I've "fixed" this problem by increasing the number of registers used on ia64 to 34. The problem will show up again if anyone finds a way to make GCC use even more registers. -heatsink >> >> Sorry, I couldn't find the rest of the preceding message. Someone >> wrote that they had to turn down cc flags to get SHA1.hs to compile on >> IA64. > > Yep. > >> What C compiler was being used, and what were the symptoms? > > GCC. > > As I recall the symptoms were that gcc used more than 32 registers and > then the mangler balked. The reason is that a registerised ia64 build > expects to only use the first 32 registers but does not take any > precautions to make sure that this is the case. It just relies on the > fact that most code coming out of the ghc backend cannot make use of > more than a handful of registers. If gcc does actually use more then the > mangler catches this. We tried some flags to make gcc restrict itself to > a subset of the registers but could not get it to obey. > > Duncan _________________________________________________________________ Talk to your Yahoo! Friends via Windows Live Messenger. Find out how. http://www.windowslive.com/explore/messenger?ocid=TXT_TAGLM_WL_messenger_ya…

1 0

ghci history.
by Venkat Ramanan 21 Aug '08

21 Aug '08

Hi all, I've just started learning haskell. I encountered a strange behaviour in ghci. 1. ghci doesn't seem to do pruning of duplicate expressions. If the same expression is entered twice, both of them are stored in the history instead of just one. Bash uses environmental variables (HISTCONTROL and HISTIGNORE ) to modify this behavior. Is there a way to get the same behaviour in ghci? 2. Does the history of ghci get stored somewhere in a file? That would be useful. I'm using ghc-6.8.2 (ubuntu 8.04). Also, is there an updated ghc package (6.8.3) for ubuntu? If not, I'll try to install the binary. Thanks, Venkat.

2 2

#ghc connection issue
by Claus Reinke 20 Aug '08

20 Aug '08

I seem to be unable to join the ghc chatroom at irc.freenode.net at the moment (using Opera). Is that an issue with my irc client or a general problem? 15:47 Joining chat room... Disconnected from chat Claus

4 4

invoking a Haskell script without a .hs extension
by prj＠po.cwru.edu 18 Aug '08

18 Aug '08

I have a Haskell script called "notify", without a .hs extension, which causes some problems. (I'm using ghc 6.8.3.) First attempt: runhaskell notify Without the .hs extension, ghc doesn't know it's a Haskell script, and so I get "Could not find module `notify'". Maybe runhaskell should automatically add "-x hs" to the ghc command? Second attempt: runhaskell -x hs notify This get me "Not in scope: `main'". runhaskell is invoking ghc with these arguments: -ignore-dot-ghci -x -e ':set prog "hs"' -e ':main ["notify"]' hs So it looks like runhaskell it treating "-x" as an argument to be relayed to ghc, "hs" as the name of the script, and "notify" as an argument to the script. I guess I need to use "--" to make it clear where the ghc arguments end and where the script and its arguments begin. Third attempt: runhaskell -- -x hs -- notify This gets me "Not in scope: `main'" again. runhaskell is invoking ghc with these arguments: -ignore-dot-ghci -x -e ':set prog "hs"' -e ':main ["--","notify"]' hs This looks like a bug in the "--" handling, unless I'm misinterpreting the usage message I get from running plain "runhaskell". Fourth attempt: ghc -ignore-dot-ghci -e ':set prog "notify"' \ -e ':main []' -x hs notify This works, but passing arguments becomes rather cumbersome. If there's a way to get runhaskell to pass "-x hs" in the right place, that would be much better. A somewhat related issue: I'd like to avoid hard-coding the path to runhaskell or ghc in the #! line. Instead, I'd like to use #!/bin/sh, and have the shell find runhaskell or ghc in $PATH. This means the first few lines of the script would have to be executed by sh, but ignored by ghc. Most scripting langauges have some way of doing this, although it's often by accident. The best way I've seen is in Guile, where "#!" starts a multi-line comment, and "!#" ends it. For Haskell, this is the best I could come up with: #!/bin/sh {- > /dev/null 2>&1 exec ghc -ignore-dot-ghci \ -e ":set prog \"$0\"" \ -e ':main []' -x hs "$0" -} But this depends on "{-" not existing as an executable command, or at least not doing anything harmful when invoked like this. I'd like to avoid depending on anything like that. Does anyone have any better ideas? paul

4 5

Any release date for GHC 6.8.3 for Mac OS X 10.5 (Leopard) on PowerPC G4 machines?
by Benjamin L.Russell 18 Aug '08

18 Aug '08

Is there any scheduled release date for GHC 6.8.3 for Mac OS X 10.5 (Leopard) on PowerPC G4 machines? I have a G4 PowerBook, currently running Mac OS X 10.2.8 (Jaguar), which I plan to upgrade next week. I also have the upgrade discs for both Mac OS X 10.4 (Tiger) and Mac OS X 10.5 (Leopard). Backing up my hard disk before upgrading takes about 20 hours; it seems a pity to upgrade to Tiger before eventually upgrading to Leopard (unless GHC is never going to be released for Tiger on PowerPC G4 machines, which would be much worse). -- Benjamin L. Russell

2 5

problems running ghc-6.8.2 on solaris 10, sparc
by Daniil Elovkov 16 Aug '08

16 Aug '08

Hello folks, Christian I'm trying to get ghc 6.8.2 running on Solaris 10 and having problems. To be precise, I'm trying to compile a 'hello world' program by ghc 6.8.2 which I got in binary form haskell.org. gcc is 2.95, it uses sun linker. I remember there were problems with that in the past. Is ghc supposed to work only with gnu ld or sun ld as well? So, how it went first I got compiler errors in many places of Reg.hs: global register variable follows a function definition Googling showed that Don Stewart used to fix it by swapping 2 includes in Stg.h - putting MachRegs.h after Regs.h instead of before. It helped Then there was assembler error: cannot use v8plus instructions in a non-v8plus target binary It was caused by -mcpu=v9, which ghc passes to gcc. I blindly added -optc -mcpu=v8 and it helped :) Then the linker complained that it could not resolve aio_fork and __aio_suspend64, referenced from librt.so. -lrt is passed by ghc to the linker. On this machine there is /lib/libaio.so. Linking with it didn't help. It doesn't really contain exactly those functions, only with slightly different names, like _aio_forkinit, _libaio_fork. Also, librt which was linked in was the one lying close to where gcc is installed. Apart from that, there is also librt.so in /lib. I thought, maybe the wrong librt was used and said -optl -L/lib to link against the one in /lib. No complaints, but the resulting binary segfaults. Does anybody have any ideas? -- Daniil Elovkov

4 5