GHC : ezyang's blog

GHC

BlockedIndefinitelyOnMVar July 27, 2011

This post was adapted from a post I made to the glasgow-haskell-users list.

According to Control.Exception, the BlockedIndefinitelyOnMVar exception (and related exception BlockedIndefinitelyOnSTM) is thrown when “the thread is blocked on an MVar, but there are no other references to the MVar so it can’t ever continue.” The description is actually reasonably precise, but it is easy to misinterpret. Fully understanding how this exception works requires some extra documentation from Control.Concurrent as well as an intuitive feel for how garbage collection in GHC works with respects to Haskell’s green threads.

In-program GC stats for GHC July 20, 2011

I’ll be at this year’s Hac Phi (coming up in a week and a half), and I am planning on working on in-program garbage collector statistics for GHC. There is nothing really technically difficult about this task (we just need to expose some functions in the RTS), but it’s not been done yet and I know quite a few performance-minded and long-running-server-minded folks who would love to see this functionality.

My question for you is this: what would you like such an API to look like? What things should it offer, how would you like to interact with it?

Tracing the compilation of Hello Factorial! April 13, 2011

It is often said that the factorial function is the “Hello World!” of the functional programming language world. Indeed, factorial is a singularly useful way of testing the pattern matching and recursive facilities of FP languages: we don’t bother with such “petty” concerns as input-output. In this blog post, we’re going to trace the compilation of factorial through the bowels of GHC. You’ll learn how to read Core, STG and Cmm, and hopefully get a taste of what is involved in the compilation of functional programs. Those who would like to play along with the GHC sources can check out the description of the compilation of one module on the GHC wiki. We won’t compile with optimizations to keep things simple; perhaps an optimized factorial will be the topic of another post!

GHC migrating to Git March 30, 2011

From cvs-ghc@haskell.org:

Hi all,

We now plan to do the git switchover this Thursday, 31 March.

Thanks
Ian

There are some things that I will miss from Darcs (darcs send and the cases where “everything is a patch” actually does work well), but all and all I’m quite pleased to see GHC moving to Git.

Type Technology Tree March 7, 2011

They say that one doesn’t discover advanced type system extensions: rather, the type system extensions discover you! Nevertheless, it’s worthwhile to know what the tech tree for GHC’s type extensions are, so you can decide how much power (and the correspondingly headache inducing error messages) you need. I’ve organized the relations in the following diagram with the following criterion in mind:

Some extensions automatically enable other extensions (implies);
Some extensions offer all the features another extension offers (subsumes);
Some extensions work really nicely with other extensions (synergy);
Some extensions offer equivalent (but differently formulated) functionality to another extension (equiv).

It’s also worth noting that the GHC manual divides these extensions into “Extensions to data types and type synonyms”, “Class and instances declarations”, “Type families” and “Other type system extensions”. I have them organized here a little differently.

Multi-day debugging February 25, 2011

Most of my hacking cycles right now are going towards debugging the new code generator for GHC. The code generation stage of GHC takes the Spineless Tagless G-machine (STG) intermediate representation (IR) to the C– high-level assembly representation; the old code generator essentially performed this step in one big bang. The new code generator is many things. It is a more modular, understandable and flexible codebase. It is a client of cutting edge research in higher-order frameworks for control-flow optimization.

It's just a longjmp to the left November 15, 2010

And then a signal to the ri-i-i-ight.

One notable wart with readline is that if you ^C during the prompt, nothing happens, and if you do a second ^C (and weren’t buggering around with the signal handlers) your entire program unceremoniously terminates. That’s not very nice! Fortunately, readline appears to be one of the rare C libraries that actually put some work into making sure that you could longjmp out of a signal handler and not completely break the library’s internal state (they do this with liberal masking and unmasking, and their own signal handler which cleans up and then rethrows the signal).

You are in a maze of twisty little passages, all alike... (a GHC hacking post) November 12, 2010

About a month ago I decided that it would be cool if I could solve the bug GHC’s runtime never terminates unused worker threads. Well, I just got around to looking at it today, and after wandering aimlessly around the twisty maze that is the GHC RTS for an hour or so, I finally found a light at the end of a tunnel, in the form of a heart-warmingly simple patch. I’ve sent mail off to Simon Marlow to make sure the light isn’t actually a train, but it occurred to me that it would be interesting to look at my command history and blog about the process by which I came to the conclusion that line 464 of Capability.c was the correct place to add my change, since this sort of mental journey is not the one that is really ever recorded anywhere in any shape or form.

Towards platform-agnostic interruptibility September 10, 2010

Last post, I talked about some of the notable difficulties in emulating pthread_cancel on Windows. Today, I want to talk about what a platform agnostic compiler like GHC actually ought to do. Recall our three design goals:

GHC would like to be able to put blocking IO calls on a worker thread but cancel them later; it can currently do this on Linux but not on Windows,
Users would like to write interrupt friendly C libraries and have them integrate seamlessly with Haskell’s exception mechanism, and
We’d like to have the golden touch of the IO world, instantly turning blocking IO code into nice, well-behaved, non-blocking, interruptible code.

I am going to discuss these three situations, described concisely as blocking system calls, cooperative libraries and blocking libraries. I propose that, due to the lack of a cross-platform interruption mechanism, the correct interruptibility interface is to permit user defined handlers for asynchronous exceptions.

pthread_cancel on Windows September 8, 2010

Edward, I’m afraid I have some bad news. Your interruptible GHC patch; it was involved in a terrible accident on the way to Windows portability. I hope you understand: we’re doing our best to patch it up, but there have been some complications…

Pop quiz! What does this pthreads code do? :

#include <pthread.h>
#include <stdio.h>

void *thread1(void *arg) { sleep(10000); }
void *thread2(void *arg) { while (1) {} }

void *psycho_killer(void *arg) {
    pthread_t *id = (pthread_t*)arg;
    pthread_cancel(*id);
    printf("[%p] Psycho killer...\n", id);
    pthread_join(*id, NULL);
    printf("[%p] ...qu'est-ce que c'est.\n", id);
}

int main(char* argv, int argc) {
    pthread_t t1, t2, k1, k2;
    pthread_create(&t1, NULL, thread1, NULL);
    printf("[%p] I can't sleep 'cause my bed's on fire\n", &t1);
    pthread_create(&t2, NULL, thread2, NULL);
    printf("[%p] Don't touch me I'm a real live wire\n", &t2);
    pthread_create(&k1, NULL, psycho_killer, &t1);
    pthread_create(&k2, NULL, psycho_killer, &t2);
    pthread_join(k1, NULL);
    pthread_join(k2, NULL);
    printf("Run run run away!\n");
    return 0;
}

It never manages to terminate the second thread…

« Newer Posts Older Posts »