hClose003 test failes on Solaris/OpenBSD on i386/x86 host

[kgardas]

Hello, it seems hClose003 expected output is kind of Linux specific hence it fails on both Solaris and OpenBSD running on x86 host where I've been able to test this. Anyway, both OSes are not that sensitive to closing pipe read end as Linux seems to be. I also expect that other Unix(-like) systems will behave more like Solaris/OpenBSD and so the behavior of Linux is specific. Hence I'm proposing attached patch which moves currect stdout to new stdout-i386-unknown-linux and provide new stdout with the Solaris/OpenBSD result. I would suggest to commit this and allow it to test on FreeBSD/Darwin as this seems to be available to GHC team as builder machines... Thanks, Karel

[kgardas]

Proposed fix for this issue

[igloo]

The test passes on Linux and OS X, and fails on Solaris (sparky).

This C program:

#include <sys/wait.h>
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <signal.h>
#include <errno.h>

int main(void) {
    int pipefd[2];
    int r;

    signal(SIGPIPE, SIG_IGN);

    if (pipe(pipefd) == -1) {
        perror("pipe");
        exit(1);
    }

    close(pipefd[0]);
    r = write(pipefd[1], "foobar", 3);
    printf("r is %d; errno is %d; %s\n", r, errno, strerror(errno));
    return 0;
}

gives me the same output on both Linux and Solaris, though:

r is -1; errno is 32; Broken pipe

Am I missing something, or is there a bug in the Haskell code on Solaris?

[simonmar]

Replying to igloo:

The test passes on Linux and OS X, and fails on Solaris (sparky). This C program: {{{ #include <sys/wait.h> #include <assert.h> #include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <string.h> #include <signal.h> #include <errno.h> int main(void) { int pipefd[2]; int r; signal(SIGPIPE, SIG_IGN); if (pipe(pipefd) == -1) { perror("pipe"); exit(1); } close(pipefd[0]); r = write(pipefd[1], "foobar", 3); printf("r is %d; errno is %d; %s\n", r, errno, strerror(errno)); return 0; } }}} gives me the same output on both Linux and Solaris, though: {{{ r is -1; errno is 32; Broken pipe }}} Am I missing something, or is there a bug in the Haskell code on Solaris?

I don't see an obvious bug in the Haskell code. Any idea what the bug is?

[igloo]

Aha, looks like it's this:

Wed Dec  1 13:08:47 GMT 2010  Simon Marlow <marlowsd@gmail.com>
  * fix a discarded exception in hClose

Fixed in HEAD and 7.0.