print doesn't overwrite $! #8919
Comments
From perlbug@der-pepe.deCreated by perlbug@der-pepe.deThis is a bug report for perl from perlbug@der-pepe.de, ----------------------------------------------------------------- open STDOUT, '>/dev/full' or die $!; This script should print "No space left on device" three times (because this is No space left on device at bug.pl line 4. It seems that once "open" sets $! to "No such file or directory", "print" (The script only runs on Unix variants because of the /dev/full thing.) Perl Info |
From perlbug@der-pepe.deI've tried strace-ing this, and I found that buffering somehow gets write(1, "x", 1) = -1 ENOSPC (No space left on device) Even if I include $|=1 in the loop just before "print", the "xx" is not printed |
From @TuxOn Thu, 31 May 2007 18:39:08 -0700, "perlbug@der-pepe.de (via RT)"
from perldoc perlvar: the key here is *only immediately after the failure*. The last *system*
-- |
|
The RT System itself - Status changed from 'new' to 'open' |
From @demerphqOn 6/1/07, H.Merijn Brand <h.m.brand@xs4all.nl> wrote:
Since stdout is autoflushing each call to print should result in a yves -- |
From mail@der-pepe.de
This bug is still in perl 5.8.8 and 5.10.0 and blead. I think I have fixed this bug. After giving this careful thought, I The problem is: pp_print calls do_print. If that is successful *and* the file-handle is This has another implication: When you get an error from print and then I've also added tests for both the $! thing and the empty-string thing. The tests don't use /dev/full as in the original report because I don't Regards, |
From @gisleOn Aug 18, 2008, at 18:40, Christoph Bussenius via RT wrote:
Is there any significance to the 'open my $f' statement here?
I don't think the patch is acceptable. If print fails the error open(my $fh, ">", $file) || die "Can't open With your patch the first print could fail and then the second could --Gisle
|
From lists@der-pepe.deOn Tue, Aug 19, 2008 at 09:45:24AM +0200, Gisle Aas wrote:
Yes, it sets $! to ENOENT (No such file or directory), to demonstrate
(This was my original report for bug #43097).
Hm, I see your point. I'll think about it, thanks. Christoph |
From lists@der-pepe.de
I've added a new patch below. Instead of clearing the error flag, it On Tue, Aug 19, 2008 at 09:45:24AM +0200, Gisle Aas wrote:
This piece of code is still a bit problematic in my opinion. In close's I think the only way to get this to work properly is to save errno along Regards, Inline Patch--- perl-5.10.0/doio.c 2007-12-18 11:47:07.000000000 +0100
+++ perl-5.10.0-debug/doio.c 2008-09-08 12:32:57.325273275 +0200
@@ -1247,7 +1247,7 @@
if (len && (PerlIO_write(fp,tmps,len) == 0))
happy = FALSE;
Safefree(tmpbuf);
- return happy ? !PerlIO_error(fp) : FALSE;
+ return happy;
}
}
--- perl-5.10.0/t/io/print.t 2007-12-18 11:47:08.000000000 +0100
+++ perl-5.10.0-debug/t/io/print.t 2008-08-18 17:49:05.093301791 +0200
@@ -6,10 +6,10 @@
}
use strict 'vars';
-eval 'use Errno';
+eval 'use Errno; use IO::Handle';
die $@ if $@ and !$ENV{PERL_CORE_MINITEST};
-print "1..21\n";
+print "1..24\n";
my $foo = 'STDOUT';
print $foo "ok 1\n";
@@ -65,3 +65,31 @@
map print(+()), ('')x68;
print "ok 21\n";
}
+
+# Create a closed pipe to test print's error reporting
+# See bug #43097
+if (exists $IO::Handle::{autoflush}) {
+ if (pipe my $rpipe, my $wpipe) {
+ eval { $SIG{PIPE} = 'IGNORE'; };
+ close $rpipe or die $!;
+ $wpipe->autoflush(1);
+
+ # printing with autoflush to a closed pipe must fail
+ print $wpipe "xy" and print "not ";
+ print "ok 22\n";
+
+ open my $h, '< /nonexistent' and die '/nonexistent exists';
+
+ print $wpipe "xy";
+ print 'not ' if $!{ENOENT};
+ print "ok 23\n";
+
+ # Printing an empty string must not fail
+ print $wpipe '' or print 'not ';
+ print "ok 24\n";
+ } else {
+ print "ok $_ # skipped: cannot pipe ($!)\n" for 22..24;
+ }
+} else {
+ print "ok $_ # skipped: no autoflush\n" for 22..24;
+} |
From @tonycozOn Mon Sep 08 04:07:40 2008, lists@der-pepe.de wrote:
I believe this change is dangerous under the following circumstances: - no autoflush to a full filesystem This produces output with some content missing. A patch that saves errno per file handle would be less dangerous, but I Tony |
Migrated from rt.perl.org#43097 (status was 'open')
Searchable as RT43097$
The text was updated successfully, but these errors were encountered: