filetests sometimes do not set _ #9069
Comments
From Mark@Overmeer.netCreated by mark@overmeer.netOn the Perl5.10.0 of today (patch 32115), I see filetest errors which Somewhere in a large program, I do something like this: sub ... { And the result is "directory"!! with -o e z s f d u g k b c l T B M A C, it is reported as file. I have tried, but was not successful in reproducing the error in a Greetings, MarkOv Perl Info |
From Mark@Overmeer.net* Mark Overmeer (perlbug-followup@perl.org) [071016 14:04]:
Hum... the error report was even more right than I expected it to be... The following script reports 'directory' twice: #!/usr/local/bin/perl5.10.0 -d '/etc'; Apparently, -r/-w/-x/-R/-W/-X implemented in filetest to not Mark Overmeer MSc MARKOV Solutions |
From @rgsOn 17/10/2007, Mark Overmeer <mark@overmeer.net> wrote:
Looking at the code, I think this bug was already present in 5.8.x. What does |
|
The RT System itself - Status changed from 'new' to 'open' |
From Mark@Overmeer.net* Rafael Garcia-Suarez (rgarciasuarez@gmail.com) [071017 08:49]:
markov@earth:&perl> perl5.10.0 -V:.*access.* markov@earth:&perl> perl5.8.7 -V:.*access.* The code which now fails was ran by cpan-testers on many many According to the cpantesters results By the way: quite some 5.9.5 tests reported successes, but actually Mark Overmeer MSc MARKOV Solutions |
From @andk
> The code which now fails was ran by cpan-testers on many many > According to the cpantesters results Binary search reveals the test fails since 25986 Change 25986 by nicholas@nicholas-saigo on 2005/11/04 13:02:42 ftrwrite, ftrexec, fteread, ftewrite and fteexec can all be merged -- |
From @nwc10On Thu, Oct 18, 2007 at 05:29:36AM +0200, Andreas J. Koenig wrote:
Mmm, I'm not sure about this. The script starts failing for me on 5.8.1 $ cat 46463 -d '/etc'; It's also failing on 5.6.2: $ /usr/local/perl/5.6.2/bin/perl ~/p4perl/perl/46463 Nicholas Clark |
From @andk
> On Thu, Oct 18, 2007 at 05:29:36AM +0200, Andreas J. Koenig wrote:
> Mmm, I'm not sure about this. The script starts failing for me on 5.8.1 > $ cat 46463 > -d '/etc'; Ahh, this is a different bug then. Mark told us that he could not ----Program---- ----Output of .../plC1Mru/perl-5.8.0@25985/bin/perl---- ----EOF ($?='0')---- ----EOF ($?='0')---- The N/A comes from the fact that the test suite failed. Whereas your program says "directory" on both 2598[56]. A binary search for your program leads us to the interval between -- |
From Mark@Overmeer.net* Andreas J. Koenig (andreas.koenig.7os6VVqR@franz.ak.mind.de) [071018 19:46]:
You may be very right. This script does shows the same problem and Remove the 'use filetest', or use f.i. -s i.s.o -w does solve the In the program the situation is far more complex: the '-d' is run I have tried and tried, but no luck on a small example which does unpack Mail::Box-*gz tests 3 and 4 fail. They call Mail::Box::Mbox::listSubFolders(). My hopes are that a fix for above bug will fix the second. MarkOv Mark Overmeer MSc MARKOV Solutions |
From @andk
> A binary search for your program leads us to the interval between ---Program---- -d '/etc'; ----Output of .../pwzF3Lg/perl-5.8.0@19391/bin/perl---- ----EOF ($?='0')---- ----EOF ($?='0')---- Change 19392 by jhi@kosh on 2003/05/03 05:33:04 Salvage 'use filetest "access"' from compiletime to runtime. -- |
From @nwc10On Wed, Oct 17, 2007 at 10:22:58AM +0200, Mark Overmeer wrote:
On Fri, Oct 19, 2007 at 02:20:26AM +0200, Andreas J. Koenig wrote:
It seems that Jarkko actually fixed a bug here - I'd infer that access() wasn't I don't believe that the current behaviour is a behavioural bug. It is documented that _ is set to the last stat. Hence one does not need to call stat(), so perl does not. *That* So I think that the current behaviour is preferable. If we changed it so that I'm curious - on a modern Unix system with decent OS caching, how much does Nicholas Clark |
From Mark@Overmeer.net* Nicholas Clark (nick@ccl4.org) [071021 20:53]:
Oh, that's a good reason for that patch, and does explain things.
Well, perldoc -f stat says: if "stat" is passed the special filehandle consisting of an I do read "filetest" here... (perl5.10.0)
On many places, Perl prefers qualilty over performance. I would even That file test behavioral change when 'use filetest' is given somewhere On the other hand, the whole use of -w / -r /-x should be avoided in
Probably only an extra context switch. (Do you known how many a simple Mark Overmeer MSc MARKOV Solutions |
From Mark@Overmeer.netGuys, What will happen with the reported bug? I don't like the silence IMO, we should decide between: (1) change the docs, reflecting the special behavior of (2) produce an error when _ is used when filetest is enabled. (3) change the internal implementation of -r/-w/-x to use IMO, (3) [filetest is always used] is strongly preferred: if you test But this probably shouldn't go into 5.10... At least the first
* Mark Overmeer (mark@overmeer.net) [071022 00:47]:
MarkOv Mark Overmeer MSc MARKOV Solutions |
From @rgsOn 02/11/2007, Mark Overmeer <mark@overmeer.net> wrote:
That's the best we can do for 5.10.0 I think. A good place to patch |
From Mark@Overmeer.net* Rafael Garcia-Suarez (rgarciasuarez@gmail.com) [071102 15:27]:
Just at the last moment, two proposed updates to perlfunc/-X and Mark Overmeer MSc MARKOV Solutions |
From Mark@Overmeer.netInline Patch--- pod/perlfunc.pod.old 2007-11-30 11:19:32.000000000 +0100
+++ pod/perlfunc.pod 2007-11-30 11:33:46.000000000 +0100
@@ -341,13 +341,16 @@
#...
}
-The interpretation of the file permission operators C<-r>, C<-R>,
-C<-w>, C<-W>, C<-x>, and C<-X> is by default based solely on the mode
-of the file and the uids and gids of the user. There may be other
-reasons you can't actually read, write, or execute the file. Such
-reasons may be for example network filesystem access controls, ACLs
-(access control lists), read-only filesystems, and unrecognized
-executable formats.
+The interpretation of the file permission operators C<-r>, C<-R>, C<-w>,
+C<-W>, C<-x>, and C<-X> is by default based solely on the mode of the
+file and the uids and gids of the user. There may be other reasons you
+can't actually read, write, or execute the file: for example network
+filesystem access controls, ACLs (access control lists), read-only
+filesystems, and unrecognized executable formats. The use of these six
+specific operators is usually a mistake in the first place, because it
+may be open to race conditions. See the filetest pragma manual L<filetest>
+for a better approach and more details, but be warned that the filetest
+pragma may cause C<_> not to be set!
Also note that, for the superuser on the local filesystems, the C<-r>,
C<-R>, C<-w>, and C<-W> tests always return 1, and C<-x> and C<-X> return 1
@@ -355,16 +358,6 @@
may thus need to do a stat() to determine the actual mode of the file,
or temporarily set their effective uid to something else.
-If you are using ACLs, there is a pragma called C<filetest> that may
-produce more accurate results than the bare stat() mode bits.
-When under the C<use filetest 'access'> the above-mentioned filetests
-will test whether the permission can (not) be granted using the
-access() family of system calls. Also note that the C<-x> and C<-X> may
-under this pragma return true even if there are no execute permission
-bits set (nor any extra execute permission ACLs). This strangeness is
-due to the underlying system calls' definitions. Read the
-documentation for the C<filetest> pragma for more information.
-
Note that C<-s/a/b/> does not do a negated substitution. Saying
C<-exp($foo)> still works as expected, however--only single letters
following a minus are interpreted as file tests. |
From @jhihttp://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2007-11/msg01085.html In principle I approve clarifications but it seems to me that you are |
From Mark@Overmeer.net* Jarkko Hietaniemi (jhi@iki.fi) [071130 12:55]:
Please be specific Mark Overmeer MSc MARKOV Solutions |
From @jhiWhat exactly are you trying to add / change? Something about the On Nov 30, 2007 8:11 AM, Mark Overmeer <mark@overmeer.net> wrote:
-- |
From Mark@Overmeer.net
* Jarkko Hietaniemi (jhi@iki.fi) [071130 14:06]:
The changed text should reflect the debates on the porters lists, which in short -r $filename no filetest 'access' # default use filetest 'access' Under the filetest pragma, -r/-w/-x do not set '_'. There was no IMO, filetest 'access' should be default, and we should try to trick The patches I provide include two major changes. Changes to the text are (of course) possible. At least, this suggestion MarkOv Mark Overmeer MSc MARKOV Solutions |
From @nwc10On Fri, Nov 30, 2007 at 05:05:31PM +0100, Mark Overmeer wrote:
Changing code always has the possibility of adding a bug. This should be
My opinion is that forcing everyone to take a speed hit to protect against I assume that it would be possible for the implementation of the filetest It, presumably, would also be possible to have the filetest pragma be able Nicholas Clark |
From Mark@Overmeer.net* Nicholas Clark (nick@ccl4.org) [071203 14:52]:
It is documented in with the access pragma.
Depends on the hit. Doing a stat() is mainly a wait-for-disk. A stat() A speed it of a few percent (extra context switch maybe) is not bad.
Of course, possible. Not so easy to get right.
I am always proud on Perl, that it does not bother people with string MarkOv Mark Overmeer MSc MARKOV Solutions |
From @hvdsNicholas Clark <nick@ccl4.org> wrote: It need not be a stat()-sized speed hit if after an access we set a flag That reduces the universal hit to checking a flag on access of C<_>, and Hugo |
From Mark@Overmeer.net* hv@crypt.org (hv@crypt.org) [071204 13:27]:
With all these solutions, you introduce a danger -r $filename; The more I think about it, the more convinced I am that -r/-w/-x in any Maybe, we should move into the direction where we add the provide The -r/-w/-x should than be documented as Mark Overmeer MSc MARKOV Solutions |
From @rgsOn 05/12/2007, Mark Overmeer wrote:
One of the goals (in my mind) of introducing stacked filetest operators
Oh, no, not File::Spec for that !
That wouldn't solve the race condition problem. I think the only sure |
From Mark@Overmeer.net* Rafael Garcia-Suarez (rgarciasuarez@gmail.com) [071205 10:33]:
Well, I think that the module is outdated as well... but it looks
This is very true... although there are situations that the readability if(open F, '<', $fn) but it is nicer to be able to do if(is_readable $fn) For instance, the auto-completion of bash only suggests files which can MarkOv Mark Overmeer MSc MARKOV Solutions |
From @rgsOn 30/11/2007, Mark Overmeer <mark@overmeer.net> wrote:
Thanks, I've applied modified/augmented/rephrased versions of those to |
|
@smpeters - Status changed from 'open' to 'resolved' |
From solutions@overmeer.net* Steve Peters via RT (perlbug-followup@perl.org) [081107 20:39]:
Well, if you declare that "documenting a problem" is equivalent to It could be solved by calling stat() also when in "use filetest 'access';" Mark Overmeer MSc MARKOV Solutions |
Migrated from rt.perl.org#46463 (status was 'resolved')
Searchable as RT46463$
The text was updated successfully, but these errors were encountered: