bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Ken Brown]

Starting with commit 3a6df2d,
tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin.  The
reason is that inode numbers tend to be very large, on the order of
10^17, at least on my systems.  As a result, the "convert inode" code in
tramp-convert-file-attributes always falls back to tramp-get-inode.
Since this increments by 1 on every call, the file-attribute comparisons
in tramp-test19-directory-files-and-attributes always fail.

The significance of commit 3a6df2d is that it makes
tramp-get-remote-stat return a non-nil value, so that tramp uses 'stat'
for file attributes; this returns the true (large) inode number.  Prior
to that commit, tramp-get-remote-stat would always return nil, so that
the tramp-perl-file-attributes script would be used for file
attributes.  This script uses whatever inode number is returned by
Perl's version of 'stat', which is presumably some sort of 32-bit
truncation of the true inode number.

Although I'm seeing this on 32-bit Cygwin, it seems to me that it could
in principle happen on any 32-bit platform.

Ken




In GNU Emacs 26.0.50 (build 11, i686-pc-cygwin, GTK+ Version 3.18.9)
  of 2017-02-23 built on desktop-new
Repository revision: 16efea3a883ebf633946ee9b9d0681eb55437878
Windowing system distributor 'The Cygwin/X Project', version 11.0.11901000

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Eli Zaretskii]

> From: Ken Brown <kbrown@cornell.edu>
> Date: Thu, 23 Feb 2017 18:19:50 -0500
> 
> Starting with commit 3a6df2d,
> tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin.  The
> reason is that inode numbers tend to be very large, on the order of
> 10^17, at least on my systems.  As a result, the "convert inode" code in
> tramp-convert-file-attributes always falls back to tramp-get-inode.
> Since this increments by 1 on every call, the file-attribute comparisons
> in tramp-test19-directory-files-and-attributes always fail.
> 
> The significance of commit 3a6df2d is that it makes
> tramp-get-remote-stat return a non-nil value, so that tramp uses 'stat'
> for file attributes; this returns the true (large) inode number.  Prior
> to that commit, tramp-get-remote-stat would always return nil, so that
> the tramp-perl-file-attributes script would be used for file
> attributes.  This script uses whatever inode number is returned by
> Perl's version of 'stat', which is presumably some sort of 32-bit
> truncation of the true inode number.
> 
> Although I'm seeing this on 32-bit Cygwin, it seems to me that it could
> in principle happen on any 32-bit platform.

Would it work to convert the large inode values to a float?

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Eli Zaretskii]

> Date: Fri, 24 Feb 2017 09:55:33 +0200
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: 25854@debbugs.gnu.org
> 
> Would it work to convert the large inode values to a float?

Sorry, this would be sub-optimal.  Instead, how about using the same
format as file-attributes uses for large inode numbers?  Michael?

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Michael Albinus]

Eli Zaretskii <eliz@gnu.org> writes:

> Sorry, this would be sub-optimal.  Instead, how about using the same
> format as file-attributes uses for large inode numbers?  Michael?

Tramp does something like this already, in
`tramp-do-file-attributes-with-ls'.

Ken, could you pls set `tramp-verbose' to 6, and rerun the test? With
the debug buffer as example, I will try to implement something similar
for your case.

Best regards, Michael.

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Ken Brown]

[-- Attachment #1: Type: text/plain, Size: 236 bytes --]

On 2/24/2017 9:04 AM, Michael Albinus wrote:
> Ken, could you pls set `tramp-verbose' to 6, and rerun the test? With
> the debug buffer as example, I will try to implement something similar
> for your case.

Message log attached.

Ken


[-- Attachment #2: test19.debug.xz --]
[-- Type: application/octet-stream, Size: 6148 bytes --]

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Ken Brown]

[-- Attachment #1: Type: text/plain, Size: 366 bytes --]

On 2/24/2017 9:49 AM, Ken Brown wrote:
> On 2/24/2017 9:04 AM, Michael Albinus wrote:
>> Ken, could you pls set `tramp-verbose' to 6, and rerun the test? With
>> the debug buffer as example, I will try to implement something similar
>> for your case.
>
> Message log attached.

Sorry, I sent the ert messages rather than the debug buffer.  Here's the 
latter.

Ken


[-- Attachment #2: test19.debug_buffer.xz --]
[-- Type: application/octet-stream, Size: 6188 bytes --]

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Michael Albinus]

Ken Brown <kbrown@cornell.edu> writes:

Hi Ken,

>> Message log attached.
>
> Sorry, I sent the ert messages rather than the debug buffer.  Here's
> the latter.

Thanks for both. I have sufficient data now; I'll look into this over
the weekend.

> Ken

Best regards, Michael.

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Michael Albinus]

Ken Brown <kbrown@cornell.edu> writes:

Hi Ken,

>>> Ken, could you pls set `tramp-verbose' to 6, and rerun the test? With
>>> the debug buffer as example, I will try to implement something similar
>>> for your case.
>>
>> Message log attached.
>
> Sorry, I sent the ert messages rather than the debug buffer.  Here's
> the latter.

I've pushed a patch to master, which should fix this. Could you, pls, check?

> Ken

Best regards, Michael.

bug#25854: 26.0.50; tramp-test19-directory-files-and-attributes fails on 32-bit Cygwin

[Ken Brown]

On 2/25/2017 9:05 AM, Michael Albinus wrote:
> I've pushed a patch to master, which should fix this. Could you, pls, check?

That fixes it.  Thanks.

I'm closing the bug.

Ken