Austin Group Defect Tracker

Aardvark Mark IV

Main | My View | View Issues | Change Log | Docs
  

Viewing Issue Simple Details [ Jump to Notes ] [ Issue History ] [ Print ]
ID Category Severity Type Date Submitted Last Update
0000478 [1003.1(2004)/Issue 6] Shell and Utilities Objection Clarification Requested 2011-07-21 11:07 2019-06-10 08:55
Reporter geoffclare View Status public  
Assigned To ajosey
Priority normal Resolution Accepted As Marked  
Status Closed  
Name Geoff Clare
Organization The Open Group
User Reference
Section cmp
Page Number 253
Line Number 9875
Interp Status Approved
Final Accepted Text See Note: 0000944
Summary 0000478: Should cmp -s write to stderr when an error occurs?
Description The description of the cmp -s option is:

    Write nothing for differing files; return exit status only.

Clearly the intention is that "return exit status only" when taken
in context is supposed to apply only to the case of differing
files. It is saying that, given that nothing is written, the only
thing that indicates the files are different is the exit status.
However, it could (at a stretch) be misinterpreted as a stand-alone
statement that extends the requirement.

The intended meaning is confirmed by the statements in the STDOUT
section, "No output shall be written to standard output when the -s
option is used", and in the STDERR section, "If file1 and file2 are
identical for the entire length of the shorter file, in the POSIX
locale the following diagnostic message shall be written, unless
the -s option is specified" (with no other mention of -s).

I.e. the intention is that nothing should be written to stdout or
stderr when the exit status is 1, but an error message should still
be written to stderr when the exit status is greater than 1.
(The intention is the same in Issue 7 although the text in the STDERR
section differs as a result of a change related to the -l option.)

However, existing practice (on the systems I have tried) is that -s
also turns off some, or possibly all, error messages related to
opening files. (I tried a non-existent file and an unreadable file.)
I also checked behaviour for a read error and got three different
results:

1. The error message produced without -s was turned off by -s.

2. The error message was written regardless of whether -s was used.

3. The read error was treated as EOF, which I would say is a bug.

The read() error behaviour was tested by causing an EIO using:

    (trap "" TTIN; cmp [-s] - somefile < /dev/tty) &

in a shell with job-control enabled. When I got result 3 I confirmed
that the read() call made by cmp did return an EIO error, by using a
system call trace utility.

Typically when a difference between the standard and existing practice
is found, and nothing in the rationale indicates that it was deliberate,
the standard is relaxed to allow existing practice. However, for cmp -s
I believe there are reasonable grounds for affirming the current
requirement. The -s option, as described in the standard, produces a
behaviour that is awkward to achieve by other means. For example, this
function could be used:
    silent_cmp() {
        # compare files with no output except error messages
        message=$(cmp "$@" 2>&1 >/dev/null)
        status=$?
        case $status in
        (0|1) ;;
        (*) printf '%s\n' "$message" ;;
        esac
        return $status
    }

By contrast, complete silence is easily achieved using simple
redirection to /dev/null.

I have produced detailed proposed changes for the two most likely
options. A third middle-ground option could also be considered
where -s is allowed to suppress error messages for non-existent
or unreadable files but not for other errors, since applications
can easily check for those conditions using "test -r". This would
mean systems that produce result type 2 above might not need to
change (depending on how they handle other open() errors such as
ENFILE).
Desired Action Option 1: affirm current requirement

Issue a "standard is clear, standard is right" interpretation,
including rationale that the phrase "return exit status only" in
the description of -s, in context, applies only to the case of
differing files. The Notes to the Editor should be:

The description of -s is considered to be unambiguous, but the
following changes are proposed for a future Issue 7 Technical
Corrigendum (or for Issue 8 if there are no further TCs) to ensure
it is not misinterpreted.

On L9875 (L81246 in Issue 7) change the -s description from:

    Write nothing for differing files; return exit status only.

to:

    Write nothing to standard output or standard error when files
    differ; indicate differing files through exit status only.

After L9964 (L81334 in Issue 7) append to RATIONALE:

    Historically, the -s option prevented some or all error messages
    from being written to standard error, not just the "EOF" message.
    With this behavior, -s was not very useful in portable
    applications; the only way to ensure consistent behavior of cmp
    when -s was used was to redirect standard error to /dev/null.
    The standard requires that -s only suppresses the "EOF" message
    to standard error (and all output to standard output). If an
    error is encountered, as indicated by an exit status greater
    than 1, a diagnostic message must still be written. This change
    from historical practice is considered to be justified because
    the required behavior is awkward to achieve by other means.
    For example, this function could be used:
    silent_cmp() {
        # compare files with no output except error messages
        message=$(cmp "$@" 2>&1 >/dev/null)
        status=$?
        case $status in
        (0|1) ;;
        (*) printf '%s\n' "$message" ;;
        esac
        return $status
    }


Option 2: allow existing practice

Issue either a "standard is clear, standard is wrong" or a "standard
is unclear" interpretation. The Notes to the Editor should be:

On L9875 (L81246 in Issue 7) change the -s description from:

    Write nothing for differing files; return exit status only.

to:

    Write nothing to standard output or standard error when files
    differ; indicate differing files through exit status only.
    It is unspecified whether a diagnostic message is written to
    standard error when an error is encountered; if a message is
    not written, the error is indicated through exit status only.

After L9924 (L81295 in Issue 7) append to STDERR:

    If the -s option is used and an error occurs, it is unspecified
    whether a diagnostic message is written to standard error.

After L9944 (L81314 in Issue 7) append to APPLICATION USAGE:

    Since the behavior of -s differs between implementations as to whether
    error messages are written, the only way to ensure consistent behavior
    of cmp when -s is used is to redirect standard error to /dev/null.

    If error messages are wanted, instead of using -s standard output
    should be redirected to /dev/null, and anything written to standard
    error should be discarded if the exit status is 1. For example:
    silent_cmp() {
        # compare files with no output except error messages
        message=$(cmp "$@" 2>&1 >/dev/null)
        status=$?
        case $status in
        (0|1) ;;
        (*) printf '%s\n' "$message" ;;
        esac
        return $status
    }

Tags tc2-2008
Attached Files

- Relationships

-  Notes
(0000944)
nick (manager)
2011-08-25 16:10
edited on: 2011-08-25 16:13

 Interpretation response
------------------------
The standard is unclear on this issue, and no conformance distinction can be made between alternative implementations based on this. This is being referred to the sponsor.

Rationale:
-------------
Existing implementations have adopted both possible interpretations of the current wording.

Notes to the Editor (not part of this interpretation):
-------------------------------------------------------


On L9875 (L81246 in Issue 7) change the -s description from:

    Write nothing for differing files; return exit status only.

to:

    Write nothing to standard output or standard error when files
    differ; indicate differing files through exit status only.
    It is unspecified whether a diagnostic message is written to
    standard error when an error is encountered; if a message is
    not written, the error is indicated through exit status only.

After L9924 (L81295 in Issue 7) append to STDERR:

    If the -s option is used and an error occurs, it is unspecified
    whether a diagnostic message is written to standard error.

After L9944 (L81314 in Issue 7) append to APPLICATION USAGE:

    Since the behavior of -s differs between implementations as to whether
    error messages are written, the only way to ensure consistent behavior
    of cmp when -s is used is to redirect standard error to /dev/null.

    If error messages are wanted, instead of using -s standard output
    should be redirected to /dev/null, and anything written to standard
    error should be discarded if the exit status is 1. For example:
    silent_cmp() {
        # compare files with no output except error messages
        message=$(cmp "$@" 2>&1 >/dev/null)
        status=$?
        case $status in
        (0|1) ;;
        (*) printf '%s\n' "$message" ;;
        esac
        return $status
    }


At page 2534 line 81336 (FUTURE DIRECTIONS), change "None" to "Future versions of this
standard may require that diagnostic messages are written to standard error when the -s option is specified"
(0001798)
ajosey (manager)
2013-09-06 04:49

 Interpretation Proposed 6 Sep 2013
(0001882)
ajosey (manager)
2013-10-14 13:03

 Interpretation approved 14 October 2013

- Issue History
Date Modified Username Field Change
2011-07-21 11:07 geoffclare New Issue
2011-07-21 11:07 geoffclare Status New => Under Review
2011-07-21 11:07 geoffclare Assigned To => ajosey
2011-07-21 11:07 geoffclare Name => Geoff Clare
2011-07-21 11:07 geoffclare Organization => The Open Group
2011-07-21 11:07 geoffclare Section => cmp
2011-07-21 11:07 geoffclare Page Number => 253
2011-07-21 11:07 geoffclare Line Number => 9875
2011-07-21 11:07 geoffclare Interp Status => ---
2011-08-25 16:10 nick Note Added: 0000944
2011-08-25 16:10 nick Status Under Review => Interpretation Required
2011-08-25 16:10 nick Resolution Open => Accepted As Marked
2011-08-25 16:10 nick Description Updated
2011-08-25 16:10 nick Desired Action Updated
2011-08-25 16:11 nick Final Accepted Text => See Note: 0000944
2011-08-25 16:11 nick Interp Status --- => Pending
2011-08-25 16:11 nick Tag Attached: tc2-2008
2011-08-25 16:13 nick Note Edited: 0000944
2013-09-06 04:49 ajosey Note Added: 0001798
2013-09-06 04:49 ajosey Status Interpretation Required => Resolution Proposed
2013-09-06 05:01 ajosey Interp Status Pending => Proposed
2013-09-06 05:04 ajosey Status Resolution Proposed => Interpretation Required
2013-10-14 13:03 ajosey Interp Status Proposed => Approved
2013-10-14 13:03 ajosey Note Added: 0001882
2019-06-10 08:55 agadmin Status Interpretation Required => Closed

Mantis 1.1.6[^]
Copyright © 2000 - 2008 Mantis Group
Powered by Mantis Bugtracker