0001229: EXIT_FAILURE should not be allowed to be all non-zero values

ID	Project	Category	View Status	Date Submitted	Last Update

0001229	1003.1(2016/18)/Issue7+TC2	Base Definitions and Headers	public	2019-02-20 16:23	2024-06-11 09:08

Reporter	geoffclare	Assigned To
Priority	normal	Severity	Objection	Type	Error
Status	Closed	Resolution	Accepted As Marked

Name	Geoff Clare
Organization	The Open Group
User Reference
Section	<stdlib.h>
Page Number	359
Line Number	12240
Interp Status	Approved
Final Accepted Text	See 0001229:0004407


Summary	0001229: EXIT_FAILURE should not be allowed to be all non-zero values
Description	The value of EXIT_FAILURE in <stdlib.h> is allowed to be any non-zero value. This is in conflict with the C standard, which requires that exit(EXIT_FAILURE) returns "unsuccessful termination status" to the host environment. Since the vast majority of applications use either wait(), waitpid(), or $? in the shell to obtain this status, and these truncate the value to the lower 8 bits, it follows that in order to match the requirement in the C standard, POSIX needs to specify that (EXIT_FAILURE & 255) is non-zero. It could also be argued that (EXIT_FAILURE & 255) should not have a value greater than 128 because when obtained via $? in the shell these values are ambiguous, in that it is not possible to distinguish a call to exit() with an argument > 128 from termination by a signal. If we are going to disallow values > 128, it would also make sense to disallow 126 and 127 since these are also ambiguous in certain circumstances (the shell uses them as the exit status for commands that it can't execute for various reasons). Unless anyone knows of an implementation where (EXIT_FAILURE & 255) is 128 we might as well disallow that as well, for simplicity, so that only the range 1 to 125 is allowed. Finally, calling exit() with a value greater than 255 is something that only programs which are specifically designed to have their exit status obtained by waitid() should do, so it seems odd to allow this to happen for all "pure ISO C" programs that call exit(EXIT_FAILURE). To prevent it, (EXIT_FAILURE & 255) should equal EXIT_FAILURE, leaving us with the range 1 to 125 as the only allowed values for EXIT_FAILURE. The proposed changes also include an editorial wording change to use "shall".
Desired Action	On page 359 line 12240 section <stdlib.h>, change: EXIT_FAILURE Unsuccessful termination for exit(); evaluates to a non-zero value EXIT_SUCCESS Successful termination for exit(); evaluates to 0. to: EXIT_FAILURE Unsuccessful termination for exit(); [CX]the value shall be between 1 and 125 inclusive[/CX]. EXIT_SUCCESS Successful termination for exit(); the value shall be 0. On page 361 line 12340 section <stdlib.h>, change RATIONALE from: None. to: The ISO C standard requires that <tt>exit(EXIT_FAILURE)</tt> returns ``unsuccessful termination status'' to the host environment. In a POSIX host environment this means that the lower 8 bits of EXIT_FAILURE must be non-zero. The standard developers decided to further restrict the allowed values for the following reasons: Exit statuses of 126, 127, and greater than 128 are ambiguous in certain circumstances because they have special meanings in the shell (see [xref to XCU 2.8.2 Exit Status for Commands]). Calling exit() with a value greater than 255 is something that only programs which are specifically designed to have their exit status obtained by waitid() should do (since it does not truncate the exit status to 8 bits). ``Pure ISO C'' programs that call <tt>exit(EXIT_FAILURE)</tt> do not meet this design criterion. The value 128 is disallowed for simplicity, making the allowed values 1 to 125 inclusive rather than 1 to 125 inclusive and 128. The requirement that the value of EXIT_SUCCESS is 0 is not shaded CX because this matches the requirement in the ISO C standard that <tt>exit(EXIT_SUCCESS)</tt> returns ``successful termination status'' to the host environment (when the host environment is a POSIX implementation).
Tags	tc3-2008

eblake 2019-02-20 16:42 manager bugnote:0004261	Historically, the Tandem/NSK system defined EXIT_FAILURE as -1 (which resulted in a status of 255 when 8-bit truncation occurs). However, this notably interferes with standardized behavior of xargs doing early termination when an 8-bit exit status of 255 is detected, enough so that gnulib intentionally declares Tandem/NSK as broken and redefines EXIT_FAILURE when building for that platform: https://git.savannah.gnu.org/cgit/gnulib.git/commit/?id=d195802a So while it is a counterexample of a platform where EXIT_FAILURE is not between 1-125, it is also evidence for why this change is appropriate, although it may be yet another bullet point of rationale to add about the xargs behavior on exit(255)/exit(-1).

geoffclare 2019-02-27 11:53 manager bugnote:0004269	Revised proposal: On page 359 line 12240 section <stdlib.h>, change: EXIT_FAILURE Unsuccessful termination for exit(); evaluates to a non-zero value EXIT_SUCCESS Successful termination for exit(); evaluates to 0. to: EXIT_FAILURE Unsuccessful termination for exit(); [CX]the value shall be between 1 and 125 inclusive[/CX]. EXIT_SUCCESS Successful termination for exit(); the value shall be 0. On page 361 line 12340 section <stdlib.h>, change RATIONALE from: None. to: The ISO C standard requires that <tt>exit(EXIT_FAILURE)</tt> returns ``unsuccessful termination status'' to the host environment. In a POSIX host environment this means that the lower 8 bits of EXIT_FAILURE must have at least one bit set. The standard developers decided to further restrict the allowed values for the following reasons: Exit statuses of 126, 127, and greater than 128 are ambiguous in certain circumstances because they have special meanings in the shell (see [xref to XCU 2.8.2 Exit Status for Commands]). The xargs utility quits when a command execution exits with status 255 (see [xref to XCU xargs]). Calling exit() with a value greater than 255 or less than 0 is something that only programs which are specifically designed to have their exit status obtained by waitid() should do (since it does not truncate the exit status to 8 bits). ``Pure ISO C'' programs that call <tt>exit(EXIT_FAILURE)</tt> do not meet this design criterion. The value 128 is disallowed for simplicity, making the allowed values 1 to 125 inclusive rather than 1 to 125 inclusive and 128. The requirement that the value of EXIT_SUCCESS is 0 is not shaded CX because this matches the requirement in the ISO C standard that <tt>exit(EXIT_SUCCESS)</tt> returns ``successful termination status'' to the host environment (when the host environment is a POSIX implementation).

nick 2019-06-06 15:20 manager bugnote:0004407	Interpretation response ------------------------ The standard does not speak to this issue, and as such no conformance distinction can be made between alternative implementations based on this. This is being referred to the sponsor. Rationale: ------------- The ISO C standard requires that <tt>exit(EXIT_FAILURE)</tt> returns ``unsuccessful termination status'' to the host environment. In a POSIX host environment this means that the lower 8 bits of EXIT_FAILURE must have at least one bit set. Notes to the Editor (not part of this interpretation): ------------------------------------------------------- Make the changes in 0001229:0004269

agadmin 2019-10-07 15:17 administrator bugnote:0004607	Interpretation proposed: 7 October 2019

agadmin 2019-11-11 12:20 administrator bugnote:0004651	Interpretation Approved: 11 Nov 2019

Date Modified	Username	Field	Change
2019-02-20 16:23	geoffclare	New Issue
2019-02-20 16:23	geoffclare	Name	=> Geoff Clare
2019-02-20 16:23	geoffclare	Organization	=> The Open Group
2019-02-20 16:23	geoffclare	Section	=> <stdlib.h>
2019-02-20 16:23	geoffclare	Page Number	=> 359
2019-02-20 16:23	geoffclare	Line Number	=> 12240
2019-02-20 16:23	geoffclare	Interp Status	=> ---
2019-02-20 16:42	eblake	Note Added: 0004261
2019-02-27 11:53	geoffclare	Note Added: 0004269
2019-06-06 15:20	nick	Note Added: 0004407
2019-06-06 15:21	nick	Interp Status	--- => Pending
2019-06-06 15:21	nick	Final Accepted Text	=> See 0001229:0004269
2019-06-06 15:21	nick	Status	New => Interpretation Required
2019-06-06 15:21	nick	Resolution	Open => Accepted As Marked
2019-06-06 15:21	nick	Tag Attached: tc3-2008
2019-06-06 15:22	nick	Final Accepted Text	See 0001229:0004269 => See 0001229:0004407
2019-10-07 15:17	agadmin	Interp Status	Pending => Proposed
2019-10-07 15:17	agadmin	Note Added: 0004607
2019-11-11 12:20	agadmin	Interp Status	Proposed => Approved
2019-11-11 12:20	agadmin	Note Added: 0004651
2019-12-04 11:49	geoffclare	Status	Interpretation Required => Applied
2020-01-29 16:40	geoffclare	Relationship added	related to 0001321
2024-06-11 09:08	agadmin	Status	Applied => Closed

View Issue Details

Relationships

Activities

Issue History