View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0000804 | 1003.1(2013)/Issue7+TC1 | Base Definitions and Headers | public | 2013-11-24 14:13 | 2014-02-06 17:00 |
| Reporter | jilles | Assigned To | |||
| Priority | normal | Severity | Objection | Type | Clarification Requested |
| Status | Closed | Resolution | Rejected | ||
| Name | Jilles Tjoelker | ||||
| Organization | FreeBSD | ||||
| User Reference | freebsd-kern/181439 | ||||
| Section | <sys/wait.h> | ||||
| Page Number | 408 | ||||
| Line Number | 13694-13695 | ||||
| Interp Status | --- | ||||
| Final Accepted Text | |||||
| Summary | 0000804: waitid() practically requires #include <signal.h> anyway | ||||
| Description | The description of <sys/wait.h> says that the <sys/wait.h> header shall define the siginfo_t type as described in <signal.h> and that it may make visible all symbols from <signal.h>. Therefore, a conforming application that compares si_status to a constant like CLD_EXITED shall #include <signal.h> anyway. However, glibc makes visible all symbols from <signal.h> in <sys/wait.h> and the application works without #include <signal.h>. | ||||
| Desired Action | Change lines 13694 and 13695: The <sys/wait.h> header shall define the siginfo_t type as described in <signal.h>. Inclusion of the <sys/wait.h> header may also make visible all symbols from <signal.h>. to one line: Inclusion of the <sys/wait.h> header shall make visible all symbols from <signal.h>. Alternatively, only the CLD_* constants could be made mandatory. However, "may make visible all symbols from" helps implementations, but hurts application portability. | ||||
| Tags | No tags attached. | ||||
|
|
The current wording is boilerplate to support that headers can use various techniques to speed up compile times when only a few items from another header are required to access and use an interface declared in the referencing header. If a portable application doesn't compile properly, that's a defect of the header, not the standard. Most implementations don't bother trying to do this as it does makes maintenance more difficult, and just #include the referenced header so they are sure they get what's needed, at the expense of processing stuff they don't need. The wording also allows this practice. While for this particular example the speed slowdown might not be all that much for various implementations by requiring the second way, I can't see making this an exception to the general practice. For others where <signal.h> also includes declarations for a number of implementation extensions it might be significant. |
|
|
Implementations are allowed, but not required, to expose all of the symbols of signal.h. Applications requiring additional symbols are required to add the appropriate header. Adding exposure of these symbols as a new requirement can break existing implementations. The "may make visible" convention is widely used in several headers and we can see no reason to single out this header for special treatment. |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2013-11-24 14:13 | jilles | New Issue | |
| 2013-11-24 14:13 | jilles | Name | => Jilles Tjoelker |
| 2013-11-24 14:13 | jilles | Organization | => FreeBSD |
| 2013-11-24 14:13 | jilles | User Reference | => freebsd-kern/181439 |
| 2013-11-24 14:13 | jilles | Section | => <sys/wait.h> |
| 2013-11-24 14:13 | jilles | Page Number | => 408 |
| 2013-11-24 14:13 | jilles | Line Number | => 13694-13695 |
| 2013-12-17 09:38 | shware_systems | Note Added: 0002078 | |
| 2013-12-18 15:25 | shware_systems | Note Edited: 0002078 | |
| 2014-02-06 16:58 | msbrown | Note Added: 0002125 | |
| 2014-02-06 16:59 | msbrown | Note Edited: 0002125 | |
| 2014-02-06 17:00 | msbrown | Interp Status | => --- |
| 2014-02-06 17:00 | msbrown | Status | New => Closed |
| 2014-02-06 17:00 | msbrown | Resolution | Open => Rejected |
