Signalhandling not detailed enough: sigaction(2) not mentioned anywhere #6018
Comments
|
Comment author: @xavierleroy This looks like a fine hair being split. All POSIX-compliant systems define signal() [from the C standard library] on top of sigaction() [from POSIX]. Likewise, OCaml defines Sys.signal in terms of sigaction() on POSIX hosts and of signal() on other hosts. We're not stupid. |
|
Comment author: oliver I did not said, you are stupid. But the problem between signal() and sigaction()/sigprocmask()/... So, who read that book and did programming in C, then switch to OCaml, will miss sigaction()-call. This is especially, because the other calls, like sigprocmask() are not also And regarding the Stevens' book (which I think even today is some kind of reference), this IMHO does not look like "a fine hair being split". It would be just some sentences in the documentation. |
Original bug ID: 6018
Reporter: oliver
Status: closed (set by @xavierleroy on 2015-12-11T18:19:35Z)
Resolution: not a bug
Priority: normal
Severity: minor
Category: documentation
Bug description
The reliable Unix signals need the function sigaction(2) to install the signal-related action.
sigaction(2) is one of the signal functions, accompanied by sigprocmask(2), sigsuspend(2), sigpending(2).
In the doc for the Unix-module, there is only mentioned, that signal numbers must be taken from Sys-module, and that signal handlers also will be set via Sys-module.
When reading the description in Sys-module it is also not mentioned that sigaction(2) is used.
So, when looking only at the docs for Sys- and Unix-module, it is not clear if signal(2) or sigaction(2) is used to establish the signal handler.
Someone coming from the C-world will miss sigaction(2) at least to be mentioned. That it is used as "hidden magic" under the hood of Sys-module is not enough, IMHO.
sigpending(2) and the other signal-functions are also available in Unix-module.
So it would be more consequently (from the Unix-systems view) to also make sigaction(2) available in Unix-module.
To have all signal-installation handling in Sys-module also makes sense somehow (regarding platform-independence), but then it should be mentioned that sigaction(2) is used in case of Unix-systems via the Sys-module. (This is not so consequently regarding platform-seperation, because Unix-stuff should belong to Unix-module).
The state of the docs and implementation looks like a compromise between platform-independence in usage of the modules vs. platform-seperation.
Put signal-installation into Sys independent of platform makes sense; seperate Unixoid systemhandler-installation intto Unix-module also makes sense.
So, the way it is done at the moment, is ok from a certain point of view.
But in in both cases, the documentation need to mention sigaction(2) somehow, somewhere; otherwise switching from C to OCaml can be confusing at this point.
Something like:
"OCaml is weird, it has sigprocmask(2) and sigpending(2) and so on, but does
not provide sigaction(2). Can this be seen as a reliable language?
Is there some hidden sigaction(2) magic? If so, it might be seen as
reliable, but weird."
The text was updated successfully, but these errors were encountered: