[IDEA] New pages for types: structs and typedfefs
colomar.6.4.3 at gmail.com
Sun Sep 13 12:53:37 UTC 2020
On 9/13/20 2:01 PM, Michael Kerrisk (man-pages) wrote:
> Hi Alex,
> On Sat, 12 Sep 2020 at 10:59, Alejandro Colomar
<colomar.6.4.3 at gmail.com> wrote:
>> Hi Michael,
>> On 2020-09-12 08:33, Michael Kerrisk (man-pages) wrote:
>> > Your not the first to suggest this. Most recently, if I recall
>> > correctly, Florian also suggested it.
>> > The idea seems reasonable, but I wonder about the best way of
>> libbsd already provides a few pages on types. Maybe we could have a
>> look at them. At least I've seen 'man timespec' (which redirects to
>> > I propose the desired information for each type would be
>> > * Type name
>> > * Short explanation of the type (often this mcould be just a
>> > few words, I think)
>> > * Whether the type is specified in POSIX; POSIX requirements on
>> > the type.
>> > * Header file that defines the type (in some cases, there
>> > may be more than one. This info can be discovered in the
>> > POSIX spec. (Alex, do you have a PDF of the POSIX spec?)
>> > * Cross references to manual pages of relevant APIs that use the type.
>> I think that would be reasonable.
>> No, I don't have a PDF. I usually search here:
> You can get a PDF by registering as a member of The Open Group. I
> think the necessary info is here:
> (I find having everything in a single PDF is useful for searching.)
>> > There are some weird corner cases. For example, clock_t:
>> > * times(2): clock_t == clock ticks (sysconf(_SC_CLK_TCK))
>> > * clock(3): clock_t measures in CLOCKS_PER_SEC
>> That would just be 1 or 2 more lines in the explanation, I guess.
> Yes, I guess.
>> That's also similar to the typical (mis?)use of size_t as a ptrdiff_t.
>> > Then, do we do one page per type? At first glance, that seems
>> > unwieldy to me. (I could be wrong.) And it seems to me that
>> > there might be benefits in having all of the information in
>> > one place rather than spread across multiple pages. (For example
>> > cantralizing the info would make it easier for the reader to
>> > get an overview.)
>> I agree in that everything should be centralized, at least in the
>> beginning. That would make it much easier to maintain and find the
>> information. If the future requires the information to be spread
>> across many pages, let the future solve that problem :)
>> > Alternatively, we could have one big page that is a list of the
>> > types with the above information. Say "system_data_types(7)".
>> > That page might be an alphabetically ordered hanging list of
>> > entries that look like this:
>> > timer_t <time.h> or <sys/types.h>
>> > POSIX timer ID.
>> > Conforming to: POSIX.1-2008.
>> > See: timer_create(2), timer_delete(2), timer_getoverrun(2),
>> > timer_settime(2)
>> I'd say here is missing the POSIX requirements on the type.
> As far as I can tell, in the case of timer_t, I think there are no
> requirements in POSIX.
>> Is it a 32-bit or 64-bit or may vary? Is it signed or unsigned?
> POSIX doesn't specify, I think.>
> One other thing the page should show of course is definition of the
> structure types.
timer_t <time.h> or <sys/types.h>
POSIX timer ID.
typedef void *timer_t;
Conforming to: POSIX.1-2008.
See: timer_create(2), timer_delete(2), timer_getoverrun(2),
Like this? Should I specify somehow if the type definition
is so for Linux only, or for all POSIX, ...?
x86_64-linux-gnu/bits/types/timer_t.h:7:typedef __timer_t timer_t;
x86_64-linux-gnu/bits/types.h:171:__STD_TYPE __TIMER_T_TYPE __timer_t;
x86_64-linux-gnu/bits/types.h:126:# define __STD_TYPE typedef
x86_64-linux-gnu/bits/typesizes.h:70:#define __TIMER_T_TYPE void *
This is so for my Debian x86-64.
However, I don't know if this is the definition for all architectures.
I'll start with what I can do, and if you see any type that you know
has a different definition in other systems, I'll let that to you :)
>> > Then of course, we'd need to have links to that page, so that
>> > people could just type 'man timer_t'. What section should the links
>> > be in? The reasonable candidates would be section 3 or 7. I'm not
>> > yet sure which is better. But the point is that we'd have files
>> > such as timer_t.3 (or timer_t.7) that are link pages containing the
>> > line:
>> > .so man7/system_data_types.7
>> Sure. And for the structs, I'd allow:
>> 'man struct timespec' (For simplicity)
>> 'man struct-timespec' (Similar to the git man pages)
>> 'man timespec' (For compatibility with libbsd)
> Mainly, I'm interested in the last case. That's the one I think that
> people would most likely use. In a follow-up mail, you expressed
> concern with conflicts with libbsd pages. I'm not too worried about
> that. There are already *many* conflicts between libbsd and man-pages.
I wasn't concerned about conflict with libbsd; that's the form libbsd
uses, and a good point for having that form would be for compatibility
(people will probably like having to write 'man timespec' in any
system and work).
I was instead concerned that some struct tag may have the same name as
some function, which I don't know for sure:
Let's say there exist a function 'int foo(void)', and a 'struct foo'.
If that is the case, which I ignore, you would need to either have
'foo.3' and 'foo.3t' or have 'foo.3' and 'struct-foo.3'.
>> > For the moment at least, I'd favor the "one big page plus
>> > links" approach.
> Would you like to take a shot at this? I'd suggest just a simple page
> covering just two or three types to start with. Maybe time_t and
> timer_t, or otherwise some types that seem good choices to you?
Yes, I'd like to. It'll be my first page from scratch, though, so
don't expect it to be soon :-}
Maybe 'timer_t', 'time_t' and 'struct timespec' would be a good start.
Do you think there's any page that has a similar format to what we want
to base on it?
More information about the libbsd