Hi Branden,
On 2020-09-28 12:40, G. Branden Robinson wrote:
> At 2020-09-28T07:59:14+0200, Michael Kerrisk (man-pages) wrote:
>>> Signed-off-by: Alejandro Colomar <colomar.6.4.3@xxxxxxxxx>
>>
>> I do think this requires an explanation saying what you are
>> trying to do with this change (and then perhaps a more expansive
>> "Subject" also).
>>
>> I can visually see what you are doing with this patch,
>> but I do wonder if there is a better way of doing it.
>>
>> I've dropped Branden into CC. Perhaps he has a comment.
>
> Hi Michael, yes--thank you!
>
> In my opinion, use of "raw" *roff requests in a man page is a "code
> smell".
>
> Let me have a look...
>
>>> diff --git a/man7/system_data_types.7 b/man7/system_data_types.7
>>> index 361e8d411..ff0403df9 100644
>>> --- a/man7/system_data_types.7
>>> +++ b/man7/system_data_types.7
>>> @@ -66,7 +66,7 @@ system_data_types \- overview of system data types
>>> .TP
>>> .I aiocb
>>> .RS
>>> -.PP
>>> +.br
>>> Include:
>>> .IR <aio.h> .
>>> .PP
>
> This is changed already from my most recent "git pull", so things are
> moving fast indeed in this area! Here's what I show currently:
>
> .\"------------------------------------- aiocb ------------------------/
> .TP
> .I aiocb
> .IP
> Include:
> .IR <aio.h> .
> .IP
> .EX
> struct aiocb {
>
> So first the indented paragraph was replaced (or proposed to be
> replaced) by a relative inset (.RS) and a paragraph break (.PP). It's
> not visible in this diff where the relative inset ends, but I reckon
> it's right before the next tagged paragraph (.TP).
>
> Now, the .PP is being replaced by .br, I guess because you want to
> eliminate some vertical space between the paragraph tag ("aiocb" in
> italics above) and the following material?
>
> However, .TP already does this for you when the tag is wider than the
> paragraph indentation. You probably discovered this yourself with
> "ptrdiff_t".
>
> I get the feeling you're going for something specific in terms of visual
> presentation.
>
> What is it that you don't like about the following pattern?
>
> .TP
> .I typename_t
> Include
> .IR <header.h> .
> .IP
> .EX
> struct typename {
> // ...
> }
> .EE
> .IP
> Notes and commentary.
That's the first thing we tried; actually if I remember correctly,
that's what Michael suggested first.
But as you said, when we documented types with a long name such as
ptrdiff_t, the alginment broke: sometimes Include was in the same line,
and sometimes in the next one.
That's when we opted for
.\"------------------------------------- aiocb ------------------------/
.TP
.I aiocb
.IP
Include:
.IR <aio.h> .
.IP
.EX
struct aiocb {
>
> The above visually aligns everything by the type name with all the
> material related to each type inset, which makes the page easy to scan.
> Literal code is not filled, so code indentation is respected and (on
> typesetters) the code be in a monospaced font so it will stand out and
> align correctly. You don't have to manage insets with .RS and .RE, and
> you can still have as many paragraphs as you want with .IP.
Then, a few days ago, I pushed the change that you still didn't see:
Replace .IP with .RS/.RE + .PP
I'll refer to the patch for the rationale: it was a bit long. Basically,
it better represents a whole block, and it's easier to further indent:
https://lore.kernel.org/linux-man/16c0890e-33c0-d052-d7ee-39385cd79299@xxxxxxxxx/T/#m21da4dba4b3a44b0a59cad1e7eacb13a68a91636
This change actually didn't have visual effects; just source changes.
>
> I have some guesses as to what might be bothering you, but I don't want
> to put words in your mouth, and my mails tend to run too long anwyay, so
> I won't voice them right now. :)
>
> Regards,
> Branden
>
Thanks,
Alex
