bug#25540: notice issue in expand -- doesn't allow for expressing tabsize value in tabstop(s)

[Pádraig Brady]

On 26/01/17 10:26, Reuti wrote:
> 
>> Am 26.01.2017 um 05:29 schrieb L A Walsh <address@hidden>:
>>
>>
>> In programs that take tabstops, as an alternative to a tabsize, I've always
>> seen tabs beyond the end of the list taken as equal to the highest tab-stop
>> difference.  So for a tabsize=8, a tabset of 1,9 would be equivalent -- with
>> tabs above "9" being "9-1" or every 8th column above 9.
>>
>> Otherwise you have no way of expression all tabs on a line that stretches 
>> out to
>> "???" 160? 240? what? other than to enumerate tabstops to infinity.
>>
>> If they want to limit tabstops above the last to size "1", they can use 
>> something like 1,9,10.  How else can one specify tabs beyond the last
>> for a size other than "1"?
>>
>> Could this be changed/fixed?
> 
> For now the behavior is like specified on the info page: "[…] and replace any 
> tabs beyond the last tab stop given with single spaces." To avoid that this 
> gets broken, I would suggest to use a modified syntax like 1,9,30,34,/4 for 
> using a width of 4 beyond 34.
> 

I like that. Explicit and extensible.
An alternative could be: --tabs=1,9,30,34,+4 ?

> This could even be expanded to: -t 2/4,120,4/5,/9
> 
> Two times a width of 4, a fixed 120, four times a width of 5 and all 
> remaining have a width of 9 (could also be */9 instead of /9).

That might be overkill.
It wouldn't provide extra functionality, only a shortcut,
and a very rarely used one at that.

> While we are on this:
> 
> expand -t 5,15,25,35 file
> expand -t 5,15,25,,,,35 file
> 
> are AFAICS both the same. I would expect the second to behave different 
> (using a space for the three tabs which have no value).

I'm not sure we should assume empty = 1.
Note FreeBSD rejects this format:

  $ seq 8 | paste -s -d$'\t' - | expand -t1,,2
  expand: bad tab stop spec

One might consider and empty value after a comma to mean take the previous tab 
stop width,
and that would be a partially backwards compat way to provide the original 
request, i.e.:

  expand -t 1,9,,

However that seems a bit hacky and non obvious.

thanks,
Pádraig