[GRASS-dev] Re: [GRASS GIS] #102: new tabs in GUI have required last

[Glynn Clements]

GRASS GIS wrote:

> #102: new tabs in GUI have required last
> -----------------------+----------------------------------------------------
>   Reporter:  cmbarton  |       Owner:  grass-dev at lists.osgeo.org
>       Type:  defect    |      Status:  new                      
>   Priority:  major     |   Milestone:  6.3.0                    
>  Component:  Tcl       |     Version:  svn-trunk                
> Resolution:            |    Keywords:                           
> -----------------------+----------------------------------------------------
> Comment (by hamish):
> 
>  Replying to [comment:4 glynn]:
>  > Replying to [comment:1 hamish]:
>  Hamish:
>  > > For r.colors, v.in.ascii the input can come from stdin which is why those
>  > > input filenames are not required by the parser, although piping to stdin isn't
>  > > really an option from a GUI window.
>  Glynn:
>  > I suggest modifying the parser so that "-" is always accepted for "old,file"
>  > options.
>  >
>  > Providing incorrect type information so that "-" works is a rather ugly hack.
> 
>  What I was trying to get at was the modules work like "if input file is
>  NULL, assume input comes from stdin". I don't think those modules actually
>  have code to deal with input=-, they will treat that as a filename not
>  something special. Some modules do recognize that trick, for consistency's
>  sake I suppose all that can read from stdin should, and if the parser
>  needs to be aware of that then so be it.
> 
>  And AFAIK the parser does not actually check to see if an "old,file"
>  option's answer actually exists. It leaves that to the module. All it
>  provides is the folder file picker GUI. (Same is true for cell, vector,
>  group, ... elements)    (??)
> 
>  So the type isn't being dropped to accomodate 'oldfile=-', ->required is
>  being dropped to accomodate "if input file is NULL, assume input comes
>  from stdin".
> 
> 
>  not sure if we're talking about the same thing here...

Okay. I misinterpreted "filenames are not required" as "files are not
required (to exist)".

Realistically, we either need a more expressive type system, with
union types, or we have to eliminate all of these special cases.

BTW, the specific issue with "mode" flags (e.g. "r.colors -l") could
be dealt with fairly easily by adding a field to "struct Flag" to
indicate that the flag suppresses any requirements. That way, you
wouldn't have to do without ->required settings because of
"exceptional" cases.

-- 
Glynn Clements <glynn at gclements.plus.com>