Re: [dev] stest review

From: Anselm R Garbe <garbeam_AT_gmail.com>
Date: Sat, 11 Feb 2012 15:39:35 +0100

On 11 February 2012 14:04, Christoph Lohmann <20h_AT_r-36.net> wrote:
> Anselm R Garbe wrote:
>> However the real point is that the getopt() style or ARGBEGIN crap
>> enables and encourages the developer to introduce a bad command flag
>> interface. Because those approaches hide the utter complexity
>> involved, the developer tends to care less here. This is my main
>> argument against getopt() or ARGBEGIN.
>>
>> If you can write a simple for() loop to process your command line
>> flags, your interface can't be that hard to grasp for the user.
>> Otherwise he will look up the weirdo flags quite often in your man
>> file and develop hate against your tool over time ;)
>
> That is plain wrong. The ARGBEGIN { } ARGEND; style is easy to read
> and adds all the convenience you want in flag parsing. On the other
> side getopt() adds a huge dependency.

Well, I agree that ARG... is simpler than getopt(), but that's beside
the whole point.
Usually the ARG.. handling can be done very similar just using a for()
loop, in most cases.

And in all other cases you should rather reconsider your whole command
line interface, as this indicates bad design or too much choice
already.

> The suckless standard library should include either the ARG* macros
> or add another function, which can be put into the switch() state-
> ment.

I really can't see that need.

> Users will rather be irritated, if the commandline argument hand-
> ling is different in every application. They then *have* to read
> the sourcecode for finding out how arguments are handled.

It's quite consistent in most suckless tools actually. One difference
I stumbled upon is exactly stest, because it uses the clunky getopt()
approach and I really wonder why it needs so many flags.

Cheers,
Anselm
Received on Sat Feb 11 2012 - 15:39:35 CET