Re: [PATCH] zsh/random module [UPDATED]

[Matthew Martin <phy1729@xxxxxxxxx>]

On Tue, Nov 22, 2022 at 08:58:30PM -0600, Clinton Bunch wrote:
> On 11/22/2022 7:23 PM, Matthew Martin wrote:
> > On Sun, Nov 20, 2022 at 08:57:25PM -0600, Clinton Bunch wrote:
> > > On 11/20/2022 8:21 PM, Matthew Martin wrote:
> > > > If a uniform random function is desired in zsh, I think it should mirror
> > > > the interface of arc4random_uniform: just take an upper_bound and return
> > > > a value in the range [0, upper_bound).
> > > The original implementation did exclude the upper bound, inadvertently, but
> > > after discussion it seemed that was counter-intuitive and something easily
> > > missed in a cursory glance at the documentation.  Adding a lower bound was
> > > easy enough and saves extra shell code even if it's not likely to be used as
> > > much.
> > arc4random_uniform excludes the upper bound so it is a direct
> > replacement for <random source> % <upper bound>. I think it would be
> > wise to follow an established API and avoid off by one errors when
> > refactoring existing code. Similarly it doesn't offer a lower bound
> > option since that's trivial to implement and difficult to get wrong.
> Instead you get off-by-one errors in new code because people assume the
> specified limit is included unless they read the documentation closely.

I think exclusive would be more useful. A common use case would be
having an array and wanting to choose a random element. If exclusive the
code is
    choices=(red green blue); print ${choices[arc4random_uniform($#choices) + 1]}
if inclusive an extra -1 is required.

> > > yes, much of getrandom *could* be implemented in shell code, much less
> > > efficiently.  For example my precmd could save the math eval by specifying
> > > getrandom -L 1 -U 7.
> > > 
> > > Once you've written the code to access the kernel random source, it seems to
> > > make sense to me to make its output as flexible as possible.  I don't get
> > > the concern about backwards compatibility in implementing a new builtin
> > > unless you know of an external command called getrandom that is in common
> > > use, and that's easily fixed by changing it to zgetrandom.
> > I disagree on making the output as flexible as possible because there
> > are already features in zsh to get those formats and the API would need
> > to be maintained into future releases even if seldom used. I think it
> > would be wise to start with the minimum interface necessary and then
> > build off it as uses arise.
> > 
> > Interfacing with libc and the kernel to get random numbers is something
> > that can only be done in C and SRANDOM builds off an existing API, so
> > I have no opposition to it. I can accept that uniformly distributed
> > random integers is a useful and non-trivial task, so probably should be
> > included, but would prefer an existing and proven API like
> > arc4random_uniform.
> > 
> > I have a patch based off yours that implements just SRANDOM and
> > a mathfunc for arc4random_uniform, but haven't yet had time to test it
> > on multipl platforms.
> Even if we go this way,  I don't think the function should be called
> arc4random_uniform.  It only implements the arc4random algorithm on *BSD.
> > 
> > With just SRANDOM and arc4random_uniform the features of getrandom can
> > be implemented in a script. I've taken a quick shot at a proof of
> > concept below.
> > 
> > 
> [snip shell code example]
> 
> That's a lot of zsh code, most of which would be difficult for a non-zsh
> expert to write.  So it would need to be provided as a function and then
> you're left with the same support problem, but a slower implementation that
> keeps converting strings to numbers and back again.

It's a lot of code because it's doing a lot. Most users would need just
a subset to do their particular task. I'm not suggesting adding the
script to zsh; it was just to show SRANDOM and arc4random_uniform are
sufficient.

> getrandom doesn't need advanced zsh programming knowledge to use.  The -U
> and -L make it easy for a relative novice. arc4random_uniform is a
> programmer's interface, not a user's interface.

I don't agree with this argument. Addition to establish a lower bound is
trivial and the subtraction to determine the range is not much harder.
Further since getrandom is a builtin not a mathfunc it's more cumbersome
to use in the array choice context.

> You also ignored the zrandom function which is definitely non-trivial to
> implement in the shell.  It's not even trivial to implement in C.

It's not clear what a use case for zrandom would be.