Output satisfies predicate?

[dsprenkels]

Thanks for this awesome library!

I would like to check if the command's output is from a specific format. I am specifically dealing with formats like this:

0156dc3c96b0939fa...
02fab783bdfc017dd...

with the first byte being a counted integer, and the rest random. So I would like to check if the binary lengths are correct and if the first byte is 01, 02, ....

Is it possible to pass a function to a OutputAssertionBuilder method which allows me to check by predicate?

If not, I would like to propose one (OutputAssertionBuilder::satisfy), which you can call with a function which checks if the string satisfies a given predicate.

Example type signature:

fn satisfies<F>(pred: F) -> Assert
    where F : FnOnce(&str) -> bool

[epage]

Currently, we only support exact matches and substring matches.

The idea is welcome. In #20, we talk about regex matches and fn matches. For an fn, we either need to Box it or have a state-less pointer.

Your suggestion for the name is satisfies. In the API, this would look like

fn all_lowercase(output: &str) -> bool {
   ...
}

...
    .stdout().satisfies(all_lowercase)
    .unwrap();

I don't (yet?) have much in the way of opinions on the name. Is there any kind of precedence that we should be consistent with?

[killercup]

The std lib has an abstraction like that. I think it was Pattern, in str or something like that. Ed Page <[email protected]> schrieb am Mo. 16. Okt. 2017 um 20:05:

…

Currently, we only support exact matches and substring matches. The idea is welcome. In #20 <#20>, we talk about regex matches and fn matches. For an fn, we either need to Box it or have a state-less pointer. Your suggestion for the name is satisfies. In the API, this would look like fn all_lowercase(output: &str) -> bool { ... } ... .stdout().satisfies(all_lowercase) .unwrap(); I don't (yet?) have much in the way of opinions on the name. Is there any kind of precedence that we should be consistent with? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#55 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AABOX6XmhTW5JEkgHd3RN-QTdh60w3GUks5ss4zjgaJpZM4P66Ki> .

[dsprenkels]

Could it be something like this?

It changes the code style, but I think that the OutputAssertion would become a bit more flexible. There is however the issue of error reporting, which gets a little harder. For the error building, I was thinking about taking an extra field in the OutputAssertion struct containing an error builder. Another option could be for pred to return a Result and putting the error building in the assert.rs code. Then the user could call two functions like predicate (-> bool) or predicate_ok (-> Result), depending on whether they want to customize their error or something.

Or ditch this plan altogether, cause closures are not very pleasant to work with. 😉

[epage]

A lot of the value is the rich error reporting, we need to be sure to preserve that.

While this might not need Debug/Clone, I still hesitate when not supporting such basic traits.

Until someone has a use case for it, we could avoid unifying the code branches and have a stateless function pointer rather than a stateful one in a Box.

[dsprenkels]

I restored the error reporting and restored the Debug/Clone trait implementations. Currently I use a Rc for the Clone functionality, which looks a bit ugly but is not really a problem (?). I don't really understand what you mean by your last statement, do you mean using &'a Fn everywhere, instead of a boxed type?

I also implemented the functions that I mentioned before (although only 'static, which I think should be enough). Shall I open a pull request?

(Compare changes: https://github.com/killercup/assert_cli/compare/master...dsprenkels:predicates?expand=1)

[epage]

Huh, somehow lost my post

I don't really understand what you mean by your last statement, do you mean using &'a Fn everywhere, instead of a boxed type?

I was thinking of a function pointer rather than a function trait unless there is a use case where we need the trait instead.

Shall I open a pull request?

Yes please, it'll provide a better forum for talking about the code.

Also, maybe I'm just being crotchety but I'm somewhat preferring the enum approach rather than unifying everything behind a boxed Fn trait. Thoughts?

[epage]

btw my current thought on what this API could look like

• satisfies(Fn(String) -> bool, message: String)
  • message is there since we can't gleam anything from the Fn
• satisfies_ok(Fn(String) -> Result<(), Box<Error>>)
  • Box<Error> so that the caller can provide a completely custom error message to explain why their assertion failed.