added automatic keyword assignment support to test macro

[DebadityaPal]

As seen in #38215 , while passing keywords without assignment to test macro, the keyword is ignored. In these cases the keyword is passed as a Symbol in the metaprogram, I modified the macro functions so that it does not get ignored.
fixes #38215

[simeonschaub]

Awesome, looks good to me! Could you also add a test for this case? Have a look in stdlib/Test/test/runtests.jl, there are already some tests cases with isapprox there.

[DebadityaPal]

Sure, I shall add the necessary tests.

[DebadityaPal]

The required tests have been added.

[JeffBezanson]

I don't understand the purpose of this whole block of code --- the escaped_kwargs get moved into a :parameters expression, which will also interpret them as keyword arguments, so why do the argument expressions need to be transformed? Replicating these cases from lowering is brittle. (Also missing the a.x => x = a.x case.)

[simeonschaub]

I think kwargs are normalized to enable pretty printing for failed tests, where the actual values passed to a function call get interpolated for printing. E.g., this produces the following output:

julia> using Test

julia> a, b = 1, 2
(1, 2)

julia> @test isapprox(a, b, rtol=a/10)
Test Failed at REPL[62]:1
  Expression: isapprox(a, b, rtol = a / 10)
   Evaluated: isapprox(1, 2; rtol = 0.1)
ERROR: There was an error during testing

I don't think there's a much better way around doing this transformation for the automatic keyword assignment syntax. The ; a.x case is a good point though, that should also be added.

[JeffBezanson]

I see. We don't need to refactor this now but it would be better to "parse" the arguments out of the expression just once, instead of parsing, re-forming an Expr, and then re-parsing in eval_test.

[DebadityaPal]

Actually when the argument expression contains an unassigned keyword like @test isapprox(1,2; atol) here atol is parsed as a symbol and not as a keyword. The first round of parsing allows us to explicitly set that along with the other keywords before evaluating the Expr. Also i am slightly confused with the a.x syntax, could you just show me an example and then i can start working on it.

[simeonschaub]

here atol is parsed as a symbol and not as a keyword

Yes, semantically it's a keyword argument, but you are right, in the AST it's just a symbol.

Also i am slightly confused with the a.x syntax, could you just show me an example and then i can start working on it.

The basic gist is that isapprox(x, y; a.atol) is lowered to be the same as isapprox(x, y; atol=a.atol) (a could be defined as a = (; atol = 0.1), for example). See #34331 for more details.

[simeonschaub]

@DebadityaPal Do you want to update this for the ; a.x case as well? Alternatively, I think we could also just merge this now and address that in a separate PR.

[DebadityaPal]

I would prefer to work on that issue over a second PR, as it could take me some time and it wouldn't be fruitful if the current changes are put on hold for that.

[DebadityaPal]

Nevermind my previous comment, I have added the a.x syntax and a unit test to go with it.

[simeonschaub]

Awesome, thank you!